Device state determination system and device state determination method

Abstract

To provide a device state determination technique by which the state of a device can be determined based even on actual measurement data including noise.SOLUTION: A device state determination system 1 comprises: an analysis unit 10 for calculating by reflecting parameters on a formulated equation based on a physical model; a physical quantity measurement unit 12 for acquiring a physical quantity actually measured by each device 51 as actual measurement data; a feature quantity extraction unit 19 for extracting a feature quantity from the calculation result of the analysis unit 10 and the actual measurement data; a noise processing unit 20 for removing noise from the actual measurement data based on the feature quantity extracted by the feature quantity extraction unit 19; a parameter identification unit 14 for identifying the parameters of the physical model so that the calculation result of the analysis unit 10 reproduces the actual measurement data from which the noise is removed by the noise processing unit 20; and a state determination unit 15 for determining the state of a device 50 based on the parameters identified by the parameter identification unit 14.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] An embodiment of the present invention relates to a state determination technique for a rotating electrical machine. [Background technology]

[0002] Conventionally, there are technologies that judge the presence or absence of an abnormality in rotating equipment or estimate the cause from measurement data detected by sensors, such as diagnostic methods that use artificial intelligence (AI) or simulation. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-182393 [Patent Document 2] JP 2020-004087 A [Patent Document 3] JP 2005-332360 A [Patent Document 4] JP 2003-149043 A [Patent Document 5] JP 2002-023845 A Summary of the Invention [Problem to be solved by the invention]

[0004] When using data related to past abnormalities to determine the presence or absence of abnormal symptoms or the cause of an abnormality, abnormality data acquired when an abnormality occurs is required. However, since it is difficult to acquire abnormality data for equipment with few abnormality occurrences, it is difficult to determine an abnormality from measurement data. For example, important equipment that may lead to long-term shutdowns in large plants, or equipment that is important for safety such as nuclear plants, is periodically disassembled and inspected, and repaired or replaced before an abnormality occurs. Therefore, there are very few cases of abnormalities occurring. In addition, various abnormality modes may occur in one piece of equipment, and data is required for all such abnormality modes.

[0005] For electric motors, technology is sometimes used to monitor vibration and determine signs of abnormality from changes in vibration level and frequency characteristics. The presence or absence of abnormal signs is determined by inspectors based on their experience, and the results of such determinations are not very valid. Furthermore, changes in vibration depend on the frequency characteristics of the equipment or the installation state of the sensor. For this reason, it is difficult to determine the presence or absence of abnormalities or estimate the cause with a diagnosis based only on predetermined data.

[0006] Actual measurement data also includes data that is unrelated to anomalies. For example, there is noise or fluctuation generated from peripheral devices or measuring instruments, and displacement vibration caused by surface shapes such as unevenness or scratches on the shaft. The effects of such noise are often not reproducible by simulation, which reduces the accuracy of the judgment. In addition, conventional filtering processes such as averaging may remove signals along with noise.

[0007] An embodiment of the present invention has been made in consideration of the above circumstances, and aims to provide an apparatus state determination technique that can determine the state of an apparatus even from actual measurement data that includes noise. [Means for solving the problem]

[0008] An equipment state determination system according to an embodiment of the present invention includes a model construction unit that constructs a physical model that reproduces the operation of an equipment composed of a plurality of equipment including at least a rotating electric machine and reproduces the state of the equipment that occurs in association with the operation, an analysis unit that performs calculations by reflecting parameters in an equation formulated based on the physical model, a physical quantity measurement unit that acquires physical quantities actually measured by each of the equipment as actual measurement data, a feature extraction unit that extracts features from a calculation result of the analysis unit and the actual measurement data, a noise processing unit that removes noise from the actual measurement data based on the feature extracted by the feature extraction unit, a parameter identification unit that identifies the parameters of the physical model so that the calculation result of the analysis unit reproduces the actual measurement data from which the noise has been removed by the noise processing unit, and a state determination unit that determines the state of the equipment based on the parameters identified by the parameter identification unit. Effect of the Invention

[0009] According to an embodiment of the present invention, an apparatus state determination technique is provided that is capable of determining the state of an apparatus even when actual measurement data includes noise. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a block diagram showing a system configuration of an apparatus state determination system. [Diagram 2] FIG. 2 is a configuration diagram showing an apparatus state determination system connected to an apparatus. [Diagram 3] FIG. 4 is an explanatory diagram showing a mode in which a physical quantity obtained from a sensor is converted into processed data. [Figure 4] FIG. 13 is an explanatory diagram showing a first example of improper installation of a coupling. [Diagram 5] FIG. 11 is an explanatory diagram showing a second example of improper installation of a coupling. [Figure 6] FIG. 11 is an explanatory diagram showing an aspect of learning based on parameters, calculation results, and feature amounts. [Figure 7] 11 is a graph for explaining extraction of a feature amount of a frequency component spectrum. [Figure 8]11 is a graph for explaining a time-series waveform caused by unevenness of a rotation shaft. [Figure 9] 11 is a graph for explaining noise processing of a frequency component spectrum. [Figure 10] 4 is a flowchart showing a device state determination method. [Figure 11] FIG. 13 is an explanatory diagram showing a modified example of setting a threshold value. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, an embodiment of an apparatus state determination system and an apparatus state determination method will be described in detail with reference to the drawings.

[0012] Reference numeral 1 in Fig. 1 denotes an equipment state determination system of this embodiment. This equipment state determination system 1 supports maintenance activities for a specific equipment 50 (Fig. 2) by optimizing the determination of the presence or absence of an abnormality and the estimation of the cause of the abnormality. In particular, the equipment state determination system 1 is used to evaluate the state of the equipment 50 from actual measurement data for equipment 50 that has never experienced an abnormality in the past. The state of the equipment 50 includes at least any one of the presence or absence of an abnormality, the presence or absence of an abnormality symptom, the abnormality event, the cause of the abnormality, the degree of the abnormality, and the type of abnormality.

[0013] The device state determination system 1 is configured with a computer having hardware resources such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), a HDD (Hard Disk Drive), and an SSD (Solid State Drive), and the CPU executes various programs to realize information processing by software using the hardware resources. Furthermore, the device state determination method of this embodiment is realized by having the computer execute the various programs.

[0014] First, the system configuration of the equipment state determination system 1 will be described with reference to the block diagram shown in FIG.

[0015] The equipment state determination system 1 includes a plurality of sensors 2 and an analytical computer 3. The analytical computer 3 includes an input unit 4, an output unit 5, a communication unit 6, a storage unit 7, and a main control unit 8.

[0016] Predetermined information is input to the input unit 4 in response to operations by a user who uses the analytical computer 3. The input unit 4 includes input devices such as a mouse and a keyboard. That is, the predetermined information is input to the input unit 4 in response to operations of these input devices.

[0017] The output unit 5 outputs predetermined information. The apparatus state determination system 1 includes a device for displaying images, such as a display that outputs the analysis results. In other words, the output unit 5 controls the images displayed on the display. The display may be separate from the computer main body, or may be integrated with the computer main body.

[0018] The apparatus state determination system 1 may control images displayed on a display of another computer connected via a network. In this case, the output unit 5 of the other computer may control the output of the analysis results.

[0019] In this embodiment, a display is exemplified as a device that displays an image, but other modes may be used. For example, an image may be displayed using a head-mounted display or a projector. Furthermore, a printer that prints information on a paper medium may be used instead of a display. In other words, the objects controlled by the output unit 5 may include a head-mounted display, a projector, or a printer.

[0020] The communication unit 6 communicates with other computers via a communication line such as the Internet. In this embodiment, the device state determination system 1 and other computers are connected to each other via the Internet, but other configurations are also possible. For example, the device state determination system 1 and other computers may be connected to each other via a LAN (Local Area Network), a WAN (Wide Area Network), or a mobile communication network.

[0021] The storage unit 7 stores various information required for evaluating the state of the device 50. The storage unit 7 includes a predetermined database. This database is a collection of information stored in a memory, a HDD, or a cloud computing resource and organized so that it can be searched or accumulated.

[0022] The main control unit 8 includes a model construction unit 9, an analysis unit 10, a learning unit 11, a physical quantity measurement unit 12, a measured data processing unit 13, a parameter identification unit 14, a state determination unit 15, a setting unit 16, a remaining life evaluation unit 17, an inspection estimation unit 18, a feature extraction unit 19, a noise processing unit 20, a state change detection unit 21, a model determination unit 22, a model update unit 23, and an abnormality cause estimation unit 24. These are realized by executing programs stored in a memory or HDD by the CPU or GPU.

[0023] Note that each component of the analysis computer 3 does not necessarily have to be provided in one computer. For example, one analysis computer 3 may be realized by a plurality of computers connected to each other via a network. For example, the learning unit 11 may be provided in a computer separate from the analysis computer 3.

[0024] 2, the device 50 to be diagnosed is, for example, one installed in a nuclear power plant. The device 50 is made up of a plurality of devices 51. For example, the device 50 is made up of a control panel 51A, an electric motor 51B, a rotating body 51C, and a group of devices 51D on the load side.

[0025] Here, the control panel 51A corresponds to a component (control section) related to control of the behavior of the electric motor 51B. For example, the control panel 51A controls the rotation speed or torque of the electric motor 51B.

[0026] Rotating body 51C is a component that transmits energy output from electric motor 51B to load side devices 51D.

[0027] The load side devices 51D correspond to components that consume the energy transmitted from the rotor 51C. The load side devices 51D include, for example, a pump, a fan, and gears.

[0028] Here, it is assumed that the rotating electric machine is an electric motor 51B. That is, the device 50 is composed of a plurality of devices 51 including at least a rotating electric machine. The rotating electric machine may be a generator. This embodiment can be applied regardless of whether the rotating electric machine is the electric motor 51B or a generator.

[0029] For example, suppose there is a parameter for a predetermined frequency related to a rotating electric machine. This frequency may be either the frequency of the power output from the rotating electric machine or the frequency of the power supplied to the rotating electric machine. If the rotating electric machine is an electric motor 51B, this frequency is the frequency of the AC supplied from the outside to this electric motor 51B. Also, if the rotating electric machine is a generator, this frequency is the frequency of the AC output from this generator to the outside.

[0030] Each sensor 2 of the equipment state determination system 1 is installed in an equipment 50. The sensor 2 detects at least one of vibration, current, voltage, magnetic flux, temperature, pressure, rotational torque, and load as a physical quantity. In this way, the sensor 2 can obtain the physical quantity for any type of device 51.

[0031] For example, sensors 2 are provided on control panel 51A, motor 51B, rotor 51C, and load side equipment group 51D. Predetermined information acquired by these sensors 2 is input to analysis computer 3. Note that sensors 2 may not only be pre-installed on device 50, but may also be installed each time inspection is performed. Also, the location of already installed sensors 2 may be changed.

[0032] These sensors 2 detect physical quantities that change depending on the state of the device 50. Specific examples of the physical quantities include the current and voltage in the control panel 51A and the electric motor 51B, the temperature of the electric motor 51B, the vibration of the electric motor 51B and the rotating body 51C, the magnetic flux, the load, and the torque of the electric motor 51B.

[0033] The model construction unit 9 (FIG. 1) constructs a physical model that reproduces the operation of the device 50 and reproduces the state of the device 51 that occurs in relation to the operation. The physical model is a virtual model that can simulate physical quantities related to the device 50. By using the physical model, it is possible to not only simulate the device 51 operating normally, but also to simulate the device 51 operating abnormally.

[0034] The physical model reproduces the behavior of the device 50 by a combination of parameters. For example, the physical model includes at least one of a dynamic model, a control model, an electrical model, a magnetic model, a thermal model, a fluid model, a physical property model, and a structural model.

[0035] Here, the mechanical model represents the behavior of the mechanical system related to at least one of the rotational motion and translational motion of each device 51 such as the electric motor 51B, the rotor 51C, and the load side device group 51D. The control model is for the control panel 51A as a control unit constituting at least one device 51 to control the behavior of the entire device 50. The electrical model represents the behavior of the electrical system of the device 50. The magnetic model represents the behavior of the magnetic system of the device 50. The thermal model represents the thermal behavior of the device 50. The fluid model represents the behavior of at least one of the flow rate and pressure of the fluid flowing through the piping connected to the pump constituting at least one device 51. The physical property model represents the material properties of each device 51. The structural model has at least one of the information on the dimensions and shape of each device 51. In this way, the behavior of the device 50 can be precisely reproduced according to various models.

[0036] In other words, the physical model is a model that models and formulates the structure or function of the device 50. A group of equations corresponding to these are then formulated.

[0037] The physical model also includes an abnormality model that indicates an abnormality in each of the devices 51 that constitute the device 50. For example, as shown in Figs. 4 and 5, a coupling can be considered as a rotating body 51C. Eccentricity e from the center of rotation caused by improper installation of the coupling c or declination d c The effect of eccentricity e is formulated as a model. c or declination d c Parameters that indicate the degree of abnormality, such as the size of the

[0038] Furthermore, a mathematical model in which input and output are associated may be used as a part of the model. In this way, the function of the device 51 for which there is no design information can be incorporated into the model and analyzed. Note that the data required to construct the mathematical model may be actual device data or may be obtained in advance by element testing. Also, the results of a separately performed analysis may be used.

[0039] Abnormalities of the rotating body 51C include misalignment of the bearings, breakage of the rotor bar or short-circuit ring of the motor 51B, protrusion of the rotor bar, protrusion of the coil inside the stator or slot, bending or cracking of the rotating body 51C, etc. Abnormalities of the rotating body 51C include unbalance (missing parts, residual unbalance), short circuit or burnout of the stator winding, unbalanced power supply voltage, irregular waves of the inverter, defects in the rolling bearings (inner ring, outer ring, rolling elements), etc. Parameters related to each abnormality are defined.

[0040] Here, the parameters related to the abnormality may be determined in advance by measurement. For example, the eccentricity occurring in a flange coupling can be measured from the displacement of the drive side and load side flanges. In this way, the number of parameters in the physical model can be reduced, which makes it possible to improve the diagnosis accuracy and reduce the calculation cost.

[0041] The analysis unit 10 (FIG. 1) performs calculations by reflecting parameters in an equation (group of equations) formulated based on a physical model. The calculations include structural analysis using the finite element method (FEM) and the like.

[0042] Here, the physical quantity measuring unit 12 (FIG. 1) acquires, as actual measurement data, physical quantities actually measured by each device 51. This physical quantity measuring unit 12 acquires, as time-series data, the physical quantities detected by the sensor 2 (FIG. 1). For example, as shown in FIG. 3, the physical quantity measuring unit 12 acquires, as actual measurement data, time-series data including physical quantities indicating current, magnetic flux, acceleration, angular velocity, torque, displacement, pressure, and the like from each device 51.

[0043] In the present embodiment, a mode in which the physical quantity measuring unit 12 acquires a physical quantity in real time using the sensor 2 during operation of the device 50 is illustrated, but other modes may be used. For example, during regular inspection, the sensor 2 may be attached to the equipment 51 to acquire and record a physical quantity, and the physical quantity measuring unit 12 may acquire the record at a later date.

[0044] That is, the physical quantity measuring unit 12 acquires, as actual measurement data, at least one of the physical quantities detected by the respective sensors 2 and the physical quantities included in the test data or inspection data obtained when a test or inspection was previously performed on the device 51. In this way, it is possible to determine the presence or absence of an abnormality in the device 50 based on at least one of the current physical quantities and the past physical quantities.

[0045] The test data or inspection data also includes information indicating whether the device 51 has failed. In other words, the physical quantity includes information indicating whether the device 51 has failed in the past. The physical quantity also includes information indicating the date, time, location, and state of the past failure.

[0046] It is preferable that the data acquired by the physical quantity measuring unit 12 includes a plurality of types of physical quantities. For example, since vibration is affected by the support structure characteristics of the device 51 or the installation state of the sensor 2, it is difficult to determine whether or not the change in physical quantity is due to an abnormality from the vibration data alone. In such a case, a comprehensive determination can be made from physical quantities such as current in addition to vibration.

[0047] Furthermore, the data acquired by the physical quantity measuring unit 12 may include data acquired while the rotating device is stopped. In this way, data on external noise unrelated to the operation of the rotating device can be acquired.

[0048] The data acquired by the physical quantity measuring unit 12 may include data acquired while the rotating equipment is operating at low speed. In this way, it is possible to acquire noise data caused by factors unrelated to vibration caused by rotation. Such factors include, for example, unevenness such as scratches on the shaft surface, and displacement vibration caused by shaft bending.

[0049] Furthermore, the actual measurement data processing unit 13 (FIG. 1) converts the actual measurement data acquired by the physical quantity measuring unit 12 into analysis data in a format compatible with the calculations of the analysis unit 10. Here, the calculations executed by the analysis unit 10 include calculations for outputting the analysis data in the same format as that output from the actual measurement data processing unit 13.

[0050] For example, as shown in Fig. 3, when the measured data processing unit 13 outputs vibration spectrum data, the corresponding calculation results of the analysis unit 10 are also subjected to a Fourier transform (Fast Fourier Transform: FFT) calculation to output the vibration spectrum data. Here, the time-series data as the measured data is converted into data for analysis. In this way, the measured data can be converted into a format optimal for calculation, allowing the analysis unit 10 to perform calculations at high speed.

[0051] The measured data processing unit 13 converts the measured data, such as vibration, current, voltage, and magnetic flux, acquired by the physical quantity measuring unit 12 into spectrum data by Fourier transform. The measured data processing unit 13 also performs filtering as necessary. The measured data processing unit 13 also performs processing to convert time-series data, such as temperature and pressure, into maximum values, minimum values, average values, effective values, overall values ​​(OA values), and the like, as necessary.

[0052] Furthermore, feature extraction unit 19 (FIG. 1) extracts features from the calculation results of analysis unit 10 and the actual measurement data. For example, feature extraction unit 19 processes the calculation results of analysis unit 10 and extracts feature amounts in normal and abnormal cases. Here, the calculations executed by analysis unit 10 include calculations for outputting the results in a format appropriate for processing in feature extraction unit 19.

[0053] When the feature extraction unit 19 extracts features related to the vibration spectrum data, the analysis unit 10 performs an FFT calculation to output the vibration spectrum data. For example, when the analysis unit 10 extracts the amplitude or OA value of the vibration data as a feature, it performs a waveform processing calculation and a filtering calculation such as a high-pass filter or a low-pass filter to select a frequency domain as the feature.

[0054] Furthermore, in the case where the time series change of the spectrum is used as the feature, the analysis unit 10 performs imaging calculation to visualize the result of the short-time FFT calculation.The analysis unit 10 also performs statistical calculation to extract statistical features based on a plurality of calculation results.The analysis unit 10 also performs correlation calculation to extract feature values ​​related to correlations between physical model parameters and calculation results, etc.

[0055] The feature extraction unit 19 extracts the feature by comparing the calculation results in the analysis unit 10 when an abnormality occurs with those in normal times. For example, the relationship between frequency and spectrum is shown in Fig. 7. Here, G1 shows a graph in normal times, and G2 shows a graph in abnormal times. As shown in this graph, when the vibration spectrum at an arbitrary position of the rotating equipment is calculated by the analysis unit 10 for a certain abnormal event, if the frequency component spectrum has changed compared to normal times, the spectrum of this frequency component is taken as the feature.

[0056] The feature amount may be extracted from the calculation results when multiple abnormal events occur. In this way, it is possible to make a judgment even when multiple abnormalities occur in the device 50.

[0057] Here, the feature extraction unit 19 may extract features from the calculation results at a plurality of positions. In this way, the state of the device 50 can be determined comprehensively from the data acquired by the plurality of sensors 2 installed in the device 50.

[0058] Furthermore, the feature extraction unit 19 may extract the calculation result of the analysis unit 10 as the feature. In this way, more information can be extracted as the feature.

[0059] The feature extraction unit 19 may also statistically extract features from the results of calculations performed under multiple conditions for a certain parameter. In this way, although there are cases where the calculation results for a certain parameter are not different from the calculation results under normal conditions, it is possible to comprehensively extract features from a wide parameter range.

[0060] The feature amount may be set by a sensitivity analysis of the calculation result for the parameter. In this way, the cause of the abnormality that has a larger effect on the device 51 can be diagnosed efficiently.

[0061] The extraction of such features can be performed not only on vibration spectrum data, but also on waveform data, image data, statistical data, correlation data, etc. Also, a combination of these data may be used.

[0062] Furthermore, the feature extraction unit 19 may extract features related to noise from the analysis data processed by the actual measurement data processing unit 13. For example, when the rotation shaft displacement is measured by the sensor 2, scratches on the surface of the rotation shaft are detected as noise.

[0063] Figure 8 is a graph showing the relationship between time and the displacement of the rotation axis. As shown in this graph, multiple spike-like vibration waveforms W are generated due to noise. For such waveforms, the magnitude of the displacement with respect to the change in time is set as the feature amount.

[0064] Furthermore, the feature extraction unit 19 extracts noise features from the correlation between the analysis data from each sensor 2 processed by the actual measurement data processing unit 13. For example, in the case where vibration spectrum data is acquired from a casing or a stand supporting the rotating body 51C and vibration spectrum data of the rotating shaft are acquired, a signal that has no correlation with these data is determined to be a noise feature.

[0065] Furthermore, the feature extraction unit 19 may determine that the data acquired while the rotating device is stopped is a feature related to noise. In this way, it is possible to extract feature amounts of noise such as vibrations of external devices unrelated to the operation of the rotating device or measurement errors of the sensor 2.

[0066] The feature extraction unit 19 may also determine that data acquired while the rotating equipment is operating at low speed is a feature related to noise. In this way, noise data due to factors unrelated to vibration caused by rotation can be acquired. For example, feature amounts of noise such as scratches or irregularities on the shaft surface, displacement vibration due to shaft bending, etc. can be extracted.

[0067] The learning unit 11 (FIG. 1) inputs the features extracted by the feature extraction unit 19 into a learning model as teacher data (training data) to perform machine learning. For example, the learning unit 11 learns the features extracted by the feature extraction unit 19 and the respective parameters corresponding thereto as teacher data. When the analysis unit 10 has derived a calculation result corresponding to a combination of a plurality of parameters in advance, the learning unit 11 learns the respective parameters for the features extracted by the feature extraction unit 19 and the calculation results corresponding to these parameters as teacher data. In this way, the accuracy of learning can be improved.

[0068] Furthermore, the learning unit 11 can learn, as teacher data, the feature amount relating to noise extracted by the feature amount extraction unit 19. Furthermore, the learning unit 11 can learn, as teacher data, the actually measured data.

[0069] Furthermore, the learning unit 11 can learn the parameters when the setting unit 16 determines that the calculation result of the analysis unit 10 is an abnormal state and the abnormal label (reference information) associated with this parameter as teacher data.

[0070] As shown in Fig. 6, the learning unit 11 generates a learning model by machine learning using, as training data, feature quantities extracted from results of calculations performed in advance under various parameter conditions in the analysis unit 10. For example, there is a learning model generated by machine learning using a neural network. Here, the learning unit 11 creates a learning model using the parameters used in the calculations performed by the analysis unit 10 and the calculation results as training data.

[0071] The calculation results of the analysis unit 10 and the corresponding parameters may be used as the teacher data for the learning unit 11. When learning is performed for a plurality of physical quantities, the calculation results and the feature quantities may be combined and used as the teacher data. In this way, the accuracy of the determination of anomalies can be improved.

[0072] The state change detection unit 21 (FIG. 1) determines (evaluates) whether there is a significant difference between newly acquired actual measurement data and previously acquired actual measurement data. The state change detection unit 21 also determines whether there is a state change from the difference between the newly acquired actual measurement data and previously acquired actual measurement data based on a learning model learned by the learning unit 11 using the actual measurement data as teacher data. The state change detection unit 21 can detect a change from the normal operating state of the device 50.

[0073] Here, the detection of the difference may be performed by a learning model that is learned in advance by the learning unit 11 using previously acquired data as teacher data. Also, the learning unit 11 may perform unsupervised learning such as clustering.

[0074] The data targeted by the state change detection unit 21 is at least one of the feature amount extracted by the feature amount extraction unit 19 and the analysis data processed by the actual measurement data processing unit 13. Here, the analysis data may be a single physical quantity or multiple physical quantities.

[0075] The noise processing unit 20 (FIG. 1) removes noise from the actual measurement data based on the features extracted by the feature extraction unit 19. For example, the noise processing unit 20 generates processed data from the analysis data processed by the actual measurement data processing unit 13, from which noise has been removed, based on the features extracted by the feature extraction unit 19.

[0076] The relationship between frequency and spectrum is shown in Fig. 9. Graphs of test run data and simulation data of the rotating equipment are shown in Fig. 9. As shown in this graph, it is assumed that the feature extraction unit 19 extracts the rotational frequency component and its double component as features from the vibration spectrum data calculated by the analysis unit 10.

[0077] In this case, the noise processing unit 20 judges other frequency component spectra included in the vibration spectrum of the analysis data to be noise. In this way, the influence of noise that cannot be reproduced by a physical model can be removed. Furthermore, noise such as vibration of external equipment unrelated to the operation of the rotating equipment or measurement error of the sensor 2 can be removed from the feature amount related to noise while the device 50 is stopped, extracted by the feature amount extraction unit 19. Furthermore, noise can be removed from the feature amount related to unevenness of the rotating shaft while the device 50 is operating at low speed.

[0078] As an example, we will show a method for removing noise related to unevenness of the rotating shaft from data acquired during low-speed rotational motion. Data during low-speed driving is less affected by vibrations caused by rotation compared to data during normal driving, so it is possible to obtain the effects of unevenness of the rotating shaft. Then, by taking the difference between the low-speed driving data and the data during normal driving, the effects of vibrations caused by rotational motion can be removed.

[0079] When calculating the difference between the spectra of the normal driving data and the low-speed driving data, the noise processor 20 performs a scale conversion of the frequency axis to remove the influence of unevenness of the rotation axis. For example, the rotation frequency is obtained by a rotation pulse meter as each piece of driving data, and the ratio of these is used to perform a scale conversion so that the frequency axis of the spectrum of the low-speed driving data matches the frequency axis of the spectrum of the normal driving data.

[0080] At this time, the number of data is adjusted so that the frequency resolution of the spectrum of the low speed driving data matches that of the normal driving data.

[0081] Moreover, the noise may be removed by a learning model related to noise learned by the learning unit 11. That is, the noise processing unit 20 may remove the noise based on a learning model learned using the feature amount related to noise extracted by the feature amount extraction unit 19 as teacher data.

[0082] These processes allow the extraction of feature quantities from the actual measurement data and the removal of unnecessary noise. In addition, processes are performed to deal with errors and variations that occur during measurement. For example, there are measurement errors or statistical errors in measuring instruments, so processes are performed to deal with these.

[0083] The device state determination system 1 includes a computer equipped with an algorithm for performing machine learning. In other words, it includes a computer equipped with artificial intelligence (AI) for performing machine learning. For example, the device state determination system 1 may be configured with one computer equipped with a neural network, or may be configured with multiple computers equipped with neural networks.

[0084] The parameter identification unit 14 (FIG. 1) identifies parameters of the physical model so that the calculation result of the analysis unit 10 reproduces the actual measurement data from which noise has been removed by the noise processing unit 20. The parameter identification unit 14 also identifies actual measurement parameters that cause the physical model to reproduce the actual measurement data, based on the learning model learned by the learning unit 11. The parameter identification unit 14 also identifies actual measurement parameters that are parameters when the calculation result of the analysis unit 10 reproduces the actual measurement data from which noise has been removed by the noise processing unit 20, based on the learned learning model.

[0085] Here, the processed data obtained by performing noise processing on the analysis data is used. That is, the parameters of the physical model are identified based on a learning model in which the parameters are learned by the learning unit 11 so that the features of the feature extraction unit 19 reproduce the processed data. In other words, the analysis unit 10 performs an analysis using the physical model so that the reproduction result matches the processed data.

[0086] Furthermore, the parameter identification unit 14 estimates an optimal combination of parameters for the physical model of the device 50 to reproduce the processed data output from the noise processing unit 20, and outputs these as actual measurement parameters. By reproducing the processed data output from the noise processing unit 20, estimation is performed taking into account noise, errors, and variations that cannot be reproduced by the physical model.

[0087] The device state determination system 1 (FIG. 1) determines, in the parameter identification unit 14, actual measurement parameters that reproduce the processed data output from the noise processing unit 20, using a trained model generated by, for example, a neural network.

[0088] For example, in the case of the electric motor 51B, the parameters include those related to the dimensions and shape of the electric motor 51B, the number of coil windings or slots, etc. The parameters also include the characteristics of the components, such as electrical resistance. The parameters also include the operating conditions or the surrounding environment of the device 50. These parameters may be obtained based on the specifications of each device 51 constituting the device 50, or may be obtained by actual testing, or may be obtained by simulation. The parameters also include those related to abnormalities, such as eccentricity and unbalance.

[0089] The learning unit 11 (FIG. 1) may be configured with a plurality of learning models. For example, the learning unit 11 may have a learning model in which a feature amount for identifying a parameter is used as teacher data, a learning model in which a feature amount related to noise for noise processing is used as teacher data, a learning model for detecting deviation from the normal operating state of the device 50, and the like.

[0090] The analysis unit 10 (FIG. 1) executes calculations by reflecting the parameters identified by the parameter identification unit 14. For example, the analysis unit 10 obtains calculation results related to time changes in the load acting on each device 51 constituting the apparatus 50 or the position of a rotating body 51C. The analysis unit 10 also evaluates numerical errors occurring in the calculations.

[0091] The model determination unit 22 (FIG. 1) determines whether or not the calculation results of the analysis unit 10 or the features extracted by the feature extraction unit 19 reproduce the actual measurement data.

[0092] For example, the model determination unit 22 determines (evaluates) the reproducibility of the physical model from the difference between the result or the feature value of the calculation reflecting the actual measurement parameters identified by the parameter identification unit 14 in the analysis unit 10 and the analysis data or the processed data. For example, when the state change detection unit 21 determines that there is a state change and the model determination unit 22 determines that the reproducibility of the physical model is low, it can determine that an abnormal event that does not exist in the abnormality model considered in the physical model has occurred. Furthermore, the presence or absence of an abnormality can be detected for abnormal events not included in the physical model.

[0093] Furthermore, the model determining section 22 may evaluate the reproducibility of a plurality of pieces of analysis data or processed data from the variation in the results when the parameters are identified by the parameter identifying section 14 .

[0094] The model update unit 23 (FIG. 1) updates the physical model when the model determination unit 22 determines that the physical model does not reproduce the actual measurement data. For example, when the model determination unit 22 determines that the degree of reproduction of the physical model is low, the model update unit 23 updates the physical model so as to reproduce the analysis data or the processed data.

[0095] The setting unit 16 (FIG. 1) sets the threshold value of the tolerance range for judging the presence or absence of an abnormality in the device 50. This tolerance range is determined based on whether the result of calculation performed on a certain parameter in the analysis unit 10 satisfies conditions such as the functional requirements, material properties, and legal restrictions of one device 51 or the device 50 composed of multiple devices 51. In this way, even when setting the tolerance range for a device 51 that has never experienced an abnormality in the past, it is possible to set it rationally rather than using standards based on the inspector's experience.

[0096] The allowable range may be set not only for the calculation result but also for the parameters. In this way, the analysis unit 10 can determine the presence or absence of an abnormality from the parameters identified by the parameter identification unit 14.

[0097] In addition, in the case where the results of calculations performed in the analysis unit 10 reflecting the actual measured parameters identified by the parameter identification unit 14 reproduce the operation of the device 50 that was previously determined to be abnormal, the allowable range may be set based on the physical quantities or parameters at that time.

[0098] There are variations and errors in sensing, simulation, parameter identification, and material properties. Therefore, the threshold of the tolerance range is set taking into consideration the statistical distribution. The threshold of the tolerance range may be set with a margin for the standard set from the required conditions. In this way, it is possible to evaluate the symptoms of an abnormality immediately before it occurs, and to repair the device 51 or replace the parts.

[0099] As an example of a method for determining the allowable range, consider a case in which vibration occurs due to centrifugal force caused by imbalance of rotor 51C, for example, and radial load P acting on the rolling bearing supporting rotor 51C changes, as shown in Fig. 11. Radial load P is calculated in analysis unit 10 based on a physical model. For example, basic rated life L10 of the rolling bearing is calculated by the following remaining life evaluation formula.

[0100] L10=(C / P) p

[0101] Here, C is the basic dynamic load rating, which indicates the dynamic load capacity of a rolling bearing. The basic dynamic load rating is a constant load that gives a basic rated life of L10 of 1 million revolutions, for example. Also, p is an index that changes depending on the type of rolling bearing. For example, for ball bearings, p = 3. For roller bearings, p = 10 / 3.

[0102] A threshold value of the allowable range for vibration can be set from the calculation result of the vibration at any position of the rotating body 51C when the life of the rolling bearing does not satisfy the required operation period due to this radial load P. In this way, a threshold value of the allowable range for vibration can be set at a position where the sensor 2 that detects vibration can be installed on the actual rotating body 51C. At this time, the threshold value for vibration may be imposed, for example, on the amplitude of the vibration or on the spectrum for each frequency of the FFT.

[0103] Furthermore, in the case where the rotating body 51C is connected to the electric motor 51B, the physical quantity acquired by the sensor 2 may be a physical quantity other than the vibration, such as the winding current of the electric motor 51B, leakage flux, etc., or may be a combination of a plurality of these. In this way, it becomes possible to set a threshold value of an allowable range for the device 50 in which the sensor 2 for detecting vibration cannot be installed, or for an event that cannot be determined by vibration. Furthermore, a physical quantity or parameter appropriate for determining an abnormality may be selected in advance by sensitivity analysis or the like.

[0104] The remaining life assessment unit 17 (FIG. 1) may assess the remaining life of any device 50 by, for example, using Miner's rule (modified Miner's rule) determined from material properties for the fluctuating stress obtained by the analysis unit 10. In this case, the stress frequency distribution of the actual stress can be obtained by using the rainflow method or the like.

[0105] Here, the number of repetitions at which a fracture occurs due to a repeated stress σi of a certain stress amplitude in the SN curve of the target material is Ni. For example, suppose that a given object is subjected to k repeated stresses σi (i=1 to k) with different amplitudes, each repeated ni (i=1 to k) times. In this case, the fatigue damage D accumulated in the object is given by the following formula. Here, the life is defined as when D reaches 1.

[0106]

number

[0107] The stress frequency distribution can be calculated in the analysis unit 10 based on a physical model. Here, a physical quantity or parameter when it is determined that the time until the fatigue damage D reaches 1 does not satisfy the required operation period may be set as a threshold value of the allowable range.

[0108] Furthermore, the load result calculated by the analysis unit 10 may be a condition on material properties, that is, a threshold value of an allowable range may be set for a physical quantity or parameter when breakage, plastic deformation, or the like occurs.

[0109] In this embodiment, the remaining life is determined not only from the load and stress, but also from the deterioration of insulating materials due to time-dependent changes in voltage or current, and deterioration of materials due to thermal cycles, etc. Based on this determination, a threshold value of the allowable range may be set.

[0110] Although the above example illustrates a method for setting the threshold value of the allowable range for the mechanical behavior of the device 50, the allowable threshold value may be set for other systems as well. For example, examples of physical quantities to be set include temperature, current, voltage, magnetic flux (density), sound, flow rate, and pressure. Threshold values ​​may be set for parameters related to these.

[0111] The setting unit 16 (FIG. 1) calculates thresholds of allowable ranges for physical quantities and parameters based on various required conditions. Then, among the thresholds calculated for the target physical quantities or parameters, the threshold with the smallest allowable range is set as the threshold to be used for determining the presence or absence of abnormality signs.

[0112] The state determination unit 15 determines whether or not there is an abnormality based on the allowable range set by the setting unit 16 .

[0113] When the state determination unit 15 determines that the device 50 is normal, the inspection estimation unit 18 (FIG. 1) estimates the progress of the abnormality from the combination and magnitude of the measured parameters, and estimates the next inspection location and inspection time from the remaining life of the equipment 51 calculated by the remaining life evaluation unit 17. In this way, the inspection location and inspection time can be automatically presented to the user.

[0114] When the state determination unit 15 determines that the device 50 is abnormal, the abnormality cause estimation unit 24 (FIG. 1) estimates the cause of the abnormality based on the abnormality label as reference information.

[0115] The setting unit 16 (FIG. 1) sets reference information for determining the state of the device 50. The reference information is set based on the calculation result of the analysis unit 10, and includes an abnormality label indicating at least one of the presence or absence of an abnormality in the device 50, the presence or absence of an abnormal symptom, an abnormality event, a cause of the abnormality, a degree of the abnormality, and a type of the abnormality.

[0116] For example, the setting unit 16 sets an abnormality label for determining the presence or absence of an abnormality in the device 50. By performing machine learning using this abnormality label, the learning model is an analysis result corresponding to a combination of multiple parameters, and it becomes possible to determine whether the calculation result of the analysis unit 10 indicates an abnormal state of the device 50.

[0117] The setting unit 16 sets an abnormal label based on the combination of parameters when the device 50 is determined to be abnormal and the calculation result of the analysis unit 10. In this way, the abnormality in the device 50 can be appropriately evaluated.

[0118] The criterion for the judgment may be whether or not the results of calculations performed by the analysis unit 10 on predetermined parameters satisfy the requirements of the apparatus 50 composed of a plurality of devices 51, such as functional requirements, material properties, and legal restrictions.

[0119] Here, examples of functional requirements include rotation speed, torque, flow rate, pressure, drive speed, response, temperature, voltage, current, etc. Examples of material properties include fatigue life, thermal cycle life, plastic deformation, load leading to fracture, etc. Examples of laws and regulations include legal restrictions such as sound volume, chemical substance emissions, and temperature.

[0120] The setting unit 16 sets an abnormality label (reference information) based on whether or not the calculation result of the analysis unit 10 satisfies at least any of the conditions of the required functions, material properties, and laws and regulations of the device 50. In this way, it becomes possible to set a rational evaluation standard, rather than a standard based on the inspector's experience, when determining whether or not there is an abnormality in the device 50 equipped with the device 51 that has never experienced an abnormality in the past.

[0121] The abnormality label may be set not only for the calculation result of the analysis unit 10 but also for the parameters of the physical model. In this way, an abnormality in the device 50 can be determined from the parameters identified by the parameter identification unit 14.

[0122] The remaining life evaluation unit 17 (FIG. 1) calculates the remaining life of each device 51 based on the calculation results of the analysis unit 10 when the parameters identified from the actual measurement data by the parameter identification unit 14 are reflected in the physical model. In this way, it is possible to evaluate the remaining life for a combination of multiple parameters.

[0123] Furthermore, when the measured parameters identified by the parameter identification unit 14 are reflected in the physical model, the remaining life evaluation unit 17 evaluates the remaining life from the results of determining at least one of the load, stress, voltage, current, and temperature related to each device 51 obtained by the analysis unit 10. In this way, the remaining life can be evaluated taking into account the load or stress applied to the devices 51.

[0124] For example, the remaining life evaluation unit 17 calculates the load and stress changes over time for each of the devices 51 constituting the apparatus 50, and judges the remaining life from the remaining life evaluation formula and material properties. In this embodiment, it is also possible to judge the remaining life from not only the load and stress, but also the deterioration of insulating materials due to changes in voltage or current over time, and material deterioration due to thermal cycles (changes in temperature over time).

[0125] The state determination unit 15 (FIG. 1) determines the state of the device 50 based on the parameters identified by the parameter identification unit 14 when the actual measurement data is reflected in the parameters of the physical model and the actual measurement data is analyzed. For example, the state determination unit 15 determines the state of the device 50 based on a learning model that has been trained using teacher data when the actual measurement parameters identified by the parameter identification unit 14 are analyzed. Note that the state determination unit 15 may determine the state of the device 50 based on reference information.

[0126] Furthermore, the setting unit 16 sets an abnormality label based on a comparison between the remaining life calculated by the remaining life evaluation unit 17 and a pre-planned operation period. In this way, it is possible to evaluate whether an abnormality will occur in the device 50 during a pre-planned operation period. For example, the presence or absence of an abnormality can be evaluated by comparing the remaining life of each device 51 with the planned operation time.

[0127] Furthermore, the state determination unit 15 determines the presence or absence of an abnormality in the device 50 by comparing the actual measurement parameters identified by the parameter identification unit 14 with the arbitrarily set allowable ranges for the parameters. In this way, the process of determining the presence or absence of an abnormality in the device 50 can be easily performed using the arbitrarily set allowable ranges for the parameters.

[0128] Furthermore, the state determination unit 15 determines the presence or absence of an abnormality in the device 50 by comparing the actual measurement data acquired by the physical quantity measurement unit 12 with an arbitrarily set allowable range for data. In this way, the process of determining the presence or absence of an abnormality in the device 50 can be easily performed using the arbitrarily set allowable range for data.

[0129] For example, the state determination unit 15 determines the presence or absence of an abnormality by comparing the actual measurement data acquired by the physical quantity measurement unit 12 with a data allowable range. The state determination unit 15 also determines the presence or absence of an abnormality by comparing the analysis data output from the data processing unit with a allowable range.

[0130] The parameter allowable ranges and data allowable ranges used for the determination by the state determination unit 15 are set in advance. These allowable ranges may be arbitrarily set in advance by the user. In addition, all of these allowable ranges may be the same for all the devices 50, or may be different for each individual device 50.

[0131] These allowable ranges may be set based on actual measurement data, may be set based on specifications relating to the behavior of the device 50, may be set based on actual test results, may be set based on simulation results, or may be set based on a combination of these results.

[0132] When it is determined that the device 50 is abnormal, the state determination unit 15 (FIG. 1) identifies the degree and cause of the abnormality. In this way, the degree and cause of the abnormality can be identified based on the actual measurement data. Also, the degree of the abnormality can be evaluated from the combination of the actual measurement parameters and the magnitude of their values.

[0133] The main control unit 8 controls the output unit 5 to output the evaluation result of the apparatus state determination system 1. For example, the main control unit 8 outputs the results derived by the state determination unit 15 and the inspection estimation unit 18.

[0134] Next, the device state determination method (processing) executed by the device state determination system 1 will be described with reference to the flowchart of Fig. 10. The above-mentioned drawings may be referenced. The following steps are at least a part of the processing included in the device state determination method, and other steps may be included in the device state determination method.

[0135] First, in step S1, the model construction unit 9 constructs a physical model that reproduces the operation of the apparatus 50 composed of a plurality of devices 51 including at least a rotating electric machine, and reproduces the state of the devices 51 that occurs in relation to the operation.

[0136] In the next step S2, the analysis unit 10 performs calculations by reflecting the parameters in the equation formulated based on the physical model.

[0137] In the next step S3, the setting unit 16 sets reference information for determining the state of the device 50. Here, the setting unit 16 sets abnormal labels as reference information, and performs processing to associate the abnormal labels with the respective teaching data.

[0138] Here, the noise processor 20 may perform a process of removing noise from teacher data that includes noise. The noise processor 20 may also perform a process of adding noise to teacher data that does not include noise. Furthermore, the noise processor 20 may perform a process of generating teacher data that includes a mixture of noisy and noisy teacher data.

[0139] In the next step S4, the learning unit 11 inputs the calculation results of the analysis unit 10 and the corresponding parameters to the learning model as teacher data. Here, machine learning is performed.

[0140] In the next step S5, the physical quantity measuring unit 12 acquires the physical quantities actually measured by each device 51 as actual measurement data.

[0141] In the next step S6, the measured data processing unit 13 converts the measured data acquired by the physical quantity measuring unit 12 into analysis data in a format compatible with the calculations of the analysis unit 10. This analysis data is reflected in the parameters of the physical model, and the analysis unit 10 analyzes the measured data.

[0142] In the next step S7, the feature extraction unit 19 extracts features from the calculation results of the analysis unit 10 and the actual measurement data.

[0143] In the next step S8, the noise processor 20 removes noise from the actual measurement data based on the feature extracted by the feature extractor 19.

[0144] In the next step S9, the parameter identification unit 14 identifies parameters of the physical model so that the calculation result of the analysis unit 10 reproduces the analysis data (actual measurement data) from which noise has been removed by the noise processing unit 20. Here, the actual measurement parameters, which are parameters for reproducing the actual measurement data, are identified based on the learning model that has been learned by the learning unit 11.

[0145] In the next step S10, the state determination unit 15 determines the state of the device 50 based on the actual measurement parameters identified by the parameter identification unit 14. For example, the state determination unit 15 determines whether or not there is an abnormality in the device 50 at the current time based on each of the actual measurement parameters identified by the parameter identification unit 14. Here, if there is no abnormality (NO in step S10), the process proceeds to step S11. On the other hand, if there is an abnormality (YES in step S10), the process proceeds to step S13.

[0146] In step S11, which is reached if the answer is NO in step S10, the remaining life evaluation unit 17 calculates the remaining life of each device 51 based on the calculation results of the analysis unit 10 when the actual measured parameters identified by the parameter identification unit 14 are reflected.

[0147] In the next step S12, the inspection estimation unit 18 estimates the progress of the abnormality from the combination and magnitude of the measured parameters, and estimates the next inspection location and inspection time from the remaining life of the device 51 calculated by the remaining life evaluation unit 17. The main control unit 8 controls the output unit 5 to output the evaluation result of the equipment state determination system 1. Then, the process ends.

[0148] In step S13, which is reached if the answer is YES in step S10, the anomaly cause estimation unit 24 identifies the degree and cause of the anomaly when it is determined that the apparatus 50 is abnormal. For example, an anomaly label is set based on the combination of parameters when the apparatus 50 is determined to be abnormal and the calculation result of the analysis unit 10. Based on this anomaly label, the anomaly cause estimation unit 24 identifies the degree and cause of the anomaly. The main control unit 8 controls the output unit 5 to output the evaluation result of the apparatus state determination system 1. Then, the process ends.

[0149] In the above-mentioned flowchart, the steps are executed in series, but the order of the steps is not necessarily fixed, and the order of some steps may be interchanged. Also, some steps may be executed in parallel with other steps.

[0150] The equipment state determination system 1 (FIG. 1) may include a deep learning unit that extracts a specific pattern from a plurality of patterns based on deep learning.

[0151] The analysis using the computer of this embodiment can use an analysis technique based on learning of artificial intelligence. For example, a learning model generated by machine learning using a neural network, a learning model generated by other machine learning, a deep learning algorithm, a mathematical algorithm such as regression analysis, etc. In addition, the form of machine learning includes forms such as clustering and deep learning.

[0152] Here, a neural network is a mathematical model that represents the characteristics of brain functions through computer simulation. For example, this model shows that artificial neurons (nodes) that form a network through synaptic connections change the strength of synaptic connections through learning, and acquire problem-solving abilities. Furthermore, neural networks acquire problem-solving abilities through deep learning.

[0153] For example, a neural network is provided with an intermediate layer having multiple layers. Each intermediate layer is composed of multiple units. In addition, by having the multi-layered neural network learn in advance using teacher data, it is possible to automatically extract feature quantities in patterns of changes in the state of a circuit or system. Note that the multi-layered neural network can be configured on a user interface with any number of intermediate layers, any number of units, any learning rate, any number of learning iterations, and any activation function.

[0154] In addition, deep reinforcement learning may be used in the neural network, in which a reward function is set for each information item to be learned, and the information item with the highest value is extracted based on the reward function.

[0155] For example, a convolutional neural network (CNN), which has a proven track record in image recognition, is used. In this CNN, the intermediate layer is composed of a convolutional layer and a pooling layer. The convolutional layer obtains a feature map by applying filtering to nearby nodes in the previous layer. The pooling layer further reduces the feature map output from the convolutional layer to create a new feature map. At this time, by obtaining the maximum value of the pixels contained in the region of interest in the feature map, it is possible to absorb slight deviations in the position of the feature amount.

[0156] The convolutional layer extracts local features from an image, and the pooling layer performs processing to consolidate local features. In these processes, the image is reduced while maintaining the features of the input image. In other words, the convolutional neural network (CNN) can significantly compress (abstract) the amount of information contained in an image. Then, the input image can be recognized and classified using the abstracted image image stored in the neural network.

[0157] Note that there are various machine learning techniques such as autoencoder, LSTM (Long Short-Term Memory), SDF (Signed Distance Function), GAN (Generative Adversarial Network), RNN (Recurrent Neural Network), etc. These techniques may be applied to the machine learning of the present embodiment.

[0158] Next, a modified example will be described with reference to Fig. 11. Note that the above-mentioned drawings will be referred to as appropriate, and the same components as those shown in the above-mentioned embodiment will be given the same reference numerals and will not be described again. The configuration applied in this modified example may be applied to the above-mentioned embodiment, or may be appropriately combined with the configuration of the above-mentioned embodiment.

[0159] The reference information in the modified example includes a predetermined threshold value. The threshold value is set based on a data tolerance range that indicates a tolerance range of the actual measurement data. The data tolerance range itself may be treated as the reference information. In the modified example, the state of the device 50 is evaluated without using a learning model.

[0160] The setting unit 16 (FIG. 1) of the modified example sets a threshold value of an allowable range for the measured data as a method of judging the presence or absence of an abnormality in the device 50. Furthermore, the state judgment unit 15 (FIG. 1) of the modified example judges the state of the device 50 based on the threshold value when the measured data is analyzed. In this way, the evaluation standard becomes clear by setting the threshold value, and the state of the device 50 can be appropriately evaluated. Furthermore, even when the parameters cannot be identified, it is possible to judge the presence or absence of an abnormality by assuming all abnormality events. For example, when there is little measured data that can be acquired from the sensor 2 and the accuracy of identifying parameters is poor, it is possible to judge the presence or absence of an abnormality from the measured data.

[0161] Furthermore, when the reference information is the data allowable range, the state determination unit 15 compares the actual measurement data with the data allowable range to determine the state of the device 50. In this way, the evaluation criteria become clear by setting the data allowable range, and the state of the device 50 can be appropriately evaluated.

[0162] The thresholds or data allowable ranges set by the setting unit 16 may be arbitrarily set in advance by the user. Furthermore, all of the thresholds or data allowable ranges may be the same for all devices 50, or may be different for each device 50.

[0163] For example, as shown in Fig. 11, the analysis unit 10 (Fig. 1) calculates the bearing life in advance for a plurality of parameters. If the result is shorter than the required operating period, a threshold value indicating the allowable range of vibration is set for at least one vibration point. At this time, the threshold value for vibration may be imposed, for example, on the amplitude of the vibration or on the spectrum for each frequency of the FFT.

[0164] The setting unit 16 calculates thresholds indicating the allowable range of vibration based on various required conditions. Then, among the thresholds calculated for the target physical quantity or parameter, the threshold with the smallest allowable range is set as the threshold to be used for determining the sign of an abnormality. In this way, even if the parameter cannot be identified, it is possible to determine the presence or absence of an abnormality while assuming all possible abnormal events.

[0165] 11 shows an example in which vibration is used as the measured data, but other measured data may be used, or these measured data may be combined. Furthermore, the measured data may be, for example, current, voltage, magnetic flux, temperature, etc.

[0166] In the above example, the computer constituting the device state determination system 1 performs various determinations (including various processes such as judgment, evaluation, estimation, and setting), but other aspects are also possible. For example, a user may perform some of the above-mentioned various determinations, and the computer may receive an input of the determination results and use them for its own determination.

[0167] The above-mentioned device state determination system 1 includes a control device, a storage device, an output device, an input device, and a communication interface. Here, the control device includes a highly integrated processor such as a central processing unit (CPU), a graphics processing unit (GPU), a field programmable gate array (FPGA), or a dedicated chip. The storage device includes a read only memory (ROM), a random access memory (RAM), a hard disk drive (HDD), a solid state drive (SSD), etc. The output device includes a display panel, a head mounted display, a projector, a printer, etc. The input device includes a mouse, a keyboard, a touch panel, etc. This device state determination system 1 can be realized by a hardware configuration using a normal computer.

[0168] The program executed by the device state determination system 1 is provided by being pre-installed in a ROM or the like. Additionally or alternatively, the program is provided by being stored in a computer-readable non-transitory storage medium as a file in an installable or executable format. The storage medium includes a CD-ROM, a CD-R, a memory card, a DVD, a flexible disk (FD), and the like.

[0169] The programs executed by this equipment status determination system 1 may be stored in a computer connected to a network such as the Internet and provided by downloading them via the network. This equipment status determination system 1 may also be configured by combining separate modules that independently fulfill the functions of the components and connect them to each other via a network or dedicated lines.

[0170] According to the embodiment described above, by providing the noise processing unit 20 that removes noise from the actual measurement data based on the feature extracted by the feature extraction unit 19, it is possible to determine the state of the device 50 even if the actual measurement data contains noise. Furthermore, by removing data unrelated to abnormalities such as noise from the actual measurement data, it is possible to improve the accuracy of determination using a simulation based on a physical model.

[0171] Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, modifications, and combinations can be made without departing from the spirit of the invention. These embodiments or modifications thereof are within the scope of the invention and its equivalents as described in the claims, as well as within the scope and spirit of the invention. [Explanation of symbols]

[0172] 1...equipment status determination system, 2...sensor, 3...analysis computer, 4...input section, 5...output section, 6...communication section, 7...memory section, 8...main control section, 9...model construction section, 10...analysis section, 11...learning section, 12...physical quantity measurement section, 13...measurement data processing section, 14...parameter identification section, 15...status determination section, 16...setting section, 17...remaining life evaluation section, 18...inspection estimation section, 19...feature extraction section, 20...noise processing section, 21...status change detection section, 22...model determination section, 23...model update section, 24...anomaly cause estimation section, 50...equipment, 51...equipment, 51A...control panel, 51B...electric motor, 51C...rotating body, 51D...group of equipment.

Claims

1. a model construction unit that reproduces the operation of an apparatus including a plurality of devices including at least a rotating electric machine and constructs a physical model that reproduces the states of the devices that occur in relation to the operation; an analysis unit that performs calculations by reflecting parameters in an equation formulated based on the physical model; a physical quantity measuring unit that acquires physical quantities actually measured by each of the devices as actual measurement data; a feature extraction unit that extracts feature values ​​from the calculation results of the analysis unit and the actual measurement data; a noise processing unit that removes noise from the actual measurement data based on the feature amount extracted by the feature amount extraction unit; a parameter identification unit that identifies the parameters of the physical model so that the calculation result of the analysis unit reproduces the actual measurement data from which the noise has been removed by the noise processing unit; a state determination unit that determines a state of the device based on the parameters identified by the parameter identification unit; Equipped with Equipment status determination system.

2. a learning unit that inputs the feature amount extracted by the feature amount extraction unit into a learning model as teacher data and performs machine learning. The apparatus state determination system according to claim 1 .

3. a state change detection unit that determines whether there is a difference between the newly acquired actual measurement data and the previously acquired actual measurement data; 3. The apparatus state determination system according to claim 1 or 2.

4. when the analysis unit has derived the calculation results corresponding to a combination of a plurality of the parameters in advance, the learning unit learns, as the teacher data, the parameters for the features extracted by the feature extraction unit and the calculation results corresponding to these parameters. The apparatus state determination system according to claim 2 .

5. The learning unit learns the feature amount related to the noise extracted by the feature amount extraction unit as the teacher data. The apparatus state determination system according to claim 2 .

6. The learning unit learns the actual measurement data as the teacher data. The apparatus state determination system according to claim 2 .

7. The parameter identification unit identifies actual measurement parameters, which are the parameters when the calculation result reproduces the actual measurement data from which the noise has been removed by the noise processing unit, based on the learned learning model. The apparatus state determination system according to claim 4 .

8. the noise processing unit removes the noise based on the learning model that has been trained using the feature amount related to the noise extracted by the feature amount extraction unit as the teacher data. The apparatus state determination system according to claim 5 .

9. a learning unit that inputs the feature amounts extracted by the feature amount extraction unit into a learning model as teacher data and performs machine learning; a state change detection unit that determines whether there is a difference between the newly acquired actual measurement data and the previously acquired actual measurement data; Equipped with the state change detection unit determines whether or not there is a state change from a difference between the newly acquired actual measurement data and the previously acquired actual measurement data based on the learning model learned by the learning unit using the actual measurement data as the teacher data. The apparatus state determination system according to claim 1 .

10. a model determination unit that determines whether the calculation result of the analysis unit or the feature amount extracted by the feature amount extraction unit reproduces the actual measurement data, 3. The apparatus state determination system according to claim 1 or 2.

11. a model updating unit that updates the physical model when the model determining unit determines that the physical model does not reproduce the actual measurement data, The apparatus state determination system according to claim 10.

12. a setting unit that sets reference information for determining a state of the device, the reference information is set based on the calculation result, and includes an abnormality label indicating at least one of the presence or absence of an abnormality in the device, the presence or absence of an abnormality symptom, an abnormality event, a cause of the abnormality, a degree of the abnormality, and a type of the abnormality; the state determination unit, when the actual measurement parameters identified by the parameter identification unit are analyzed, determines a state of the device based on the learning model that has been trained using the teacher data to which the anomaly label is associated. The apparatus state determination system according to claim 7.

13. The learning unit learns, as the teacher data, the parameters and the abnormal label when the calculation result of the analysis unit is determined to be an abnormal state by the setting unit. The equipment state determination system according to claim 12.

14. The setting unit sets the abnormal label based on whether or not the calculation result of the analysis unit satisfies at least any of the conditions of the required function of the device, the material characteristics, and laws and regulations. The apparatus state determination system according to claim 12 or 13.

15. a remaining life evaluation unit that calculates a remaining life of each of the devices based on the calculation result when the parameters identified by the actual measurement data by the parameter identification unit are reflected in the physical model, The apparatus state determination system according to claim 12 or 13.

16. the setting unit sets the abnormal label based on a comparison between the remaining life calculated by the remaining life evaluation unit and a pre-planned operation period. The apparatus state determination system according to claim 15.

17. the remaining life evaluation unit evaluates the remaining life from a result of determining at least one of a load, a stress, a voltage, a current, and a temperature associated with each of the devices, which is obtained by the analysis unit when the parameters identified by the parameter identification unit are reflected in the physical model; The apparatus state determination system according to claim 15.

18. an anomaly cause estimation unit that estimates a cause of the anomaly based on the anomaly label when the state determination unit determines that the device is abnormal; The apparatus state determination system according to claim 12 or 13.

19. The reference information includes a predetermined threshold value, the state determination unit determines a state of the device based on the threshold value when the actual measurement data is analyzed. The apparatus state determination system according to claim 12 or 13.

20. the reference information includes a data tolerance range indicating a tolerance range of the actual measurement data, the state determination unit determines the state of the device by comparing the actual measurement data with the data allowable range. The apparatus state determination system according to claim 12 or 13.

21. a measurement data processing unit that converts the measurement data acquired by the physical quantity measuring unit into analysis data in a format compatible with the calculation by the analysis unit, 3. The apparatus state determination system according to claim 1 or 2.

22. The physical model is A dynamic model representing the behavior of a dynamic system related to at least one of rotational motion and translational motion of each of the devices; A control model for a control unit constituting at least one of the devices to control the behavior of the entire device; an electrical model representing the behavior of an electrical system of the device; a magnetic model representing the behavior of the magnetic system of the device; a thermal model describing the thermal behavior of the device; A fluid model representing at least one of the behaviors of a flow rate and a pressure of a fluid flowing through a pipe connected to a pump constituting at least one of the devices; A material model representing material properties of each of the devices; A structural model having information on at least one of dimensions and shape of each of the devices; At least one of Reproducing the behavior of the device by a combination of the parameters; 3. The apparatus state determination system according to claim 1 or 2.

23. a model construction unit constructing a physical model that reproduces an operation of an apparatus including a plurality of devices including at least a rotating electric machine and reproduces a state of the devices that occurs in association with the operation; an analysis unit performing a calculation by reflecting parameters in an equation formulated based on the physical model; A physical quantity measuring unit acquires physical quantities actually measured by each of the devices as actual measurement data; A feature extraction unit extracts feature amounts from the calculation result of the analysis unit and the actual measurement data; a noise processing unit removing noise from the actual measurement data based on the feature amount extracted by the feature amount extraction unit; a parameter identification unit identifying the parameters of the physical model such that the calculation result of the analysis unit reproduces the actual measurement data from which the noise has been removed by the noise processing unit; A state determination unit determines a state of the device based on the parameters identified by the parameter identification unit; Including, A method for determining the state of an apparatus.

Images