Motor diagnosis device and motor diagnosis system

By performing frequency analysis on the current and voltage signals of the motor and removing sideband wave components, the problem of false detection under the influence of inverter noise in the existing technology is solved, and high-precision motor anomaly diagnosis is achieved.

CN123125499APending Publication Date: 2026-11-17MITSUBISHI ELECTRIC CORP
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202480091144.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-04-26
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing motor diagnostic devices struggle to achieve high-precision diagnosis when removing inverter noise, leading to an increased likelihood of false detections.

Method used

By detecting the current and voltage signals in the motor, frequency analysis is used to remove sideband wave components, and high-precision spectrum analysis is performed in conjunction with the noise determination unit to identify abnormalities in the motor.

Benefits of technology

It achieves high-precision diagnosis of motor abnormalities, reduces the possibility of false detections, and improves the reliability and accuracy of diagnosis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123125499A_ABST
    Figure CN123125499A_ABST
Patent Text Reader

Abstract

The electric motor diagnostic device (60) of the present invention includes a detection unit (20) for detecting current signals and voltage signals of an electric motor (1); and a control unit (50) for detecting whether the electric motor (1) has any abnormality based at least on the current signal. The control unit (50) detects sideband wave components based on a first spectrum obtained by frequency analysis of the current signal, performs first control to remove a first sideband wave component whose frequency changes by a first set amount or more according to the load fluctuation of the electric motor (1) from the first spectrum, and detects whether the electric motor (1) has any abnormality based on the first spectrum after removing the first sideband wave component.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to electric motor diagnostic devices and electric motor diagnostic systems. Background Technology

[0002] Electric motors are used to power devices and machinery in production lines of process industries. Loads for electric motors include pumps, compressors, fans, and industrial robots, with a wide range of applications and increasing demand. For stable and continuous operation of these devices and machinery, the electric motor needs to run reliably and continuously. However, not all electric motors operate in suitable environments; they are frequently subjected to high-stress environments such as high temperature, high humidity, heavy loads, corrosion, and abrasion.

[0003] Previously, condition diagnosis of equipment and machinery was mostly conducted by maintenance personnel using sensory inspections based on time-based maintenance plans (TBM). However, for particularly critical components like electric motors, it is necessary to periodically diagnose the presence or absence of faults. Therefore, this method of diagnosis by monitoring personnel presents a significant cost problem.

[0004] Therefore, Condition-Based Maintenance (CBM) technology for motors is receiving increasing attention. Currently, the diagnosis of inverter-driven motors is achieved by installing various sensors and other measuring devices on each motor. These measuring devices include torque meters, speed and acceleration vibration sensors, etc. However, this method is impractical for motor control centers that centrally manage hundreds or thousands of motors. Therefore, there is a need for a device that diagnoses inverter-driven motors based on measurable current, voltage, and other information available in the motor control center, without requiring additional specialized sensors or measuring instruments, to ensure reliability, production efficiency, and operational integrity.

[0005] Motor diagnostics using current is called Motor Current Signature Analysis (MCSA), which diagnoses the motor based on the analysis results of the frequency components of the current. However, due to the influence of the environment in which the motor is installed, the noise superimposed in the power supply increases, leading to the possibility of false detections. Therefore, to improve the accuracy of motor diagnostics, the following diagnostic aids have been disclosed in the past.

[0006] That is, the diagnostic auxiliary device of the conventional electric motor diagnostic device has: an acquisition unit for acquiring the q-axis current of the power conversion device that controls the rotating electric motor; an amplitude calculation unit for calculating the amplitude of each frequency related to the acquired q-axis current based on the acquired q-axis current; and an output unit for outputting information related to the amplitude of each frequency. Between the power conversion device and the amplitude calculation unit, there is a filter unit for removing predetermined frequencies (for example, see Patent Document 1).

[0007] [Existing Technical Documents]

[0008] [Patent Document 1] International Publication No. WO2019 / 198563 Summary of the Invention

[0009] The technical problem that the invention aims to solve

[0010] In the aforementioned existing diagnostic auxiliary devices used as motor diagnostic devices, the focus is on the q-axis current of the power conversion device, and a method is adopted to improve diagnostic accuracy by removing specific noise components through filters such as low-pass filters and high-pass filters.

[0011] However, while this method can remove noise in a specific frequency band through filters such as low-pass and high-pass filters, it sometimes fails to remove inverter noise caused by the inverter driving the motor with high precision.

[0012] This disclosure discloses a technique for solving the above-mentioned problems, with the aim of providing a motor diagnostic device and motor diagnostic system that can accurately diagnose the presence or absence of abnormalities in a motor.

[0013] Technical means for solving technical problems

[0014] The electric motor diagnostic device disclosed herein includes: The detection unit detects a current signal of the current flowing in the motor and a voltage signal of the voltage applied to the motor; and The control unit detects whether there is any abnormality in the motor based on at least the detected current signal. The control unit detects the sideband wave component of the power supply frequency of the motor based on a first spectrum obtained by frequency analysis of the current signal, performs a first control to remove the first sideband wave component from the first spectrum, and detects whether the motor has any abnormality based on the first spectrum after removing the first sideband wave component. The first sideband wave component is the sideband wave component whose frequency changes by more than a first set amount according to the load fluctuation of the motor.

[0015] Furthermore, the electric motor diagnostic system disclosed herein includes: The electric motor diagnostic device configured as described above; and The notification department, which is connected to the motor diagnostic device via a network, displays the diagnostic results of the motor diagnostic device to the administrator.

[0016] Invention Effects

[0017] According to the electric motor diagnostic device and electric motor diagnostic system disclosed herein, it is possible to obtain an electric motor diagnostic device and electric motor diagnostic system that can accurately diagnose the presence or absence of abnormalities in an electric motor. Attached Figure Description

[0018] Figure 1 This is a block diagram showing the general structure of the motor diagnostic system according to Embodiment 1.

[0019] Figure 2 This is a diagram showing an example of the hardware of the arithmetic processing unit in Embodiment 1.

[0020] Figure 3 This is a block diagram showing the structure of the arithmetic processing unit of the motor diagnostic device according to Embodiment 1.

[0021] Figure 4 This is a flowchart illustrating the control of the arithmetic processing unit in Embodiment 1.

[0022] Figure 5 This is a flowchart showing the control of the noise determination process performed by the noise determination unit of the computational processing unit in Embodiment 1.

[0023] Figure 6 This is a flowchart showing the control of the floor noise estimation process performed by the noise determination unit of the computational processing unit in Embodiment 1.

[0024] Figure 7 This is a flowchart showing the control of the load fluctuation noise estimation process performed by the noise determination unit of the computational processing unit in Embodiment 1.

[0025] Figure 8 This is a flowchart showing the control of the load fluctuation noise determination execution process performed by the noise determination unit of the computational processing unit in Embodiment 1.

[0026] Figure 9 This is a graph showing the first spectrum of the base noise estimation process performed by the noise determination unit of the computational processing unit for explaining Embodiment 1.

[0027] Figure 10 This is a graph showing the first spectrum of the base noise estimation process performed by the noise determination unit of the computational processing unit for explaining Embodiment 1.

[0028] Figure 11middle, Figure 11 A is a histogram showing the basis noise estimation process performed by the noise determination unit of the computational processing unit in embodiment 1. Figure 11 B is a graph showing the first spectrum of the base noise estimation process performed by the noise determination unit of the computational processing unit for explaining Embodiment 1.

[0029] In Figure 12, Figure 12A This is a graph showing the first spectrum at a motor load rate of 29.9% during the load fluctuation noise estimation process performed by the noise determination unit of the computational processing unit for explaining Embodiment 1. Figure 12B This is a graph showing the first spectrum at a motor load rate of 42.2% during the load fluctuation noise estimation process performed by the noise determination unit 530 of the computational processing unit for explaining Embodiment 1. Figure 12C This is a graph showing the first spectrum at a motor load rate of 55.2% during the load fluctuation noise estimation process performed by the noise determination unit of the computational processing unit for explaining Embodiment 1. Figure 12D This is a graph showing the first spectrum at a motor load rate of 75.3% during the load fluctuation noise estimation process performed by the noise determination unit of the computational processing unit for explaining Embodiment 1. Figure 12E This is a graph showing the first spectrum at a motor load rate of 88.3% during the load fluctuation noise estimation process performed by the noise determination unit of the computational processing unit for explaining Embodiment 1.

[0030] Figure 13 This is a graph showing the relationship between the load rate and the frequency of the characteristic frequency component in Embodiment 1, and the relationship between the load rate and the frequency of the inverter noise component.

[0031] Figure 14 This is a diagram illustrating the voltage noise estimation process of Embodiment 1.

[0032] Figure 15 This is a schematic diagram of the normal waveform prediction unit in Implementation 1 when it has AI functionality.

[0033] Figure 16 middle, Figure 16 A is a graph showing the characteristic quantities of the input data under a load rate of 29.9% for the normal waveform estimation unit of Embodiment 1. Figure 16 B is a graph showing the characteristic quantities of the input data at a load rate of 42.2% for the normal waveform estimation section of Embodiment 1. Figure 16 C is a graph showing the characteristic quantities of the input data at a load rate of 55.2% for the normal waveform estimation section of Embodiment 1.

[0034] Figure 17 This diagram illustrates the AI ​​function performed by the normal waveform estimation unit in Embodiment 1.

[0035] Figure 18 This is a diagram illustrating AI control in the normal waveform prediction unit of Embodiment 1.

[0036] Figure 19 This diagram illustrates the averaging process performed by the averaging calculation unit in Embodiment 1.

[0037] Figure 20 This is a flowchart illustrating the control of the abnormal state comparison unit of the arithmetic processing unit in Embodiment 1. Detailed Implementation

[0038] Implementation method 1.

[0039] Figure 1 This is a block diagram showing the schematic structure of the motor diagnostic system 100 according to Embodiment 1.

[0040] The motor diagnostic system 100 includes a motor diagnostic device 60 and a notification unit 70. The diagnostic device 60 diagnoses whether there are any abnormalities such as bearing damage in the motor 1 in the main circuit 10, and the notification unit 70 notifies the monitoring personnel who are performing maintenance of the diagnostic results.

[0041] The following description illustrates an example of using the motor diagnostic system 100 to monitor one motor 1, but it is also possible to monitor multiple motors 1.

[0042] like Figure 1 As shown, the main circuit 10 connected to the power system K includes a motor 1 such as a three-phase induction motor as a load, a circuit breaker 2 for disconnecting the circuit, an electromagnetic contactor 3 for opening and closing the circuit via an electromagnet, and a detector 4 for changing the current flowing through the circuit to the motor 1 and the voltage applied to the motor 1. The motor 1 is connected to a mechanical device 6 as a load and drives the operation of the mechanical device 6.

[0043] The motor diagnostic device 60 includes a measurement circuit 20 as a detection unit and an arithmetic processing unit 50 as a control unit.

[0044] The measurement circuit 20 detects the current signal and voltage signal transformed by the detector 4.

[0045] The arithmetic processing unit 50 uses the current signal or voltage signal input from the measurement circuit 20 to detect the presence or absence of abnormalities in loads such as the motor 1 and mechanical equipment 6.

[0046] The structure of the notification section 70 will be explained.

[0047] The notification unit 70 includes an output circuit unit 71, a drive circuit 72, a display unit 73, and a communication circuit 74.

[0048] The output circuit unit 71 outputs signals such as signals indicating abnormal status of the motor 1 and warnings input from the arithmetic processing unit 50 to the outside to notify the administrator.

[0049] The drive circuit 72 outputs a control signal that causes the electromagnetic contactor 3 to open and close, based on the signal indicating the state of the motor 1 input from the arithmetic processing unit 50.

[0050] The display unit 73 displays an alarm indicating the abnormal state based on the signal indicating the state of the motor 1 input from the arithmetic processing unit 50, for example, when an abnormality of the motor 1 is detected, and notifies the monitoring personnel.

[0051] The communication circuit 74 outputs information such as the diagnostic results of the motor 1 input from the arithmetic processing unit 50 to the external monitoring device 80.

[0052] Furthermore, the notification unit 70 is not limited to the structure described above, which includes the output circuit unit 71, the drive circuit 72, the display unit 73, and the communication circuit 74. For example, the notification unit 70 may also include only the display unit 73. Alternatively, the notification unit 70 may include a speaker that outputs sound, as long as it has the function of notifying the monitoring personnel of the calculation results in the calculation processing unit 50.

[0053] Furthermore, the external monitoring device 80, which is composed of a PC (personal computer) or similar device, receives information from the processing unit 50 via the communication circuit 74 to monitor the operating status of the motor diagnostic system 100. The connection between the external monitoring device 80 and the motor diagnostic system 100 can be via cable or wireless connection. Alternatively, a network can be established between multiple motor diagnostic systems 100, connecting the external monitoring device 80 via the Internet.

[0054] Next, the general structure of the arithmetic processing unit 50 will be explained.

[0055] The arithmetic processing unit 50 includes a processor 51, a storage device 52 serving as a storage unit, and a rated information setting circuit 53.

[0056] The rated information setting circuit 53 receives inputs such as the power supply frequency, rated output, rated current, number of poles, and rated speed of the motor 1. The rated information input to the rated information setting circuit 53 is stored in the storage device 52. This rated information is easily obtained by consulting the manufacturer's product catalog or the nameplate affixed to the motor 1.

[0057] In addition, when there are multiple motors 1 that are to be diagnosed, the rated information of all motors 1 that are to be diagnosed needs to be input into the rated information setting circuit 53 in advance.

[0058] Figure 2 This is a diagram showing an example of the hardware of the arithmetic processing unit 50.

[0059] As described above, the arithmetic processing unit 50, which serves as a control device, is configured with a processor 51, a storage device 52, and a rated information setting circuit 53. The storage device 52 includes volatile storage devices such as random access memory (not shown) and non-volatile auxiliary storage devices such as flash memory.

[0060] Alternatively, an auxiliary storage device, such as a hard disk, may be included instead of flash memory. The processor 51 executes a program input from the storage device 52. In this case, the program is input to the processor 51 from the auxiliary storage device via a volatile storage device.

[0061] In addition, the processor 51 sometimes uses data recorded in the storage device 52 or the rated information setting circuit 53 to execute programs. Furthermore, the processor 51 can output calculation results and other data to the volatile storage device of the storage device 52, or save data to an auxiliary storage device via the volatile storage device.

[0062] Next, the detailed structure and operation of the arithmetic processing unit 50 will be explained.

[0063] Figure 3 This is a block diagram showing the structure of the arithmetic processing unit 50 of the motor diagnostic device 60 according to Embodiment 1.

[0064] Figure 4 This is a flowchart illustrating the control of the arithmetic processing unit 50 in Embodiment 1.

[0065] Figure 5 This is a flowchart showing the control of the noise determination process performed by the noise determination unit 530 of the computation processing unit 50 in Embodiment 1.

[0066] Figure 6 This is a flowchart showing the control of the background noise estimation process performed by the noise determination unit 530 of the computation processing unit 50 in Embodiment 1.

[0067] Figure 7 This is a flowchart showing the control of the load fluctuation noise estimation process performed by the noise determination unit 530 of the computation processing unit 50 in Embodiment 1.

[0068] Figure 8 This is a flowchart showing the control of the load fluctuation noise determination execution process performed by the noise determination unit 530 of the computation processing unit 50 in Embodiment 1.

[0069] Figure 9This is a graph showing the first spectrum of the noise estimation process of the noise determination unit 530 of the computational processing unit 50 in Embodiment 1.

[0070] Figure 10 This is a graph showing the first spectrum of the noise estimation process of the noise determination unit 530 of the computational processing unit 50 in Embodiment 1.

[0071] Figure 11 A is a histogram showing the base noise estimation process of the noise determination unit 530 of the computational processing unit 50 in Embodiment 1.

[0072] Figure 11 B is a graph showing the first spectrum of the base noise estimation process of the noise determination unit 530 of the computational processing unit 50 in embodiment 1.

[0073] Figure 12A This is a graph showing the first spectrum of the motor 1 at a load rate of 29.9% for the load fluctuation noise estimation process of the noise determination unit 530 of the calculation processing unit 50 in Embodiment 1.

[0074] Figure 12B This is a graph showing the first spectrum of the motor 1 at a load rate of 42.2% for the load fluctuation noise estimation process of the noise determination unit 530 of the calculation processing unit 50 in Embodiment 1.

[0075] Figure 12C This is a graph showing the first spectrum of the motor 1 at a load rate of 55.2% for the load fluctuation noise estimation process of the noise determination unit 530 of the calculation processing unit 50 in Embodiment 1.

[0076] Figure 12D This is a graph showing the first spectrum of the motor 1 at a load rate of 75.3% for the load fluctuation noise estimation process of the noise determination unit 530 of the computational processing unit 50 in Embodiment 1.

[0077] Figure 12E This is a graph showing the first spectrum of the motor 1 at a load rate of 88.3% for the load fluctuation noise estimation process of the noise determination unit 530 of the calculation processing unit 50 of Embodiment 1.

[0078] Figure 13 It is a graph showing the relationship between load rate and the frequency of characteristic frequency component (fs±fr), and the relationship between load rate and the frequency of inverter noise component.

[0079] like Figure 3As shown, the processor 51 of the arithmetic processing unit 50 includes a load rate calculation unit 501, a sampling frequency calculation unit 502, an FFT (Fast Fourier Transform) analysis unit 503, a normalization processing unit 504, a peak detection calculation unit 506, a rotating frequency band detection unit 507, a rotating spectrum value detection unit 508, a frequency axis transformation calculation unit 509, an averaging calculation unit 511, a rotating frequency σ value calculation unit 512, a threshold calculation unit 513, a sideband wave extraction unit 520, a noise determination unit 530, a matrix selection unit 540, an FFT analysis result 541, a normal waveform prediction unit 550, and an abnormal state comparison unit 560.

[0080] In addition, the storage device 52 includes a rotating spectrum value moving average cache 590, a correction value data storage unit 591, and a normal state storage unit 592.

[0081] The diagnosis of motor 1 can be roughly divided into two categories: the learning stage of learning the initial state (normal state) of motor 1, and the diagnostic stage of diagnosing the presence or absence of abnormalities.

[0082] Figure 4 The flowchart shown illustrates the control during the learning phase. First, the initial state of motor 1 is learned as the normal state of motor 1. Then, in the diagnostic phase described later, the mechanism of the arithmetic processing unit 50 is to diagnose the presence or absence of an abnormality based on the difference from the normal state.

[0083] <Learning Stage>

[0084] When the control of the arithmetic processing unit 50 is started, the arithmetic processing unit 50 first detects whether the motor 1 has started to operate (step S101).

[0085] Next, the arithmetic processing unit 50 acquires the current signal of the current flowing in the motor 1 and the voltage signal of the voltage applied to the motor 1 via the measurement circuit 20 (step S102).

[0086] Next, the load rate calculation unit 501 calculates the load rate of the motor 1 based on the acquired current signal and voltage signal (step S103).

[0087] Next, the sampling frequency calculation unit 502 calculates the power supply frequency fs based on the input current signal or voltage signal, and calculates the sampling frequency used to determine the measurement interval of the measurement circuit 20.

[0088] Next, the FFT analysis unit 503 performs FFT analysis on the input current signal and derives the first spectrum after decomposing the current signal into frequency components (step S104).

[0089] Next, the normalization processing unit 504 performs normalization processing on the derived first spectrum (step S105).

[0090] Next, the peak detection calculation unit 506 detects all peak components in the normalized first spectrum (step S106).

[0091] The peak processing procedure (step S106A) for processing the peak portion detected by the peak detection calculation unit 506 will be described below.

[0092] The rotating frequency band detection unit 507 extracts the peak position caused by the rotation frequency fr of the motor 1 from the multiple detected peak positions.

[0093] Specifically, the rotating frequency band detection unit 507 calculates the rotational frequency fr of the motor 1 based on the rated rotational speed recorded in the storage device 52. Then, the rotating frequency band detection unit 507 extracts peak portions (fs±fr) generated at positions corresponding to the calculated rotational frequency fr on both sides of the power supply frequency fs of the motor 1. The rotating frequency band detection unit 114 defines the frequency band containing this peak portion (fs±fr) and its error as the rotating frequency band W.

[0094] Next, the rotation spectrum value detection unit 508 detects the spectrum value of the signal strength peak value of the selected peak part (fs±fr) caused by the rotation frequency fr.

[0095] Next, the frequency axis transformation calculation unit 509 performs a correction process that transforms the frequency axis of the peak part (fs±fr) caused by the rotation frequency fr in the first spectrum that has been derived multiple times to match, for example, the load rate of the motor 1 when it is unloaded.

[0096] The frequency axis transformation calculation unit 509 stores the spectrum values ​​of the multiple peak portions (fs±fr) aligned with the frequency axis after the above correction into the moving average cache 590.

[0097] Next, the averaging calculation unit 511 averages the spectral values ​​of multiple peak portions (fs±fr) aligned with the frequency axis stored in the moving average cache 590.

[0098] For example, the averaging operation unit 511 reduces the spectral value of a peak that occurs only once to one-tenth of its value by performing averaging processing 10 times, thereby reducing the floor noise.

[0099] Next, the rotation frequency σ value calculation unit 512 uses the stored value of the moving average cache 121 to calculate the dispersion σ of the spectrum value of the peak part (fs±fr) caused by the rotation frequency fr.

[0100] Next, the threshold calculation unit 513 determines the threshold for abnormality diagnosis based on the dispersion σ of the spectral values ​​of the peak position (fs±fr).

[0101] Specifically, the difference between the spectral value of the power supply frequency fs and the spectral value of the peak part (fs±fr) is related to the degree of deterioration of the motor 1. Therefore, the threshold calculation unit 513 can determine the threshold based on the above difference as described above.

[0102] Next, the sideband wave extraction unit 520 extracts whether there are peaks on both sides of the power supply frequency fs in the frequency region outside the rotating frequency band W based on the peak portion (fs±fr) in the rotating frequency band W after averaging by the averaging calculation unit 511.

[0103] Subsequently, the peak portions (fs±fr) generated on both sides of the power frequency fs within the rotating frequency band W, and the peak portions (fs±A·fr) generated outside the rotating frequency band W, will all be referred to as sideband wave components.

[0104] Furthermore, as mentioned above, the spectral values ​​of the sideband wave component (fs±fr) are related to the degree of degradation of motor 1 and can be used to diagnose the presence or absence of abnormalities in motor 1. Therefore, in the following description, this sideband wave component (fs±fr) will sometimes be referred to as the characteristic frequency component (fs±fr).

[0105] The above is the processing performed in step S106A for the peak portion within the first spectrum.

[0106] The following describes the control process performed by the noise determination unit 530 in the noise determination procedure.

[0107] The noise determination unit 530 performs a noise determination process for detecting noise in the first frequency spectrum (step S107).

[0108] like Figure 5 As shown, the noise determination process includes a base noise estimation process (step S130), a load fluctuation noise estimation process (step S131), and a voltage noise estimation process (step S132).

[0109] First, use Figure 6 , Figures 9-11 The process of inferring the base noise (step S130) is described.

[0110] Generally speaking, the first spectrum derived from the current signal of the motor 1 driven by the inverter will generate a great deal of background noise caused by factors such as timing deviations of the inverter's switching, and the signal strength of this background noise is extremely strong.

[0111] Therefore, this base noise will be superimposed on the characteristic frequency component (fs±fr) in the first spectrum used for anomaly determination of motor 1, which will sometimes lead to misdiagnosis of anomalies.

[0112] To prevent misdiagnosis, a method is needed to distinguish between the characteristic frequency component (fs±fr) used for anomaly detection in motor 1 and the floor noise. This floor noise estimation process sets a signal strength threshold for distinguishing between the characteristic frequency component (fs±fr) and the floor noise.

[0113] First, the noise determination unit 530 acquires the first spectrum derived by the FFT analysis unit 503 (step S130A).

[0114] like Figure 9 As shown, in the first spectrum, mainly between -70dB and -100dB in signal strength, a pulsation in signal strength that increases with decreasing frequency can be confirmed.

[0115] Next, the noise determination unit 530 derives a second regression equation representing the relationship between the frequency components in the first spectrum and the signal strength by performing a low-order polynomial approximation regression analysis, thereby calculating the pulsation component of the base noise (step S130B).

[0116] Next, the noise determination unit 530 performs a first control (step S130C) by subtracting the calculated pulsating component of the base noise, i.e., the signal strength of each frequency component determined by the second regression equation, from the first spectrum. Through this subtraction process, the pulsating component of the base noise is removed from the first spectrum.

[0117] like Figure 10 As shown, it can be confirmed that the pulsations that increase signal strength as the frequency decreases are eliminated. In this example, a cubic polynomial approximation using the least squares method was performed. By using polynomial approximation in this way, dynamic pulsation prediction can be made.

[0118] Next, the noise determination unit 530 generates, based on the first spectrum after removing the pulsating component of the base noise, as shown in the figure. Figure 11 Figure A shows a histogram representing the data density of the frequency components that occur as the signal strength.

[0119] exist Figure 11 In example A, the noise determination unit 530, based on the signal strength (around -90dB) where the data density of the occurrence frequency reaches its maximum (8.5%) in the histogram, sets a data density (3%) that is one set value smaller than the reference value towards the side where the signal strength increases (step S130D).

[0120] Furthermore, the aforementioned reference signal strength, the aforementioned set value, the aforementioned data density, etc., are not limited to the values ​​shown above, and can be set to any value according to the control state of the motor, the installation environment, etc.

[0121] Next, the noise determination unit 530 sets the signal strength at the data density (3%) as the first signal strength (-80dB) (step S130E), and in the first control, removes the signal strength of the frequency component with a signal strength smaller than the first signal strength (-80dB) from the first spectrum (step S130F).

[0122] The signal strength that generates the base noise has a high data density on the histogram. Therefore, conversely, the signal strength band with lower data density can be considered as a signal strength band with a higher signal-to-noise ratio (S / N) relative to the base noise. Thus, as described above, the noise determination unit 530 subtracts the signal strength of the frequency component with a signal strength lower than the first signal strength (-80dB) from the first spectrum. Therefore, in the detection of the presence or absence of an abnormality in the motor 1 described later, only the peak portion generated at or above the first signal strength (-80dB) is used, thus preventing false detections due to the influence of base noise.

[0123] Furthermore, in the above description, the noise determination unit 530 performs two processes in the first control: removal of the pulsation component of the base noise using the second regression equation and removal of the pulsation component of the base noise using the data density. However, it is not limited to this; for the first spectrum, only one of the processes, removal of the pulsation component of the base noise using the second regression equation or removal of the pulsation component of the base noise using the data density, may be performed.

[0124] Furthermore, when the floor noise is assumed to be low, the above two processes are sometimes not performed.

[0125] Next, for Figure 5 The load fluctuation noise estimation process shown (step S131) ​​uses... Figure 7 , Figure 8 And Figure 12 will be used for explanation.

[0126] First, the noise determination unit 530 acquires the first spectrum derived by the FFT analysis unit 503, and acquires the load rate of the motor 1 calculated by the load rate calculation unit 501 when the first spectrum is derived (steps S131A and S131B).

[0127] Next, the noise determination unit 530 confirms whether the set flag described later has been set (step S131C). If the flag has not been set (step S131C: Yes), the noise determination unit 530 acquires the sideband wave component extracted by the sideband wave extraction unit 520 (step S131D).

[0128] Next, the noise determination unit 530 stores the frequencies of multiple sideband wave components generated corresponding to the load fluctuations of the motor 1 together with the load rate at the time of acquisition in a table in the storage device 52 (step S131E).

[0129] By repeating steps S131A to S131E, the frequencies of the multiple sideband wave components extracted from the first spectrum are stored in correspondence with the load rate of motor 1 at the time of extraction.

[0130] in addition, Figure 7 The flowchart does not illustrate the control process of repeating the above steps S131A to S131E.

[0131] Next, after the noise determination unit 530 stores the frequencies of the sideband wave components under different load rates into tables, it extracts the sideband wave components that disappear, are newly generated, or have changed frequencies according to the changes in load rate based on the stored tables, and records them into the storage device 52 (step S131F).

[0132] After extracting the number of sideband wave components corresponding to the set load rate (step S131G: Yes), the noise determination unit 530 uses methods such as least squares to deduce the first regression equation with the load rate as the explanatory variable based on these sideband wave components (step S131H).

[0133] Inverter noise such as overmodulation noise caused by load fluctuations of motor 1 Figure 13 As shown, it has the characteristic that the frequency and load rate increase in a linear manner and change with a large negative slope above the first set value.

[0134] In subsequent descriptions, this inverter noise will sometimes be referred to as load fluctuation noise as the first sideband wave component.

[0135] The noise determination unit 530 compares the actual frequency with the frequency of the load fluctuation noise determined by the univariate regression equation obtained by estimation, and evaluates the estimation accuracy of the univariate regression equation by relative error.

[0136] When the predicted univariate regression equation meets a certain accuracy, the noise determination unit 530 determines the regression coefficients in the univariate regression equation.

[0137] When the relative error is not met, i.e., when the frequency of the sideband wave component actually detected at each load rate deviates significantly from the residual of the frequency determined by the univariate regression equation (step S131I: No), it can be considered that the decrease in accuracy is caused by insufficient number of sideband wave components used to infer the univariate regression equation. At this time, the noise determination unit 530 determines whether the number of sideband wave components used to infer the univariate regression equation at each load rate is below a set number. When the number of sideband wave components at each load rate is below a set number (step S131L: No), the process returns to step S131A to obtain the frequency of the sideband wave component corresponding to each load rate.

[0138] Furthermore, if the residuals of the frequencies actually detected at each load rate deviate significantly from the frequencies determined by the univariate regression equation, and if the number of sideband wave components used to infer the univariate regression equation at each load rate exceeds the set number (step S131L: Yes), it can be assumed that the frequencies of the detected sideband wave components contain a large error. In this case, there is also a method to set an upper limit on the frequencies of the sideband wave components used for inference and then infer the regression coefficients again. However, if it is completely difficult to infer, the data corresponding to the load rate in the load rate data recorded in the table in S131E is deleted (step S131M), and the regression coefficients are inferred again.

[0139] Thus, before deriving regression coefficients that minimize the residuals of the frequencies actually detected at each load rate that deviate from the frequencies determined by the univariate regression equation, the noise determination unit 530 performs at least one of the following: detection of the sideband wave components at each load rate, or deletion of the sideband wave component data stored in the storage device 52. This allows for the high-precision estimation of a first value, which is the slope in the univariate regression equation representing the frequency variation of the load fluctuation noise.

[0140] The noise determination unit 530 calculates and records the regression coefficients for each power supply frequency fs (step S131J). When the number of calculations reaches a certain set number (step S131K: Yes), a flag is set, and the calculation of the regression coefficients in the univariate regression equation for load fluctuation noise prediction at that power supply frequency fs is completed (step S131N). The flag indicates that the derivation of the univariate regression equation for load fluctuation noise prediction for each power supply frequency fs of the motor 1 has been completed.

[0141] In step S131C, if there is a flag setting at the current power supply frequency fs of motor 1 (step S131C: No), the process is in a state where the superposition determination of load fluctuation noise to characteristic frequency (fs±fr) can be performed, and therefore the process of load fluctuation noise determination execution (step S140) is entered.

[0142] Next, use Figure 8The procedure for determining load fluctuation noise (step S140) is explained.

[0143] The noise determination unit 530 acquires the first spectrum derived in S104 and acquires the load rate of the motor 1 when the first spectrum is derived (steps S140A and S140B).

[0144] Next, the noise determination unit 530 determines whether a flag has been set at the power supply frequency fs of the motor 1 at the current moment based on the first spectrum, that is, whether the regression coefficient used for load fluctuation noise estimation at the power supply frequency fs has been derived (step S140C).

[0145] Furthermore, the steps S140A to S140C described above correspond to the aforementioned Figure 7 Steps S131A to S131C are performed, but due to inverter control and other reasons, the power supply frequency fs may deviate. Therefore, in order to improve the accuracy of load fluctuation noise estimation, the presence or absence of a flag is determined based on the power supply frequency fs in the most recently derived first spectrum.

[0146] If the flag for the current power frequency fs has been set (step S140C: Yes), the noise determination unit 530 refers to the regression coefficients stored in S131J (step S140D).

[0147] Then, the noise determination unit 530, based on a univariate regression equation using the inferred regression coefficients, infers at what frequency the load fluctuation noise occurs under the current power supply frequency fs and load rate (step S140E).

[0148] For example, according to Figure 13 At a load rate of 90%, the frequency of load fluctuation noise (inverter noise) is around 30Hz, which can be inferred to be superimposed with the characteristic frequency component (fs±fr) of about 30Hz.

[0149] Additionally, in step S140C, if no flag is set at the current power frequency fs (step S140C: No), the load fluctuation noise determination execution process ends and returns. Figure 7 The step S131A shown is illustrated.

[0150] The above describes the execution steps for determining load fluctuation noise.

[0151] Next, use Figure 14 right Figure 5 The voltage noise estimation process (step S132) shown is explained.

[0152] Figure 14This is a diagram illustrating the voltage noise estimation process of Embodiment 1.

[0153] If the first sideband wave component, which is load fluctuation noise, cannot be inferred at the first spectral peak derived by the current FFT, the FFT analysis unit 503 also performs a voltage FFT.

[0154] The noise determination unit 530 compares the analysis results of the current FFT with the voltage FFT. If the sideband wave component generated in the voltage FFT is also generated in the current FFT, the sideband wave component is detected as the first sideband wave component as load fluctuation noise.

[0155] Next, the noise determination unit 530 calculates six characteristic quantities: the frequency, signal strength, load rate, and power supply frequency fs of the load fluctuation noise generated in the current FFT, and the frequency and signal strength of the load fluctuation noise generated in the voltage FFT.

[0156] Noise determination unit 530 passes through Figure 5 The noise determination process ends after the following steps: the base noise estimation process (step S130), the load fluctuation noise estimation process (step S131), and the voltage noise estimation process (step S132).

[0157] As described above, the noise determination unit 530 performs a process to determine a signal strength threshold to remove the base noise. This signal strength threshold can distinguish between the base noise and the frequency component used for anomaly detection. Then, based on the base noise removal process described above, the noise determination unit 530 detects a sideband wave component whose frequency changes with the fluctuation of the motor load rate. Then, if there is the aforementioned correlation between the change in load rate and the change in the frequency of the sideband wave component, the noise determination unit 530 performs a process to determine that the sideband wave component as load fluctuation noise, which is the first sideband wave component.

[0158] However, the noise determination process is not limited to making the processes in steps S130 and S132 mandatory controls. For example, the process of removing the base noise in step S130 may be omitted, and the load fluctuation noise estimation process in step S131 may be performed instead.

[0159] Alternatively, control can be adopted that does not perform the voltage noise estimation process in step S132.

[0160] return Figure 4 The normal waveform estimation process (step S108) performed by the normal waveform estimation unit 550 after the above noise determination process (step S107) will be explained.

[0161] In the normal waveform estimation process (step S108), the normal waveform estimation unit 550 performs first control by subtracting the signal strength of the load fluctuation noise estimated by the noise estimation process (step S107) from the first spectrum.

[0162] Since the load fluctuation noise is inferred using a regression equation, even if the signal strength of the characteristic frequency component (fs±fr) and the signal strength of the load fluctuation noise are superimposed in the same frequency band, the signal strength of the load fluctuation noise can be removed from the first spectrum with high precision.

[0163] The following describes the normal waveform estimation process (step S108) performed by the normal waveform estimation unit 550 using AI (Artificial Intelligence).

[0164] Figure 15 This is a schematic structural diagram of the normal waveform estimation unit 550 in Embodiment 1 when it has AI functionality.

[0165] Figure 16 A is a graph showing the characteristic quantities of the input data at a load rate of 29.9% obtained by the normal waveform estimation unit 550 of Embodiment 1. Figure 16 B is a graph showing the characteristic quantities of the input data at a load rate of 42.2% obtained by the normal waveform estimation unit 550 of Embodiment 1. Figure 16 C is a graph showing the characteristic quantities of the input data at a load rate of 55.2% obtained by the normal waveform estimation unit 550 of Embodiment 1. In this graph, the load fluctuation noise is represented as N.

[0166] Figure 17 This is a diagram illustrating the AI ​​control performed by the normal waveform prediction unit 550 in Embodiment 1.

[0167] Figure 18 This is a diagram showing the neural network in the normal waveform estimation unit 550 of Embodiment 1.

[0168] Figure 19 This is a diagram illustrating Implementation 1, which stores a database containing correction values ​​based on load fluctuations.

[0169] Figure 20 This is a flowchart illustrating the control of the abnormal state comparison unit of the arithmetic processing unit in Embodiment 1.

[0170] As mentioned above, inverter noise, which is load fluctuation noise, is strongly correlated with the frequency of occurrence of load fluctuations. Furthermore, signal strength is also correlated to some extent with the degree of noise occurrence; however, this value fluctuates due to factors such as the surrounding environment and the inverter's PWM control. Therefore, an advanced regression inference method utilizing AI (Artificial Intelligence) is used to infer the signal strength of inverter noise alone.

[0171] In particular, based on the load rate variation of motor 1, even if the inverter noise, which is load fluctuation noise, is superimposed on the characteristic frequency component (fs±fr) used in the anomaly determination of motor 1, or is close to the characteristic frequency component (fs±fr), the inverter noise component can be removed from the first spectrum with higher accuracy.

[0172] like Figure 15 As shown, the normal waveform prediction unit 550 includes a data acquisition unit 550A, a model generation unit 550B, and a learned model storage unit 550C.

[0173] Data Acquisition Unit 550A Acquisition Figure 16 The following data are used as learning data: the frequency of load fluctuation noise, signal strength, load rate, power supply frequency fs in the first spectrum (current FFT), the frequency of load fluctuation noise generated in the second spectrum (voltage FFT) and the signal strength of the characteristic frequency component (fs±fr) as the sideband wave component in the first spectrum after removing the load fluctuation noise.

[0174] Model generation unit 550B, such as Figure 17 As shown, based on the learning data created from the combination of input 1 and input 2 (correct answer) output from the data acquisition unit 550A, the signal strength (output 1) of the characteristic frequency as a sideband wave component in the second spectrum after removing load fluctuation noise is learned. That is, a learned model that predicts the optimal output based on input 1 and input 2 (correct answer) is generated. Here, the learning data is the data that correlates input 1 and input 2 (correct answer).

[0175] The model generation unit 550B, for example, learns the output C using a neural network model through so-called supervised learning. Here, supervised learning refers to a method of inferring the result based on the input by providing a set of data containing input and result (label) to a learning device and learning the features present in the learning data.

[0176] A neural network consists of an input layer containing multiple neurons, an intermediate layer (hidden layer) containing multiple neurons, and an output layer containing multiple neurons. The intermediate layer can be one layer or two or more layers.

[0177] For example, if it is Figure 18 The three-layer neural network shown in the diagram, when multiple inputs are input to the input layer (X1-X3), multiply the value by weight W1 (w11-w16) and input it to the intermediate layer (Y1-Y2). The result is then multiplied by weight W2 (w21-w26) and output from the output layer (Z1-Z3). The output result varies according to the values ​​of weights W1 and W2.

[0178] In this disclosure, the neural network learns the output through so-called supervised learning, based on learning data created from a combination of input 1 and input 2 (correct answer) acquired by the data acquisition unit 550A.

[0179] That is, the neural network learns by adjusting the weights W1 and W2, so that the result of inputting input 1 into the input layer and outputting from the output layer is close to input 2 (the correct answer).

[0180] The model generation unit 550B generates and outputs a learned model by performing the above-described learning process. The learned model storage unit 550C stores the learned model output from the model generation unit 550B.

[0181] Next, the normal waveform estimation unit 550 estimates the signal strength of the characteristic frequency component (fs±fr) of the first spectrum after removing the load fluctuation noise component, based on the input characteristics of the load fluctuation noise frequency, signal strength, load rate, power supply frequency fs in the first spectrum (current FFT), the load fluctuation noise frequency generated in the second spectrum (voltage FFT), and the signal strength (input 1).

[0182] Furthermore, in this embodiment, the application of supervised learning to the learning algorithm used by the model generation unit has been described, but this disclosure is not limited to this. In addition to supervised learning, reinforcement learning, unsupervised learning, or semi-supervised learning may also be applied to the learning algorithm.

[0183] Here, the signal strength of the sideband wave component in the first spectrum after removing the load fluctuation noise component, which is the above input 2 (correct answer), can also be derived from the first regression equation that includes the load rate of motor 1 as an explanatory variable.

[0184] As described above, the input features are set as follows: the signal strength and frequency of the sideband wave component generated in the voltage FFT and the sideband wave component generated at the same frequency in the current FFT, the signal strength and frequency of the voltage FFT, the power supply frequency fs of the inverter, and the load rate.

[0185] By subtracting the spectral peak value of the load fluctuation noise derived from the voltage FFT, which is calculated by AI through inference, from the current FFT analysis results, a current FFT waveform with load fluctuation noise removed is generated by AI.

[0186] Therefore, the normal waveform estimation unit 550 can also perform the first control to remove the load fluctuation noise component from the first spectrum by using AI.

[0187] In addition, as examples of AI for time series regression problems, LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), XGBoost, LightGBM (Light Gradient Boosting Machine) and other AI methods can be used, or even any method such as statistical methods can be used for inference, not limited to AI.

[0188] After the above steps, the normal waveform estimation unit 550 determines the first spectrum after removing load fluctuation noise as the spectrum for subsequent anomaly diagnosis (step S109).

[0189] Furthermore, if the noise determination unit 530 determines in step S107 that there is no load fluctuation noise in the first spectrum, the normal waveform estimation process in step S108 will not be performed.

[0190] Next, the averaging unit 511 extracts the signal strength of the characteristic frequency component (fs±fr) based on the frequency components of the first spectrum after removing the load fluctuation noise component (step S110), such as Figure 19 As shown, for each power supply frequency fs, the characteristic frequency component (fs±fr) under each load rate is averaged multiple times (step S111).

[0191] The averaging operation unit 511 accumulates the averaged value multiple times over a certain period of time and performs initial learning as a determined value for the normal state (step S112).

[0192] The normal state storage unit 592 stores the first spectrum of the signal strength after subtracting the load fluctuation noise component as the normal state measurement value for the purpose of determining the abnormality of the motor 1 described later.

[0193] After the initial learning is completed, the matrix selection unit 540 calculates the correction value of the characteristic frequency component (fs±fr) under each load rate, i.e. the matrix correction value, and the judgment threshold of abnormal vibration based on the derived average value (step S113).

[0194] The so-called matrix correction value refers to the correction value of the signal strength of the detected characteristic frequency component (fs±fr) to the signal strength under the set first power frequency fs and first load rate, based on the power frequency fs and load rate at which the signal strength is detected.

[0195] The matrix selection unit 540 databases the correction value and stores the judgment threshold for anomaly determination in the correction value data storage unit 591 based on the standard deviation of the averaged feature frequencies.

[0196] The database storing the correction values ​​could be, for example, in... Figure 19 The database with correction values ​​is entered in the AVE column shown. As long as a correction value corresponding to each load rate is set for each power frequency fs, it is sufficient.

[0197] Next, use Figure 20 The diagnostic initiation process is explained.

[0198] The process from confirming that motor 1 starts operating (step S101) to step S111, which calculates the signal strength of the characteristic frequency component (fs±fr) under each power supply frequency fs and load rate, is set to be the same as the aforementioned initial learning process.

[0199] The abnormal state comparison unit 560 performs a correction matrix position selection process (step S212) to read the correction value under the current load rate from a pre-set database containing correction matrix values ​​learned through the initial learning process, and reads the correction value under the current load rate and the determination threshold. That is, the abnormal state comparison unit 560 performs correction control to correct the signal strength of the characteristic frequency component (fs±fr) under the current load rate to the pre-set first power supply frequency fs and the signal strength under the first load rate.

[0200] Then, the abnormal state comparison unit 560 performs an initial value-current value comparison (step S213) that compares the initial learning value with the corrected measurement value.

[0201] The abnormal state comparison unit 560 compares the value stored in the normal state storage unit 592 with the current value, and diagnoses whether the motor is abnormal based on the judgment threshold (step S129).

[0202] In addition, the judgment threshold used in the anomaly determination can be the judgment threshold derived in step S106A, but if it is speculated that there is load fluctuation noise superimposed in the characteristic frequency component, the judgment threshold derived in step S113 can be used.

[0203] The motor diagnostic device of this embodiment, configured as described above, includes: The detection unit detects a current signal of the current flowing in the motor and a voltage signal of the voltage applied to the motor; and The control unit detects whether there is any abnormality in the motor based on at least the detected current signal. The control unit detects the sideband wave component of the power supply frequency of the motor based on a first spectrum obtained by frequency analysis of the current signal, performs a first control to remove the first sideband wave component from the first spectrum, and detects whether the motor has any abnormality based on the first spectrum after removing the first sideband wave component. The first sideband wave component is the sideband wave component whose frequency changes by more than a first set amount according to the load fluctuation of the motor.

[0204] Through repeated research, the inventors discovered that overmodulation noise, which is easily generated when an electric motor operates under high load and is overmodulated, is produced as a sideband wave, and its frequency changes with load fluctuations (degree of overmodulation). Furthermore, the magnitude of this frequency change corresponds to the magnitude of the load rate change, exceeding a certain threshold. Therefore, it is clear that the correlation between load rate and frequency can be used to determine whether something is noise.

[0205] In this embodiment, the control unit performs a first control to remove a first sideband wave component from the first spectrum. This first sideband wave component is a sideband wave component whose frequency changes by a first set amount or more according to the load fluctuations of the motor, obtained through frequency analysis. Specifically, the control unit performs the first control, detecting sideband wave components whose frequency changes by a set first amount or more according to the load fluctuations as noise components, and removing these noise components from the first spectrum. Therefore, by accurately detecting and removing noise components generated by motor load fluctuations that were previously difficult to determine, the diagnostic accuracy for motor abnormalities can be significantly improved.

[0206] Furthermore, although the characteristic frequency used to determine motor malfunctions changes with the load rate, the amount of change is extremely small. Therefore, by setting the first setpoint to the frequency change, noise components and characteristic frequencies can be distinguished and determined with high precision.

[0207] Furthermore, based on the highly accurate motor diagnostic results described above, monitoring personnel can grasp the status of the motor through the display unit and other notification units, enabling them to perform efficient motor maintenance.

[0208] Furthermore, in the motor diagnostic device of this embodiment configured as described above, The control unit detects the first sideband wave component whose frequency changes by more than a first set amount according to the load fluctuation of the motor, among the multiple sideband wave components in the first spectrum, based on a first regression equation with the load rate of the motor as the explanatory variable.

[0209] The frequency of the noise component generated by the motor during overmodulation operation will change according to the load fluctuation of the motor. Therefore, in the first spectrum, the frequency band of the characteristic frequency component used for motor anomaly diagnosis may overlap with the frequency band of the noise component. In this case, false detection may occur in the diagnosis of whether the motor has an anomaly.

[0210] The motor diagnostic device of this embodiment detects the first sideband wave component, which is a noise component, based on a first regression equation that uses the motor's load rate as an explanatory variable. The frequency characteristics of the first sideband wave component, which is a noise component, are thus determined using the regression equation.

[0211] Therefore, by pre-determining the frequency characteristics of the first sideband wave component in the frequency band where the first sideband wave component is not superimposed on the characteristic frequency component using the first regression equation, it is possible to infer the signal strength of the first sideband wave component in the frequency band where the first sideband wave component is superimposed on the characteristic frequency component. Thus, based on the motor's load condition, even when the first sideband wave component is superimposed on the characteristic frequency component, noise components can be precisely removed from the first spectrum, enabling high-precision motor diagnosis independent of the motor's load condition.

[0212] Furthermore, in the motor diagnostic device of this embodiment configured as described above, When the relationship between the load rate of the motor and the frequency of the sideband wave component is represented by a univariate regression equation that is a linear function of the first regression equation, and the regression coefficient representing the slope in the univariate regression equation is a first value or higher than the first set value having a set polarity, the control unit detects the sideband wave component in the first spectrum that satisfies the relationship between the load rate and the frequency in the univariate regression equation, and identifies it as the first sideband wave component.

[0213] Through repeated research, the inventors discovered that the frequency variation of overmodulation noise, such as inverter noise, which is easily generated when a motor operates under high load, sometimes exhibits the following characteristics: the frequency increases linearly and proportionally with the increase in load rate. Furthermore, it is known that the characteristic frequency used for detecting abnormalities in motors decreases slightly and linearly with the increase in load rate.

[0214] In this embodiment, the control unit is pre-set such that the frequency characteristics of the first frequency component, which is a noise component, satisfy a regression coefficient of at least a first value having a set polarity, and satisfy a univariate regression equation expressed by a linear function. Then, among the detected sideband wave components, the sideband wave component that satisfies this univariate regression equation is detected as the first sideband wave component, which is a noise component, and is removed from the first spectrum.

[0215] Therefore, by predicting the frequency characteristics of the first frequency component, which is a noise component, in advance with higher accuracy, it is possible to distinguish the noise component and the characteristic frequency component with higher accuracy based on the determination of frequency change and the difference in the direction of frequency change, thereby making anomaly judgment.

[0216] Furthermore, in the motor diagnostic device of this embodiment configured as described above, The control unit performs at least one of the following processes: detecting the sideband wave component at each load rate of the motor and deleting the frequency of the sideband wave component at each load rate recorded in the storage unit, until the regression coefficient in the univariate regression equation that minimizes the residual of the frequency of the sideband wave component detected at each load rate from the frequency determined by the univariate regression equation is derived.

[0217] Depending on the motor's installation environment and the control status of the inverter driving the motor, noise will often be superimposed on the first frequency spectrum.

[0218] The control unit in this embodiment repeatedly performs operations such as detecting the sideband wave components used to predict the univariate regression equation, or deleting data of the sideband wave components, as described above, until the regression coefficients of the detected sideband wave components from the univariate regression equation with the smallest residual deviation from the frequency determined by the univariate regression equation are derived. Therefore, when there are insufficient sideband wave components used to predict the univariate regression equation, data of those sideband wave components can be acquired until a sufficient number is obtained to complete the prediction. Furthermore, for example, frequency components of external noise that occur only once are removed in the prediction of the univariate regression equation, thus enabling the setting of a highly accurate univariate regression equation.

[0219] Furthermore, in the motor diagnostic device of this embodiment configured as described above, The control unit includes: A data acquisition unit acquires learning data, which includes at least one of the frequency and signal strength of the sideband wave component in the first spectrum, the power supply frequency, and the load rate of the motor, as well as the signal strength of the sideband wave component in the first spectrum after removing the first sideband wave component; and A model generation unit generates a learned model, which is used in the first control to infer the signal strength of the sideband wave component in the first spectrum after removing the first sideband wave component, based on at least one of the frequency and signal strength of the sideband wave component in the first spectrum, the power supply frequency, and the load rate, using the learned data.

[0220] Therefore, in the first control that removes the first frequency component from the first spectrum, by using AI-based control, the influence of noise environment, inverter control, etc., can be eliminated, and the noise component can be removed from the first spectrum with high precision.

[0221] Furthermore, in the motor diagnostic device of this embodiment configured as described above, The control unit derives a second regression equation representing the relationship between frequency components in the first spectrum and signal strength by performing a polynomial approximation regression analysis. In the first control, the control unit removes the signal strength of the first sideband wave component and the signal strength of each frequency component determined by the second regression equation from the first spectrum.

[0222] Therefore, by performing regression analysis using polynomial approximation, noise components such as the base noise, whose frequency characteristics do not change linearly, can be inferred. This allows for the high-precision and dynamic inference of noise components that do not exhibit unidirectional increasing or decreasing characteristics, thus enabling more accurate motor diagnostics.

[0223] This disclosure describes exemplary embodiments, but the various features, methods, and functions described in the embodiments are not limited to specific embodiments and can be applied to embodiments individually or in various combinations.

[0224] Therefore, numerous variations not illustrated are contemplated to fall within the scope of the technology disclosed in this specification. For example, this includes variations, additions, or omissions of at least one constituent element.

[0225] Label Explanation

[0226] 1 Electric motor, 20 Measurement circuit (detection unit), 50 Calculation and processing unit (control unit), 52 Storage device (storage unit), 550A Data acquisition unit, 550B Model generation unit, 60 Electric motor diagnostic device, 70 Notification unit (display unit), 100 Electric motor diagnostic system.

Claims

1. A motor diagnostic device, characterized in that, include: The detection unit detects the current signal of the current flowing in the motor and the voltage signal of the voltage applied to the motor. as well as The control unit detects whether there is any abnormality in the motor based on at least the detected current signal. The control unit detects the sideband wave component of the power supply frequency of the motor based on a first spectrum obtained by frequency analysis of the current signal, performs a first control to remove the first sideband wave component from the first spectrum, and detects whether the motor has any abnormality based on the first spectrum after removing the first sideband wave component. The first sideband wave component is the sideband wave component whose frequency changes by more than a first set amount according to the load fluctuation of the motor.

2. The motor diagnostic device as described in claim 1, characterized in that, The control unit detects the first sideband wave component whose frequency changes by more than a first set amount according to the load fluctuation of the motor, among the multiple sideband wave components in the first spectrum, based on a first regression equation with the load rate of the motor as the explanatory variable.

3. The motor diagnostic device as described in claim 2, characterized in that, When the relationship between the load rate of the motor and the frequency of the sideband wave component is represented by a univariate regression equation that is a linear function of the first regression equation, and the regression coefficient representing the slope in the univariate regression equation is a first value or higher than the first set value having a set polarity, the control unit detects the sideband wave component in the first spectrum that satisfies the relationship between the load rate and the frequency in the univariate regression equation, and identifies it as the first sideband wave component.

4. The motor diagnostic device as described in claim 3, characterized in that, The control unit performs at least one of the following processes: detecting the frequency of the sideband wave component at each load rate of the motor and deleting the frequency of the sideband wave component at each load rate recorded in the storage unit, until the regression coefficient in the univariate regression equation that minimizes the residual of the frequency of the sideband wave component detected at each load rate from the frequency determined by the univariate regression equation is derived.

5. The motor diagnostic device as described in any one of claims 1 to 4, characterized in that, The control unit includes: A data acquisition unit acquires learning data, which includes at least one of the frequency and signal strength of the sideband wave component in the first spectrum, the power supply frequency, and the load rate of the motor, as well as the signal strength of the sideband wave component in the first spectrum after removing the first sideband wave component; and A model generation unit generates a learned model, which is used in the first control to infer the signal strength of the sideband wave component in the first spectrum after removing the first sideband wave component, based on at least one of the frequency and signal strength of the sideband wave component in the first spectrum, the power supply frequency, and the load rate, using the learned data.

6. The motor diagnostic device as described in claim 5, characterized in that, The signal strength of the sideband wave component in the first spectrum after removing the first sideband wave component, which is the learning data, is derived based on a first regression equation that includes the load rate of the motor as an explanatory variable.

7. The motor diagnostic device according to any one of claims 1 to 6, characterized in that, The control unit derives a second regression equation representing the relationship between frequency components in the first spectrum and signal strength by performing a polynomial approximation regression analysis. In the first control, the control unit removes the signal strength of the first sideband wave component and the signal strength of each frequency component determined by the second regression equation from the first spectrum.

8. The motor diagnostic device according to any one of claims 1 to 7, characterized in that, The control unit generates a histogram representing the frequency of occurrence of frequency components of each signal strength based on the first spectrum. In the first control, the signal strength of the frequency component with a smaller signal strength than the first signal strength set based on the signal strength with the highest frequency of occurrence in the histogram is removed from the first spectrum.

9. The motor diagnostic device as described in any one of claims 1 to 8, characterized in that, In the sideband wave components of the second spectrum obtained by frequency analysis of the voltage signal The control unit takes the sideband wave component generated at the same frequency in the first spectrum as the first sideband wave component, and removes the first sideband wave component from the first spectrum in the first control.

10. The motor diagnostic device according to any one of claims 1 to 9, characterized in that, Based on the first spectrum after removing the first sideband wave component, the control unit derives the average signal strength of the sideband wave component at various load rates of the motor for each power supply frequency. The control unit includes a database containing correction values. These correction values ​​are based on the derived average value and, according to the power supply frequency at which the signal strength of the detected sideband wave component is detected and the load rate of the motor, correct the detected signal strength to a signal strength at a set first power supply frequency and a first load rate. The control unit performs correction control based on the database, correcting the signal strength of the detected sideband wave component to the signal strength at the first power supply frequency and the first load rate.

11. The motor diagnostic device as described in claim 10, characterized in that, The control unit detects whether there is any abnormality in the motor based on the signal strength of the sideband wave component after correction by the correction control.

12. A motor diagnostic system, characterized in that, include: The electric motor diagnostic device as described in any one of claims 1 to 11; as well as The notification department, which is connected to the motor diagnostic device via a network, notifies the administrator of the diagnostic results of the motor diagnostic device.

Citation Information

Patent Citations

  • Diagnosis support device, rotating machine system, and diagnosis support method

    WO2019198563A1