Predictive maintenance device and predictive maintenance method

Abstract

To provide a predictive maintenance device and predictive maintenance method capable of accurately detecting abnormal signs in a target device.SOLUTION: A predictive maintenance device 1 comprises: a time waveform data acquisition unit 35 that acquires time waveform data of target devices 12, 14 measured by a measurement circuit 30; an FFT unit 52 that obtains a spectrum of the time waveform data; a logarithmic power spectrum calculation unit 53 that calculates a logarithmic power spectrum from the spectrum; a cepstrum calculation unit 54a that obtains a cepstrum from the logarithmic power spectrum; a disturbance removal unit 54b for acquiring a disturbance-removed power spectrum; a feature quantity extraction unit 56 that extracts two-dimensional feature quantities from waveforms of the disturbance-removed power spectrum and the cepstrum; a Mahalanobis distance calculation unit 58 that calculates the Mahalanobis distance of the feature quantities; an abnormality determination unit 61 that determines that the target device is abnormal when the Mahalanobis distance exceeds a threshold; and a display unit 40 that displays an abnormality determination result determined by the abnormality determination unit 61.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a predictive maintenance device and a predictive maintenance method, and more particularly to a predictive maintenance device and a predictive maintenance method that can perform predictive maintenance on various components that make up an article inspection device. [Background technology]

[0002] Conventionally, maintenance for product inspection equipment has consisted of either corrective maintenance, in which various components comprising the equipment are replaced after failure, or preventive maintenance, in which various components are replaced periodically before failure occurs. Recently, to improve line productivity and reduce running costs, there has been a demand for predictive maintenance, in which abnormal parts are detected in advance and replaced before the equipment stops functioning due to a failure.

[0003] Known technologies relating to predictive maintenance include those disclosed in, for example, Patent Documents 1 to 3. The technology in Patent Document 1 extracts features from sensor data on a frame-by-frame basis and uses normal class data as a discriminator to determine whether the extracted features on a frame-by-frame basis are normal or abnormal.

[0004] The technology in Patent Document 2 detects waveform data such as vibrations and sounds generated when a rotating device rotates, measures temporal changes in the spectrum of the waveform data using short-time Fourier transform or wavelet transform, and determines whether an abnormality exists based on feature quantities obtained from the temporal changes in the spectrum of the waveform data. The feature quantities here are the frequency of the power peak of the spectrum or cepstrum and the time interval (period) at which the power at that frequency exceeds a preset threshold.

[0005] The technology of Patent Document 3 detects abnormalities using Mahalanobis distance based on data collected using various types of sensors, and performs deterioration diagnosis and analysis. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Application Laid-Open No. 2017-102765 [Patent Document 2] Japanese Patent Application Publication No. 10-274558 [Patent Document 3] Japanese Patent Application Laid-Open No. 2000-259222 Summary of the Invention [Problem to be solved by the invention]

[0007] However, the conventional techniques disclosed in Patent Documents 1 to 3 have the problem that it is difficult to accurately detect abnormalities when the measurement data is unstable due to changes in the usage environment caused by disturbances such as changes in temperature and humidity.

[0008] The present invention has been made to solve such conventional problems, and aims to provide a predictive maintenance device and a predictive maintenance method that can accurately detect signs of abnormality in a target device. [Means for solving the problem]

[0009] In order to solve the above problems, the predictive maintenance device according to the present invention includes a measurement circuit (30) that measures a current or a voltage of a target device (12, 14), a time waveform data acquisition unit (35) that acquires time waveform data of the current or the voltage measured by the measurement circuit, an FFT unit (52) that performs a Fourier transform on the time waveform data to obtain a spectrum of the time waveform data, a logarithmic power spectrum calculation unit (53) that calculates a logarithmic power spectrum from the spectrum, and a cepstrum calculation unit (54a) that obtains a cepstrum from the logarithmic power spectrum. The system includes a disturbance removal unit (54b) that obtains a disturbance-removed power spectrum by subtracting a DC component of the logarithmic power spectrum from the logarithmic power spectrum, a feature extraction unit (56) that extracts two-dimensional feature quantities of the disturbance-removed power spectrum and the waveform of the cepstrum, a Mahalanobis distance calculation unit (58) that calculates a Mahalanobis distance of the feature quantities, an abnormality determination unit (61) that determines that the target device is abnormal when the Mahalanobis distance exceeds a threshold, and a display unit (40) that displays the abnormality determination result determined by the abnormality determination unit.

[0010] With this configuration, the predictive maintenance device of the present invention can accurately detect signs of abnormality in the target device, making it possible to replace parts of the target device that have not yet developed an abnormality with new parts at the appropriate time before the target device breaks down, thereby significantly reducing the downtime of item inspection devices equipped with the target device.

[0011] Furthermore, the predictive maintenance device of the present invention can eliminate the effects of disturbances that are mainly caused by changes in the operating environment inside and outside the target device, so it is possible to acquire feature values ​​and perform predictive maintenance immediately after the target device starts operating.

[0012] Moreover, the predictive maintenance device according to the present invention may further include an interval division unit (55) that calculates a plurality of peak frequencies in the disturbance-removed power spectrum according to the operating state of the target device, divides the disturbance-removed power spectrum into a plurality of frequency intervals each including each of the peak frequencies, calculates peak times of a plurality of quefrencies in the cepstrum according to the operating state of the target device, and divides the cepstrum into a plurality of time intervals each including each of the peak times, and the feature extraction unit may be configured to extract the feature of the waveform of the disturbance-removed power spectrum in at least one of the plurality of frequency intervals, and extract the feature of the waveform of the cepstrum in at least one of the plurality of time intervals.

[0013] With this configuration, the predictive maintenance device according to the present invention can detect small changes in the disturbance-removed power spectrum and the cepstrum by dividing the data into sections to reduce the number of data to be processed when extracting waveform features of the disturbance-removed power spectrum and the cepstrum.

[0014] Furthermore, the predictive maintenance device according to the present invention may be configured such that the feature is the center of gravity of the disturbance-removed power spectrum in at least one of the plurality of frequency intervals, or the center of gravity of the cepstrum in at least one of the plurality of time intervals.

[0015] With this configuration, the predictive maintenance device according to the present invention can obtain the center of gravity of the disturbance-removed power spectrum or cepstrum for each divided section, thereby capturing the feature quantities of the entire waveforms of the divided disturbance-removed power spectrum and cepstrum.

[0016] Furthermore, the predictive maintenance device according to the present invention may further include a Mahalanobis ellipse estimation unit (59) that estimates an ellipse indicating the Mahalanobis distance equal to the threshold value in a two-dimensional space of the feature quantity from the feature quantity obtained based on normal time waveform data, and the display unit may be configured to further display the ellipse and the feature quantity obtained based on the time waveform data of the object to be determined as abnormal on a graph having a first axis indicating a first component of the feature quantity and a second axis indicating a second component of the feature quantity.

[0017] With this configuration, the predictive maintenance device according to the present invention can visually indicate how far a feature obtained based on time waveform data of a target for abnormality detection is from the threshold for abnormality detection.

[0018] Furthermore, the predictive maintenance device according to the present invention may be configured so that the cepstrum calculation unit performs an inverse Fourier transform on the logarithmic power spectrum to obtain the cepstrum, and the disturbance removal unit subtracts the DC component corresponding to a zeroth-order component of the cepstrum from the logarithmic power spectrum to obtain the disturbance-removed power spectrum.

[0019] The predictive maintenance device according to the present invention may further include a logarithmic power cepstrum calculation unit (54c) that calculates a logarithmic power cepstrum from the cepstrum.

[0020] The predictive maintenance method according to the present invention also includes a time waveform data acquisition step (S5, S25) of acquiring time waveform data of a current or voltage of a target device (12, 14) measured by a measurement circuit (30), an FFT step (S6, S26) of Fourier transforming the time waveform data to obtain a spectrum of the time waveform data, a logarithmic power spectrum calculation step (S7, S27) of calculating a logarithmic power spectrum from the spectrum, a cepstrum calculation step (S8, S28) of calculating a cepstrum from the logarithmic power spectrum, and a cepstrum calculation step (S9, S99) of calculating a cepstrum from the logarithmic power spectrum. The method includes a disturbance removal step (S9, S29) of subtracting the DC component of the logarithmic power spectrum to obtain a disturbance-removed power spectrum, a feature extraction step (S11, S31) of extracting two-dimensional feature quantities of the disturbance-removed power spectrum and the cepstrum waveform, a Mahalanobis distance calculation step (S32) of calculating the Mahalanobis distance of the feature quantities, an abnormality determination step (S35) of determining that the target device is abnormal when the Mahalanobis distance exceeds a threshold, and a display step (S36) of displaying the abnormality determination result determined in the abnormality determination step. [Effects of the Invention]

[0021] The present invention provides a predictive maintenance device and a predictive maintenance method that can accurately detect signs of abnormality in a target device. [Brief explanation of the drawings]

[0022] [Figure 1] 1 is a diagram illustrating a configuration of a predictive maintenance device according to an embodiment of the present invention. [Figure 2] FIG. 1(a) is a diagram showing the configuration of a motor and motor drive circuit, which are target devices of a predictive maintenance device according to an embodiment of the present invention, and the configuration of a measurement circuit using a shunt resistor, and FIG. 1(b) is a diagram showing the configuration of a measurement circuit using a clamp meter and a current transformer. [Figure 3] 10A is a table showing an example of various parameters relating to the rotation of the motor and the drive roller, and FIG. 10B is a table showing an example of various parameters relating to the bearings of the drive roller. [Figure 4] 1A is a graph showing an example of a logarithmic power spectrum calculated by a predictive maintenance system according to an embodiment of the present invention, and FIG. 1B is a graph showing an example of a logarithmic power cepstrum calculated by a predictive maintenance system according to an embodiment of the present invention. [Figure 5] 10 is a graph for explaining changes in time waveforms and spectrum waveforms due to the influence of disturbances; [Figure 6] 1 is a graph showing an example of a disturbance-removed power spectrum calculated by a predictive maintenance system according to an embodiment of the present invention. [Figure 7A] FIG. 10 is a diagram of a maintenance screen showing an example of a logarithmic power spectrum and a plurality of frequency intervals before disturbance removal. [Figure 7B] FIG. 10 is a diagram of a maintenance screen showing an example of a logarithmic power cepstrum and multiple time intervals. [Figure 8] FIG. 10 is a diagram showing the position of the FFT center of gravity in a specific frequency interval of the disturbance-removed power spectrum. [Figure 9] FIG. 1 is a diagram illustrating the concept of the dynamic center of gravity. [Figure 10] FIG. 10 is a diagram showing an example of a graph display in which feature quantities obtained from a target device are plotted. [Figure 11] 10(a) to 10(c) are diagrams showing examples of the correspondence between the Mahalanobis distance and the degree of abnormality obtained from the target device. [Figure 12] 10 is a graph showing an example of a change in the degree of abnormality over time. [Figure 13] 10 is a table showing an example of a correspondence relationship between an abnormality level, a symbol, and a character string. [Figure 14A] FIG. 2 is a diagram showing a normal operation screen displayed on a display unit included in the predictive maintenance device according to the embodiment of the present invention. [Figure 14B] FIG. 10 is a diagram showing an abnormality sign display screen displayed by a display unit included in the predictive maintenance device according to the embodiment of the present invention. [Figure 14C] FIG. 10 is a diagram illustrating a maintenance screen displayed on a display unit included in the predictive maintenance device according to the embodiment of the present invention. [Figure 14D]FIG. 10 is a diagram illustrating a condition monitoring screen displayed on a display unit included in the predictive maintenance device according to the embodiment of the present invention. [Figure 14E] FIG. 10 is a diagram illustrating a mode selection screen displayed on a display unit included in the predictive maintenance device according to the embodiment of the present invention. [Figure 14F] FIG. 10 is a diagram showing a data collection execution screen displayed on a display unit included in the predictive maintenance device according to the embodiment of the present invention. [Figure 14G] FIG. 10 is a diagram showing a data registration screen displayed on a display unit included in the predictive maintenance device according to the embodiment of the present invention. [Figure 15] 10 is a flowchart illustrating processing in a normal data collection mode of a predictive maintenance method using a predictive maintenance device according to an embodiment of the present invention. [Figure 16] 10 is a flowchart illustrating processing in a normal use abnormality determination mode of a predictive maintenance method using a predictive maintenance device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0023] Hereinafter, embodiments of a predictive maintenance device and a predictive maintenance method according to the present invention will be described with reference to the drawings.

[0024] As shown in FIG. 1, the predictive maintenance device 1 of this embodiment detects abnormal signs in a target device, and includes a measurement circuit 30, a time waveform data acquisition unit 35, and a control unit 50 having a display unit 40 and an operation unit 46.

[0025] The target device is, for example, a component that constitutes various kinds of article inspection devices or sorting machines, and more specifically, for example, a motor and drive roller mounted on a conveying unit 10 that various kinds of article inspection devices are equipped with.

[0026] The conveying section 10 is a belt conveyor that includes, for example, a loop-shaped conveying belt 11, a drive roller 12 and a driven roller 13 around which the loop-shaped conveying belt 11 is wound, a motor 14 that drives the drive roller 12, and a motor drive circuit 20, and sequentially conveys the objects W placed on the conveying surface 11a of the conveying belt 11.

[0027] 2(a), the motor 14 is driven by a motor drive circuit 20 that converts a DC voltage from a DC power supply 21 into a three-phase AC voltage. The DC power supply 21 outputs the DC voltage to an inverter circuit 24 via a pair of power supply lines 22 and 23. A shunt resistor 32 is provided on the negative side of the power supply line 23.

[0028] The inverter circuit 24 has a three-phase bridge configuration in which a series circuit of switching elements 24a and 24b, a series circuit of switching elements 24c and 24d, and a series circuit of switching elements 24e and 24f are connected in parallel between the power supply lines 22 and 23.

[0029] The inverter circuit 24 has a U-phase output terminal 24g, a V-phase output terminal 24h, and a W-phase output terminal 24i connected to coils of each phase wound around a stator provided in the motor 14.

[0030] In the following, an example will be described in which the motor 14 is a 14-pole, 12-slot motor. Here, the number of poles is the number of magnetic poles of the rotor of the motor 14, and the number of slots is the number of coils wound around the stator of the motor 14.

[0031] As the phase current flowing through each coil makes one revolution, the rotor is attracted toward the coil and rotates by 2 divided by the number of poles (2 / number of poles). In other words, the phase current makes seven revolutions while a 14-pole, 12-slot motor makes one revolution, so a peak appears in the phase current spectrum at a frequency seven times the frequency equivalent to the rotation speed [rpm] of the motor 14. Hereafter, the frequency obtained by multiplying the frequency equivalent to the rotation speed [rpm] of the motor 14 by the number of poles / 2 is called the "cogging frequency."

[0032] 3(a) is a table showing an example of various parameters related to the rotation of the motor 14 and the drive roller 12. When the main shaft rotation speed of the motor 14, which is a 14-pole, 12-slot motor, is 1250 rpm, the rotation frequency of the main shaft of the motor 14 is 20.84 Hz, the rotation period of the main shaft of the motor 14 is 0.048 s, and the cogging frequency of the main shaft of the motor 14 is 145.88 Hz.

[0033] Furthermore, when the gear ratio between the motor 14 and the drive roller 12 is 3.29, the rotation speed of the drive roller 12 (the rotation speed of the motor 14 itself) is 380.07 rpm, the rotation frequency of the drive roller 12 is 6.33 Hz, and the rotation period of the drive roller is 0.158 s. In this case, the cogging frequency of the motor 14 itself is 44.3 Hz. Furthermore, the conveying speed of the conveying unit 10 is 40 m / min.

[0034] Figure 3(b) is a table showing an example of various parameters related to the bearings of the drive roller 12. As shown in Figure 3(b), the frequency due to rotation of the cage is 2.431 Hz, the frequency due to rotation of the rolling elements is 12.898 Hz, the frequency due to revolution of the rolling elements is 19.452 Hz, the frequency due to damage to the inner ring is 31.224 Hz, the frequency due to damage to the outer ring is 19.452 Hz, and the frequency due to damage to the rolling elements is 25.795 Hz.

[0035] 2(a) measures a current flowing through one of the three phases of the motor 14 (hereinafter also referred to as a "single-phase current") and a voltage proportional to the sum of the magnitudes of the currents flowing through the phases of the motor 14 (hereinafter also referred to as a "total current"). The measurement circuit 30 includes, for example, shunt resistors 31a to 31c for measuring the single-phase current, a shunt resistor 32 for measuring the total current, analog amplifiers 33a to 33c, and a controller 34.

[0036] Note that, in place of the shunt resistors 31a to 31c, a current transformer (CT) or a clamp meter may be used to measure one-phase current. For example, as shown in Fig. 2(b), the magnetic core 36 of the CT 35 may be attached to the electric wire 15 of one of the three phases of the motor 14, and the voltage across the coil 37 wound around the magnetic core 36 may be measured to measure the one-phase current.

[0037] The single-phase current and total current of the measurement circuit 30 are either measured as current or converted into voltage and measured. Furthermore, the measurement circuit 30 may include an acceleration sensor that measures the vibration of the drive roller 12 or the motor 14, or a microphone that measures the sound of the drive roller 12 or the motor 14.

[0038] The time waveform data acquiring unit 35 has, for example, a low-pass filter for removing aliasing and an ADC (Analog-Digital Converter), and acquires time waveform data of current, voltage, vibration, sound, or the like measured by the measurement circuit 30 at a predetermined sampling frequency (for example, 4 kHz), and performs window function processing. Note that hereinafter, an example will be described in which the time waveform data acquiring unit 35 acquires time waveform data of one-phase current or time waveform data of total current measured by the measurement circuit 30.

[0039] The control unit 50 includes a mode selection unit 51, an FFT unit 52, a logarithmic power spectrum calculation unit 53, a cepstrum calculation unit 54a, a disturbance removal unit 54b, a logarithmic power cepstrum calculation unit 54c, an interval division unit 55, a feature extraction unit 56, a feature data storage unit 57, a Mahalanobis distance calculation unit 58, a Mahalanobis ellipse estimation unit 59, an abnormality degree calculation unit 60, an abnormality determination unit 61, and an external output unit 62.

[0040] 14E, the mode selection unit 51 switches the operation mode of the predictive maintenance device 1 between a normal data collection mode and a normal use anomaly determination mode. For example, the mode selection unit 51 selects the operation mode of the predictive maintenance device 1 in response to an operation input by a user to the operation unit 46.

[0041] Here, the normal data collection mode is an operation mode for collecting the features of normal time waveform data (hereinafter also referred to as "normal data") obtained from a device of the same type as the target device and in a normal state. It is desirable to collect the features of the normal data, for example, when shipping an article inspection device equipped with the target device, during idle operation before the start of production, or immediately after replacing the target device.

[0042] The abnormality determination mode is an operating mode in which abnormal signs of a target device are detected and an abnormality in the target device is determined by comparing the features of the time waveform data of the target device that is obtained from the target device and that may be abnormal with the features of normal data.

[0043] The FFT unit 52 performs a fast Fourier transform (FFT) on the time waveform data acquired by the time waveform data acquisition unit 35 to obtain the spectrum of the time waveform data.

[0044] The logarithmic power spectrum calculation unit 53 calculates a logarithmic power spectrum expressed in dB as shown in FIG. 4( a ) from the spectrum obtained by the FFT unit 52 .

[0045] The waveforms of the single-phase current and total current flowing through the motor 14 fluctuate due to the influence of disturbances. Possible disturbance influences are mainly those caused by changes in the usage environment inside and outside the device, such as contraction or expansion of the conveyor belt 11 or adhesion of dirt, changes in the hardness of the grease inside the drive system such as the drive roller 12 and motor 14, and changes in temperature and humidity in circuits such as the motor drive circuit 20.

[0046] As shown in Figure 5, if the current value of the original waveform a0 of one phase current or total current is multiplied by a constant as a whole by the influence of a disturbance, as in the waveforms a1 and a2, the power of those spectrum waveforms will shift up or down from the original spectrum b0 to the spectra b1 and b2. Such disturbance influences will also cause the distribution of feature quantities such as the FFT centroid, which will be described later, to shift from the original distribution.

[0047] The cepstrum calculation unit 54a performs an inverse Fast Fourier transform (IFFT) on the logarithmic power spectrum expressed in dB calculated by the logarithmic power spectrum calculation unit 53 to calculate a cepstrum.

[0048] The disturbance removal unit 54b subtracts the DC component corresponding to the zeroth-order component of the cepstrum calculated by the cepstrum calculation unit 54a from the logarithmic power spectrum calculated by the logarithmic power spectrum calculation unit 53, thereby obtaining a disturbance-removed power spectrum in which the influence of disturbances has been removed.

[0049] The logarithmic power cepstrum calculation unit 54c performs logarithmic transformation on the cepstrum calculated by the cepstrum calculation unit 54a using 20log to obtain a logarithmic power cepstrum for display on the screen.

[0050] Here, the time waveform data g obtained by the n-th measurement by the time waveform data acquisition unit 35 is n (t) is expressed as the following equation (1), where X(n) is a constant due to the influence of disturbances that is independent of time and frequency, and f(t) is the original time waveform data that is not influenced by disturbances.

[0051]

number

[0052] Time waveform datag n (t) is subjected to FFT by the FFT unit 52 and converted into a spectrum as shown in the following equation (2).n (f) and F(f) are respectively g n are the spectra of f(t) and f(t).

[0053]

number

[0054] Spectrum G n Since (f) is treated as logarithmic power, the logarithmic power spectrum calculation unit 53 calculates the logarithmic power spectrum L in dB notation as shown in the following equation (3): PS is converted to

[0055]

number

[0056] The logarithmic power spectrum L in dB shown in equation (3) PS is subjected to IFFT by the cepstrum calculation unit 54a and converted into a cepstrum as shown in the following equation (4a), and is calculated using equation (4b) to obtain a logarithmic power cepstrum as shown in FIG. 4(b).

[0057]

number

[0058] Here, q is the quefrency, and the left side of equation (4a) is gn (q), when q≠0, the right hand side is c f If (q), then equation (4a) can be expressed as equation (5) below.

[0059]

number

[0060] Cepstrum c gn (q) is the logarithmic power spectrum L expressed in dB as given by equation (3). PSIt can be said that the constant X(n) due to the disturbance effect is the cepstrum c gn (q) is summarized in the quefrency q=0 component. That is, c gn (0) is a DC component containing a constant X(n) component, and the logarithmic power spectrum L PS In 10^(c gn (0) / 20). On the other hand, c f (q≠0) is a component derived from the original time waveform data f(t) that does not include the component of the constant X(n).

[0061] The disturbance removal unit 54b extracts, from the logarithmic power spectrum calculated by the logarithmic power spectrum calculation unit 53, a DC component 10^(c gn (0) / 20) is subtracted to obtain the disturbance-free power spectrum shown in Fig. 6. The disturbance-free power spectrum shown in Fig. 6 is obtained by shifting the entire logarithmic power spectrum so that the center of power oscillation is 0, and the components due to the influence of disturbances have been removed.

[0062] Furthermore, the disturbance removal unit 54b can also obtain a disturbance-removed spectrum in which the influence of disturbances has been removed from the spectrum of the time waveform data output from the FFT unit 52. In this case, the logarithmic power spectrum calculation unit 53 calculates a logarithmic spectrum instead of a logarithmic power spectrum, the cepstrum calculation unit 54a calculates a cepstrum from the logarithmic spectrum, and the disturbance removal unit 54b subtracts a DC component corresponding to the zeroth-order component of the cepstrum from the logarithmic spectrum of the original time waveform data.

[0063] The interval dividing unit 55 calculates a plurality of peak frequencies corresponding to the operating state of the target device in the disturbance-removed power spectrum obtained by the disturbance removing unit 54b, and divides the disturbance-removed power spectrum into a plurality of frequency intervals each including each peak frequency. Alternatively, the interval dividing unit 55 may calculate a plurality of quefrency peak times corresponding to the operating state of the target device in the cepstrum obtained by the cepstrum calculating unit 54a, and divide the cepstrum into a plurality of quefrency time intervals each including each peak time. Here, the operating state of the target device is represented by parameters related to the rotation of the motor 14 and the drive roller 12, for example, as shown in FIG. 3.

[0064] That is, the main peaks of the disturbance-removed power spectrum and cepstrum appear at regular intervals determined by various parameters related to the rotation of the motor 14 and the drive roller 12 shown in Fig. 3. For this reason, the interval dividing unit 55 sets frequency intervals and quefrequency time intervals with interval widths that allow the main peaks to be captured one by one.

[0065] For example, the interval dividing unit 55 calculates integer multiples of the cogging frequency of the spindle of motor 14 as the multiple peak frequencies of the disturbance-rejection power spectrum based on the number of poles and the spindle rotation speed of motor 14. The interval dividing unit 55 divides the disturbance-rejection power spectrum into multiple frequency intervals, each having the cogging frequency of the spindle of motor 14 as the interval width, so that the center of each interval includes an integer multiple of the cogging frequency of the spindle of motor 14, as shown in Fig. 7A.

[0066] Alternatively, the interval dividing unit 55 calculates, based on the spindle rotation speed of the motor 14, integer multiples of the reciprocal of the spindle rotation speed of the motor 14, i.e., integer multiples of the time required for one rotation of the motor 14, as multiple peak times of the cepstrum. In the display example of the logarithmic power cepstrum in Fig. 7B, the interval dividing unit 55 divides the cepstrum into multiple quefrequency time intervals, each having an interval width equal to the reciprocal of the spindle rotation speed of the motor 14, so that the center of the interval includes an integer multiple of the reciprocal of the spindle rotation speed of the motor 14.

[0067] Similarly, the interval dividing unit 55 can divide the disturbance-removed power spectrum or cepstrum based on the rotation speed of the motor 14 itself, i.e., the rotation speed of the drive roller 12. Alternatively, the interval dividing unit 55 may divide the disturbance-removed power spectrum or cepstrum into any frequency interval or time interval of quefrency input via the operation unit 46.

[0068] 7A is a diagram showing an example of multiple frequency intervals (interval 1, interval 2, interval 3, interval 4, interval 5, interval 6, ...) of the disturbance-rejected power spectrum of one phase current of motor 14. The interval width of each frequency interval is equal to the cogging frequency (145.88 Hz) of the main shaft of motor 14, except for interval 1 which starts at a frequency of 0 Hz.

[0069] 7B is a diagram showing an example of multiple quefrency time intervals (interval 1, interval 2, interval 3, interval 4, interval 5, ...) of the cepstrum of one phase current of motor 14. The interval width of each time interval is equal to the reciprocal of the spindle rotation speed of motor 14 (0.048 s), except for interval 1 which starts at time 0 s.

[0070] The feature extractor 56 extracts two-dimensional feature values ​​of the waveform of the disturbance-removed power spectrum or disturbance-removed spectrum calculated by the disturbance remover 54b in at least one of the plurality of frequency intervals. Alternatively, the feature extractor 56 may extract two-dimensional feature values ​​of the waveform of the cepstrum calculated by the cepstrum calculator 54a in at least one of the plurality of quefrency time intervals.

[0071] For example, the feature extraction unit 56 is configured to extract features of the waveform of the disturbance-removed power spectrum, the disturbance-removed spectrum, or the cepstrum in a frequency interval or a quefrequency time interval selected in response to an operation input to the operation unit 46.

[0072] The feature data storage unit 57 classifies and stores the feature amounts extracted by the feature extraction unit 56 in the normal data collection mode according to the measurement conditions. Here, the measurement conditions include, for example, the sampling frequency in the time waveform data acquisition unit 35, and parameters related to the rotation of the motor 14 and the drive roller 12 as shown in FIG.

[0073] Hereinafter, a case will be described in which the feature quantity extractor 56 extracts the FFT centroid, which is the centroid of the waveform of the disturbance-removed power spectrum or the disturbance-removed spectrum in a certain frequency range, as a feature quantity of the waveform of the disturbance-removed power spectrum or the disturbance-removed spectrum.

[0074] FFT center of gravity G x ,G y is defined by the following equations (6a) to (6c).

[0075]

number

[0076] Here, n is the number of frequency points included in the frequency interval of interest (excluding 0 Hz). i is the frequency of the i-th frequency point in the frequency interval of interest, i.e., the frequency given by frequency resolution × i. i is the intensity (power or amplitude) of the disturbance-removed power spectrum or the disturbance-removed spectrum at the i-th frequency point. S is the sum of the intensity (power or amplitude) of the disturbance-removed power spectrum or the disturbance-removed spectrum at all frequency points included in the frequency interval of interest.

[0077] As shown in FIG. 8, the feature extraction unit 56 extracts, for example, data in section 2 of the logarithmic power spectrum after the disturbance removal shown in FIG. 7A, and calculates the FFT center of gravity G x ,G y Calculate.

[0078] FFT center of gravity G x ,G yis similar to the mechanical center of gravity. Figure 9 shows the concept of the mechanical center of gravity. The coordinates of the center of gravity (X G ,Y G ) are expressed as the following equations (7a) to (7c).

[0079]

number

[0080] where m i is the mass of the i-th weight. X i is the X coordinate of the i-th weight. i is the Y coordinate of the i-th weight. M is the sum of the masses of n weights. That is, the mass m i Power L i , X coordinate X i The frequency f i , Y coordinate Y i Power L i , and if we replace the sum of mass M with the sum of power S, the FFT center of gravity G x ,G y Equations (6a) to (6c) can be derived.

[0081] FFT center of gravity G x ,G y Unlike a simple power spectrum or a peak frequency of a spectrum, this shows the disturbance-removed power spectrum or disturbance-removed spectrum within a desired frequency interval compressed into two values ​​that well represent the shape of the entire waveform.

[0082] The feature extracted by the feature extracting unit 56 in at least one of the plurality of frequency intervals may be not only the FFT centroid but also the variance, skewness, or kurtosis of the FFT centroid.

[0083] The variance V of the FFT centroid extracted by the feature extraction unit 56 x ,V y is defined by the following equations (8a) to (8d).

[0084]

number

[0085] where L Vi is the squared amplitude of the undisturbed power spectrum at the i-th frequency point. Pi is the power of the undisturbed power spectrum at the ith frequency point. S V is the sum of the amplitudes of all frequency points included in the frequency interval of interest. P is the sum of the powers of all frequency points included in the frequency interval of interest.

[0086] That is, the variance V x and V y is the center of gravity of the power of the disturbance-removed power spectrum in the frequency interval of interest minus the square of the center of gravity of the amplitude of the disturbance-removed power spectrum raised to the 1 / 2 power.

[0087] The skewness α of the FFT centroid extracted by the feature extraction unit 56 x ,α y and the kurtosis β of the FFT centroid x ,β y is defined by equations (9a) to (9f) below m.

[0088]

number

[0089] The Mahalanobis distance calculation unit 58 calculates the Mahalanobis distance D M The Mahalanobis distance D M is a distance that takes into account the correlation between multidimensional variables and is different from the Euclidean distance. M The square of is basically χ 2 It is known to follow a distribution, e.g., D M The probability of being ≥ 3 is approximately 1%.

[0090] In the abnormality determination mode, the Mahalanobis distance calculation unit 58 substitutes the feature quantity x (bold characters are vector notation) based on the time waveform data obtained from the target device that may be abnormal into the following equation (10) to calculate the Mahalanobis distance D M Calculate.

[0091]

number

[0092] x (bold) is the FFT centroid G extracted by the feature extraction unit 56 x ,G y , the variance of the FFT centroid V x ,V y , skewness α of the FFT centroid x ,α y , or the kurtosis β of the FFT centroid x ,β y μ (bold) is a vector whose elements are the averages of elements x1 and x2 of multiple x (bold) collected in normal data collection mode.

[0093] For example, x1 and x2 are the FFT centroids G x ,G y Then, μ1 and μ2 are the FFT centroids G x ,G y is the average of Σ -1 is the inverse matrix of the variance-covariance matrix Σ, and is a matrix that rotates and scales the distribution of x (bold)-μ (bold) collected in the normal data collection mode to convert it into a distribution with mean 0 and standard deviation 1.

[0094] That is, the Mahalanobis distance D calculated by the Mahalanobis distance calculation unit 58 in the abnormality determination mode M is an index indicating how far the calculated feature value is from the feature value distribution of normal data collected in the normal data collection mode.

[0095] The Mahalanobis ellipse estimator 59 calculates μ (in bold) in Equation (10) and the inverse matrix Σ of the variance-covariance matrix from a plurality of feature quantities x (in bold) obtained based on normal time waveform data stored in the feature quantity data storage unit 57 in the normal data collection mode. -1 This is to calculate the following.

[0096] Furthermore, the Mahalanobis ellipse estimation unit 59 calculates the calculated μ (bold) and the inverse matrix Σ of the variance-covariance matrix. -1 Using the above, the threshold value T used in the abnormality determination unit 61 described later is calculated as shown in the following equation (11). M An ellipse showing a Mahalanobis distance equal to (hereinafter also referred to as a "Mahalanobis ellipse") is estimated in the two-dimensional space of the feature.

[0097]

number

[0098] The display unit 40 displays a scatter diagram in which the feature values ​​obtained based on the time waveform data of the object to be determined as an abnormality are plotted in the abnormality determination mode, with the first axis representing the first component of the feature value and the second axis representing the second component of the feature value. The scatter diagram also displays the Mahalanobis ellipse estimated by the Mahalanobis ellipse estimation unit 59.

[0099] For example, the feature is the FFT centroid G x ,G y If so, the first component of the feature is the FFT centroid G x The second component of the feature is the FFT centroid G y In this case, the first axis of the scatter plot is frequency, and the second axis of the scatter plot is power.

[0100] Hereinafter, when the feature quantity extraction unit 56 extracts the FFT centroid, which is the centroid of the waveform in a time interval of a certain quefrequency, as a feature quantity of the waveform of the cepstrum, f i Aq i By changing the formula to

[0101] FFT centroid G of cepstrum waveform x ,G y is defined by the following equations (12a) to (12c).

[0102]

number

[0103] Here, n is the number of quefrency measurement points included in the time interval of interest (excluding 0 Hz). i is the quefrency of the i-th measurement point in the time interval of interest, i.e., the quefrency given by the quefrency resolution × i. i is the strength (power or amplitude) of the cepstrum at the i-th measurement point. S is the sum of the strength (power or amplitude) of the cepstrum at all measurement points included in the time interval of interest.

[0104] As shown in FIG. 8, the feature extraction unit 56 extracts, for example, data from section 2 of the cepstrum shown in FIG. 7B, and calculates the FFT center of gravity G x ,G y Calculate.

[0105] FFT centroid G of cepstrum waveform x ,G y Unlike a simple cepstrum peak time, this represents the cepstrum within a desired time interval compressed into two values ​​that well represent the shape of the entire waveform.

[0106] The feature extracted by the feature extractor 56 in at least one of the multiple time intervals may be not only the FFT centroid of the cepstral waveform described above, but also the variance, skewness, or kurtosis of the FFT centroid of the cepstral waveform.

[0107] The variance V of the FFT centroid of the cepstrum waveform extracted by the feature extraction unit 56 x ,V yis defined by the following equations (13a) to (13d).

[0108]

number

[0109] where L Vi is the amplitude of the cepstrum at the i-th measurement point. Pi is the power of the cepstrum at the i-th measurement point. S V is the sum of the amplitudes of all measurement points included in the time interval of interest. S P is the sum of the powers of all measurement points included in the time interval of interest.

[0110] That is, the variance V x and V y is the centroid of the power of the cepstrum in the time interval of interest minus the square of the centroid of the amplitude of the cepstrum raised to the power of 1 / 2.

[0111] The skewness α of the FFT center of gravity of the cepstrum waveform extracted by the feature extraction unit 56 x ,α y and the kurtosis β of the FFT centroid of the cepstrum waveform x ,β y is defined by equations (14a) to (14f) below m.

[0112]

number

[0113] Regarding the cepstrum waveform, the Mahalanobis distance calculation unit 58 calculates the Mahalanobis distance D M The Mahalanobis ellipse estimation unit 59 can estimate the Mahalanobis ellipse of equation (11).

[0114] 10 shows an example of a graph displayed on the display unit 40, in which feature quantities obtained based on time waveform data of an object to be determined as abnormal are plotted. The three Mahalanobis ellipses shown in the scatter plot of FIG. 10 are plotted at the Mahalanobis distance D M are ellipses 1, 2, and 3, respectively.

[0115] For example, the threshold value T used in the abnormality determination unit 61 described later M If is 3, the abnormality determination unit 61 M Any feature plotted outside the Mahalanobis ellipse of σ = 3 will be judged to be abnormal.

[0116] The abnormality degree calculation unit 60 calculates the Mahalanobis distance D obtained based on the time waveform data of the object to be determined as abnormal. M The degree of abnormality of the time waveform data to be judged as abnormal is calculated from the above.

[0117] The degree of abnormality is the Mahalanobis distance D M For example, the degree of anomaly is determined by the Mahalanobis distance D M may be equal to

[0118] FIG. 11(a) and equation (15) show that the degree of abnormality A calculated by the abnormality degree calculation unit 60 is smaller than the Mahalanobis distance D M In this example, the minimum value of the abnormality level A is 0 and the maximum value diverges.

[0119]

number

[0120] Figure 11(b) and equation (16) show that the degree of anomaly A is proportional to the Mahalanobis distance D M In this example, the anomaly level A has a minimum value of 0 and a maximum value of less than 100.

[0121]

number

[0122] Figure 11(c) and equation (17) show that the degree of anomaly A is proportional to the Mahalanobis distance D M This shows an example where the anomaly level A increases in a stepwise manner, with a broad monotonic increase. In this example, the minimum value of the anomaly level A is 0 and the maximum value diverges.

[0123]

number

[0124] The display unit 40 may display the time change of the degree of abnormality A calculated by the degree of abnormality calculation unit 60, for example, as shown in FIG. 12, with the first axis representing time and the second axis representing the degree of abnormality. Here, the time on the first axis is, for example, the operating time of the predictive maintenance device 1 when the feature quantity extraction unit 56 extracts the feature quantity from the time waveform data of the object to be determined as abnormal. Note that FIG. 12 shows the case where the degree of abnormality A is calculated based on the Mahalanobis distance D M shows an example equivalent to

[0125] The display unit 40 may convert the degree of abnormality calculated by the abnormality degree calculation unit 60 into characters or symbols and display them. Fig. 13 is a table showing an example of the correspondence between the degree of abnormality and symbols or character strings. Depending on the operation input by the user to the operation unit 46, the display format of the degree of abnormality on the display unit 40 may be selected from, for example, numerical values, characters, or symbols.

[0126] The abnormality determination unit 61 determines the Mahalanobis distance D M Or the abnormality level is threshold T M When this threshold T M may be set to an arbitrary value in response to an operation input to the operation unit 46 by the user, for example.

[0127] The display unit 40 displays the abnormality determination result determined by the abnormality determination unit 61.

[0128] 14A shows a normal operation screen 42 displayed by the display unit 40 when an article inspection device equipped with the target device inspects a transported item W. When the abnormality determination unit 61 determines that the target device is abnormal, the display unit 40 displays an abnormality sign alarm as the abnormality determination result in the alarm display area 41. At this time, the external output unit 62 may be configured to output the abnormality sign alarm externally via serial communication or external contact output.

[0129] When the user presses the alarm display area 41 by operating the operation unit 46, the display screen of the display unit 40 transitions from the normal operation screen 42 to an abnormality sign display screen 43, as shown in Fig. 14B. The abnormality sign display screen 43 displays a message 43a urging the user to replace the motor 14, which is the target device, a diagram 43b showing the location of the target device, and a details button 43c.

[0130] When the user presses the details button 43c by operating the operation unit 46, the display screen of the display unit 40 transitions from the abnormality sign display screen 43 to the maintenance screen 44, as shown in Fig. 14C. The maintenance screen 44 displays a table showing the name of the target device, the degree of abnormality in each section (frequency section or time section), etc.

[0131] When the user operates the operation unit 46 to select, for example, the display location of motor section 1 on the maintenance screen 44, the display screen of the display unit 40 transitions from the maintenance screen 44 to a status monitoring screen 45, as shown in Fig. 14D. On the status monitoring screen 45, for example, a scatter diagram 45a of feature quantities as shown in Fig. 10 and an abnormality degree 45b calculated by the abnormality degree calculation unit 60 are displayed.

[0132] FIG. 14E shows a mode selection screen 48a displayed on the display unit 40 for allowing the mode selection unit 51 to select an operation mode.

[0133] When the user selects "YES" on the mode selection screen 48a by operating the operation unit 46, the operating mode switches to the normal data collection mode, and the display screen of the display unit 40 transitions from the mode selection screen 48a to the data collection execution screen 48b, as shown in FIG. 14F.

[0134] When the user selects "Execute" on the data collection execution screen 48b by operating the operation unit 46, collection of the feature amounts of normal data starts, and the elapsed time since the start of collection of the feature amounts of normal data is displayed.

[0135] 14G shows a data registration screen 48c that is displayed on the display unit 40 when collection of the feature amounts of normal data is completed in the normal data collection mode. When the user selects "Register" on the data registration screen 48c by operating the operation unit 46, the feature amounts of the collected normal data are stored in the feature data storage unit 57.

[0136] An example of the processing of a predictive maintenance method using the predictive maintenance device 1 will be described below with reference to the flowcharts of Figures 15 and 16. Note that descriptions that overlap with the description of the configuration of the predictive maintenance device 1 described above will be omitted as appropriate.

[0137] The flowchart of Figure 15 shows the processing that is performed when the mode selection unit 51 selects the normal data collection mode as a result of the user selecting "YES" by operating the operation unit 46 on the mode selection screen 48a shown in Figure 14E.

[0138] First, the user inputs operation into the operation unit 46 to set various measurement conditions such as the sampling frequency in the time waveform data acquisition unit 35, the rotation speed and number of poles of the motor 14, the conveying speed of the conveying unit 10, and the number of collections (several hundred points) (step S1).

[0139] Next, the interval division unit 55 calculates the interval widths and center frequencies (peak frequencies) of multiple frequency intervals, or the interval widths and center times (peak times) of multiple quefrency time intervals, based on the measurement conditions set in step S1 (step S2).

[0140] Next, by operating the operation unit 46, one or more desired frequency intervals or quefrency time intervals are selected from the plurality of frequency intervals or quefrency time intervals calculated by the interval dividing unit 55 (step S3).

[0141] Next, when an operation input instructing the start of measurement is made to the operation unit 46, the transport unit 10 starts driving the motor 14 (step S4). The measurement is performed with the transport unit 10, which is a transport belt conveyor, running idle. The timing of the start of measurement is determined by a user's arbitrary operation, regular measurement, or immediately after operation has started.

[0142] Next, the time waveform data acquiring unit 35 acquires the time waveform data of the one-phase current or the time waveform data of the total current measured by the measuring circuit 30 (time waveform data acquiring step S5).

[0143] Next, the FFT unit 52 performs FFT on the time waveform data acquired by the time waveform data acquisition unit 35 to obtain the spectrum of the time waveform data (FFT step S6).

[0144] Next, the logarithmic power spectrum calculation unit 53 calculates a logarithmic power spectrum expressed in dB from the spectrum obtained by the FFT unit 52 (logarithmic power spectrum calculation step S7).

[0145] Next, the cepstrum calculation unit 54a calculates a cepstrum from the logarithmic power spectrum calculated by the logarithmic power spectrum calculation unit 53 (cepstrum calculation step S8). Also, in step S8, the logarithmic power cepstrum calculation unit 54c calculates a logarithmic power cepstrum for screen display in dB from the cepstrum calculated by the cepstrum calculation unit 54a.

[0146] Next, the disturbance removal unit 54b subtracts the DC component corresponding to the zeroth-order component of the cepstrum calculated by the cepstrum calculation unit 54a from the logarithmic power spectrum calculated by the logarithmic power spectrum calculation unit 53, thereby obtaining a disturbance-removed power spectrum (disturbance removal step S9).

[0147] Next, the interval dividing unit 55 divides the disturbance-removed power spectrum calculated by the disturbance removing unit 54b into a plurality of frequency intervals (step S10). Alternatively, in step S10, the interval dividing unit 55 divides the cepstrum calculated by the cepstrum calculating unit 54a into a plurality of quefrequency time intervals.

[0148] Next, the feature extraction unit 56 extracts two-dimensional features of the waveform of the disturbance-removed power spectrum in one or more frequency intervals selected in step S3 from among the multiple frequency intervals divided by the interval division unit 55, and stores the extracted features in the feature data storage unit 57 (feature extraction step S11). Alternatively, in step S11, the feature extraction unit 56 extracts two-dimensional features of the waveform of the cepstrum in one or more quefrequency time intervals selected in step S3 from among the multiple quefrequency time intervals divided by the interval division unit 55, and stores the extracted features in the feature data storage unit 57.

[0149] Next, the Mahalanobis ellipse estimation unit 59 calculates the mean μ (in bold) of the feature in Equation (10) and Equation (11) and the inverse matrix Σ of the variance-covariance matrix from the feature stored in the feature data storage unit 57 for each frequency interval or quefrency time interval selected in step S3. -1 is calculated (step S12).

[0150] Next, when the user inputs an instruction to end the collection of the feature amounts of normal data through the operation unit 46 (step S13: YES), the series of processes ends. On the other hand, when the user does not input an instruction to end the collection of the feature amounts of normal data through the operation unit 46 (step S13: NO), the processes from step S5 onwards are executed again.

[0151] The flowchart in Figure 16 shows the processing that is performed when the user selects "NO" by operating the operation unit 46 on the mode selection screen 48a shown in Figure 14E, causing the mode selection unit 51 to select the abnormality determination mode for normal use.

[0152] First, various measurement conditions such as the sampling frequency in the time waveform data acquisition unit 35, the rotation speed and number of poles of the motor 14, the conveying speed of the conveying unit 10, and the number of judgment processes (several to several tens of points) are set by operating the operation unit 46 (step S21).

[0153] Next, the interval division unit 55 calculates the interval widths and center frequencies (peak frequencies) of multiple frequency intervals, or the interval widths and center times (peak times) of multiple quefrency time intervals, based on the measurement conditions set in step S21 (step S22).

[0154] Next, by operating the operation unit 46, one or more desired frequency intervals or quefrency time intervals are selected from the plurality of frequency intervals or quefrency time intervals calculated by the interval dividing unit 55 (step S23).

[0155] Next, when an operation input instructing the start of measurement is made to the operation unit 46, the transport unit 10 starts driving the motor 14 (step S24). The measurement is performed with the transport unit 10, which is a transport belt conveyor, running idle. The timing of the start of measurement is determined by a user's arbitrary operation, regular measurement, or immediately after operation has started.

[0156] Next, the time waveform data acquiring unit 35 acquires the time waveform data of the one-phase current or the time waveform data of the total current measured by the measuring circuit 30 (time waveform data acquiring step S25).

[0157] Next, the FFT unit 52 performs FFT on the time waveform data acquired by the time waveform data acquisition unit 35 to obtain the spectrum of the time waveform data (FFT step S26).

[0158] Next, the logarithmic power spectrum calculation unit 53 calculates a logarithmic power spectrum expressed in dB from the spectrum obtained by the FFT unit 52 (logarithmic power spectrum calculation step S27).

[0159] Next, the cepstrum calculation unit 54a calculates a cepstrum from the logarithmic power spectrum calculated by the logarithmic power spectrum calculation unit 53 (cepstrum calculation step S28). Also, in step S28, the logarithmic power cepstrum calculation unit 54c calculates a logarithmic power cepstrum for screen display in dB notation from the cepstrum calculated by the cepstrum calculation unit 54a.

[0160] Next, the disturbance removal unit 54b subtracts the DC component corresponding to the zeroth-order component of the cepstrum calculated by the cepstrum calculation unit 54a from the logarithmic power spectrum calculated by the logarithmic power spectrum calculation unit 53, thereby obtaining a disturbance-removed power spectrum (disturbance removal step S29).

[0161] Next, the interval dividing unit 55 divides the disturbance-removed power spectrum calculated by the disturbance removing unit 54b into a plurality of frequency intervals (step S30). Alternatively, in step S30, the interval dividing unit 55 divides the cepstrum calculated by the cepstrum calculating unit 54a into a plurality of quefrency time intervals.

[0162] Next, the feature extraction unit 56 extracts two-dimensional feature amounts of the waveform of the disturbance-removed power spectrum in one or more frequency intervals selected in step S23 from among the multiple frequency intervals divided by the interval division unit 55 (feature extraction step S31). Alternatively, in step S31, the feature extraction unit 56 extracts two-dimensional feature amounts of the waveform of the cepstrum in one or more quefrequency time intervals selected in step S23 from among the multiple quefrequency time intervals divided by the interval division unit 55.

[0163] Next, the Mahalanobis distance calculation unit 58 calculates the Mahalanobis distance D of the feature extracted by the feature extraction unit 56 in step S31 for each frequency interval or quefrency time interval selected in step S23. M is calculated using equation (10) (Mahalanobis distance calculation step S32).

[0164] Next, the abnormality degree calculation unit 60 calculates the Mahalanobis distance D calculated by the Mahalanobis distance calculation unit 58. M The degree of abnormality of the time waveform data to be judged as abnormal is calculated from the above (step S33). In the abnormality judgment mode, abnormality is judged from several to several tens of points.

[0165] Next, the display unit 40 plots and displays the feature quantities extracted by the feature quantity extraction unit 56 in step S31 on a scatter plot in which the Mahalanobis ellipse estimated by the Mahalanobis ellipse estimation unit 59 is displayed, with the first axis representing the first component of the feature quantity and the second axis representing the second component of the feature quantity. Furthermore, the display unit 40 displays the latest degree of abnormality calculated by the degree of abnormality calculation unit 60 in step S33 (step S34).

[0166] Next, the abnormality determination unit 61 calculates the Mahalanobis distance D M is the threshold T M If it is determined that the number of times exceeds the specified number (the abnormality determination step S35: YES), the process of step S36 is executed. M is the threshold T MIf it is determined that the specified number of times has not been exceeded (abnormality determination step S35: NO), the processing from step S25 onwards is executed again.

[0167] In step S36, the display unit 40 displays an abnormality symptom alarm in the alarm display area 41 (display step S36).

[0168] As described above, the predictive maintenance device 1 according to this embodiment acquires time waveform data of the current or voltage of the target device, calculates a disturbance-removed power spectrum from the acquired time waveform data by removing the influence of disturbances, and extracts two-dimensional feature quantities such as the FFT centroid from the disturbance-removed power spectrum and cepstrum. Furthermore, the predictive maintenance device 1 according to this embodiment calculates a two-dimensional feature quantity such as the FFT centroid from the disturbance-removed power spectrum and cepstrum when the Mahalanobis distance of the feature quantity is less than or equal to a threshold T M If the value exceeds this, the target device is determined to be abnormal.

[0169] As a result, the predictive maintenance device 1 of this embodiment can accurately detect signs of abnormality in the target device, making it possible to replace the target device with a new one at the appropriate time before the target device breaks down, and significantly reducing the downtime of an item inspection device equipped with the target device.

[0170] Furthermore, the predictive maintenance device 1 according to this embodiment can eliminate the effects of disturbances that are mainly caused by changes in the operating environment inside and outside the target device, and therefore can acquire features and perform predictive maintenance immediately after the target device starts operating.

[0171] Furthermore, the predictive maintenance device 1 according to this embodiment divides the disturbance-removed power spectrum into a plurality of frequency intervals each including a center frequency (peak frequency) corresponding to the operating state of the target device, and extracts waveform feature quantities of the disturbance-removed power spectrum for each of these frequency intervals. Furthermore, it calculates peak times of a plurality of quefrencies in the cepstrum corresponding to the operating state of the target device, and extracts waveform feature quantities of the cepstrum in a plurality of quefrency time intervals each including each peak time.

[0172] As a result, the predictive maintenance device 1 according to this embodiment can detect small changes in the disturbance-removed power spectrum or cepstrum by reducing the number of data items to be processed when extracting waveform features of the disturbance-removed power spectrum or cepstrum.

[0173] Furthermore, the predictive maintenance device 1 according to this embodiment can obtain the FFT centroid, which is the center of gravity of the disturbance-removed power spectrum or cepstrum, for each divided frequency interval or quefrequency time interval, thereby capturing the overall feature values ​​of the waveform of the divided disturbance-removed power spectrum or cepstrum.

[0174] Furthermore, the predictive maintenance device 1 according to this embodiment displays the threshold value T for anomaly determination on a graph having a first axis indicating the first component of the feature amount and a second axis indicating the second component of the feature amount. M and a feature value obtained based on the time waveform data of the object to be determined to be abnormal.

[0175] As a result, the predictive maintenance device 1 according to this embodiment calculates the feature amount obtained based on the time waveform data of the object to be determined as abnormality by using the threshold value T M This gives a visual indication of how far away you are from the target.

[0176] In addition, the predictive maintenance device 1 of this embodiment can calculate a cepstrum from the logarithmic power spectrum of the time waveform data of the current or voltage of the target device, and generate a disturbance-removed power spectrum by subtracting the DC component corresponding to the zeroth-order component of the cepstrum from the logarithmic power spectrum. [Explanation of symbols]

[0177] 1. Predictive maintenance equipment 10 Conveyor 11 Conveyor belt 11a Conveying surface 12 Drive roller (target device) 13 Driven roller 14 Motor (target device) 20 Motor drive circuit 21 DC power supply 22 Positive side power supply 23 Negative power supply 24 Inverter circuit 24a~24f Switching elements 24g U phase output terminal 24h V phase output terminal 24i W phase output terminal 32 Shunt resistor 30 Measurement circuit 35 Time waveform data acquisition section 40 Display section 41 Alarm display area 46 Control section 50 control section 51 Mode selection section 52 FFT section 53 Logarithmic power spectrum calculation section 54a Cepstrum calculation unit 54b Disturbance removal section 54c Logarithmic power cepstrum calculation unit 55 Section division section 56 Feature Extraction Unit 57 Feature data storage unit 58 Mahalanobis distance calculation section 59 Mahalanobis ellipse estimator 60 Abnormality calculation part 61 Abnormality determination section 62 External output section W Conveyed object

Claims

1. a measurement circuit (30) for measuring a current or a voltage of the target device (12, 14); a time waveform data acquisition unit (35) that acquires time waveform data of the current or the voltage measured by the measurement circuit; an FFT unit (52) that performs a Fourier transform on the time waveform data to obtain a spectrum of the time waveform data; a logarithmic power spectrum calculation unit (53) that calculates a logarithmic power spectrum from the spectrum; a cepstrum calculation unit (54a) for calculating a cepstrum from the logarithmic power spectrum; a disturbance removal unit (54b) for subtracting a DC component of the logarithmic power spectrum from the logarithmic power spectrum to obtain a disturbance-removed power spectrum; a feature extraction unit (56) for extracting two-dimensional feature quantities of the waveforms of the disturbance-removed power spectrum and the cepstrum; a Mahalanobis distance calculation unit (58) for calculating the Mahalanobis distance of the feature amount; an abnormality determination unit (61) that determines that the target device is abnormal when the Mahalanobis distance exceeds a threshold; A predictive maintenance device comprising: a display unit (40) that displays the abnormality determination result determined by the abnormality determination unit.

2. a section dividing unit (55) that calculates a plurality of peak frequencies in the disturbance-removed power spectrum according to the operating state of the target device, divides the disturbance-removed power spectrum into a plurality of frequency sections each including each of the peak frequencies, calculates peak times of a plurality of quefrencies in the cepstrum according to the operating state of the target device, and divides the cepstrum into a plurality of time sections each including each of the peak times, 2. The predictive maintenance device according to claim 1, wherein the feature extractor extracts the feature of the waveform of the disturbance-removed power spectrum in at least one of the plurality of frequency intervals, and extracts the feature of the waveform of the cepstrum in at least one of the plurality of time intervals.

3. 3. The predictive maintenance device according to claim 2, wherein the feature is a center of gravity of the disturbance-removed power spectrum in at least one of the plurality of frequency intervals or a center of gravity of the cepstrum in at least one of the plurality of time intervals.

4. a Mahalanobis ellipse estimation unit (59) for estimating an ellipse indicating the Mahalanobis distance equal to the threshold value in a two-dimensional space of the feature amount from the feature amount obtained based on the normal time waveform data, The predictive maintenance device according to claim 1 or 2, characterized in that the display unit further displays the ellipse and the feature obtained based on the time waveform data of the object to be determined as an anomaly on a graph having a first axis indicating a first component of the feature and a second axis indicating a second component of the feature.

5. the cepstrum calculation unit performs an inverse Fourier transform on the logarithmic power spectrum to obtain the cepstrum; 3. The predictive maintenance device according to claim 1, wherein the disturbance removal unit calculates the disturbance-removed power spectrum by subtracting the DC component corresponding to the zeroth-order component of the cepstrum from the logarithmic power spectrum.

6. 3. The predictive maintenance device according to claim 1, further comprising a logarithmic power cepstrum calculation unit (54c) that calculates a logarithmic power cepstrum from the cepstrum.

7. a time waveform data acquisition step (S5, S25) of acquiring time waveform data of the current or voltage of the target device (12, 14) measured by the measurement circuit (30); an FFT step (S6, S26) of Fourier transforming the time waveform data to obtain a spectrum of the time waveform data; a logarithmic power spectrum calculation step (S7, S27) of calculating a logarithmic power spectrum from the spectrum; a cepstrum calculation step (S8, S28) for obtaining a cepstrum from the logarithmic power spectrum; a disturbance removal step (S9, S29) of subtracting a DC component of the logarithmic power spectrum from the logarithmic power spectrum to obtain a disturbance-removed power spectrum; a feature extraction step (S11, S31) of extracting two-dimensional feature values ​​of the waveforms of the disturbance-removed power spectrum and the cepstrum; a Mahalanobis distance calculation step (S32) of calculating a Mahalanobis distance of the feature amount; an abnormality determination step (S35) of determining that the target device is abnormal when the Mahalanobis distance exceeds a threshold; A display step (S36) of displaying the abnormality determination result determined in the abnormality determination step.

Images