TWI481828B - Abnormal sound diagnostic device - Google Patents

Description

Abnormal sound diagnostic device

The present invention relates to a determination device that collects sound generated from a device to be inspected, and determines a possibility of occurrence of an abnormal sound of a device in operation based on a time-frequency analysis of the sound of the collected sound.

Regarding the abnormal sound diagnosis device, an abnormal sound diagnosis device as disclosed in Patent Document 1 is known. In the abnormal sound diagnostic apparatus disclosed in Patent Document 1, the memory device stores, as a reference value, the sound data of the device to be inspected that the human ear hears as the normal sound, and the reference value and the measuring device in the device stored in advance by the memory device. The degree of coincidence between the measurement data of the sound generated by the inspection target device at the time of the diagnosis is calculated by the processing device, and the determination device determines the abnormal sound of the inspection target device based on the degree of matching of the results calculated by the processing device. Good or not.

Further, the processing device that calculates the degree of matching between the reference value of the sound data of the inspection target device and the sound measurement data at the time of diagnosis in the inspection target device is input with the amplitude value of each frequency of the reference value, and has the same amplitude value. The straight line of the scale of the output is used as a reference straight line, and the difference between the set straight lines of the amplitude values from the respective frequencies of the above-mentioned measurement data is calculated by the least square method, and the result is an index of the degree of coincidence.

In addition, when the amplitude value of each of the measured frequencies is lower than the set straight line, the amplitude value of each of the above-mentioned frequencies is assumed to be corrected on the set straight line to prevent the correction. The degree of agreement between the sound data of the reference values generated by zero or low amplitude due to poor machine is not integrated.

[advance technical documents]

[Patent Document 1] Japanese Laid-Open Patent Publication No. 2005-283227

However, the conventional abnormal sound diagnosis, according to various conditions of temperature, air pressure, humidity, speed, acceleration, pressure, tension, load, etc., checks the time-frequency characteristics of the sound emitted by the target machine at a specific specific time and frequency because Regardless of the difference in the change, the time-frequency characteristic of the reference sound material is different from the above-described respective conditions of the time-frequency characteristics of the input sound data at the time of diagnosis, and the change in the specific time and frequency is erroneously detected as an abnormality.

The abnormal sound diagnostic apparatus according to the present invention acquires a sound generated by a device to be inspected at the time of learning and at the time of diagnosis, and compares the sound at the time of learning and the diagnosis, and diagnoses the abnormality of the sound, and includes a sound data acquiring device and the above-mentioned inspection target machine. Synchronizing the movements to obtain the sound data generated by the inspection target machine; the analysis device obtains the specimen series composed of the intensity of each time based on the above-mentioned sound data; A memory device, a series of specimens for memory learning, as a reference specimen series; a correction device that corrects a target specimen series or a reference specimen based on a correction amount or a correction amount series estimated from a target specimen series of the specimen series at the time of diagnosis and the above-described reference specimen series. And at least one of the series; and the determination device compares the corrected target specimen series or the reference specimen series with the corresponding reference specimen series or the target specimen series, calculates the abnormality, and compares the calculated abnormality with the predetermined critical value. The output determines the degree of abnormality.

According to the abnormal sound diagnostic apparatus of the present invention, it is possible to change the environmental conditions, speed, acceleration, pressure, tension, load, and the like of the inspection target machine such as temperature, air pressure, humidity, and the like of the time and frequency characteristics of the sound emitted by the inspection target device. Each condition of the target machine operating condition is considered to be different in the time and frequency of the correction sound correction data at a specific specific time and frequency, and the change in the specific time and frequency generated by the above different conditions can be reduced. The effect of false detection of the possibility of an abnormality.

1‧‧‧concentrator

2‧‧‧ Waveform acquisition department

3‧‧‧ Waveform data

4‧‧‧Time Frequency Analysis Department

5‧‧‧Time frequency distribution

15‧‧‧Abnormality calculation department

16‧‧‧ anomalies

17‧‧‧Decision Department

18‧‧‧Results

501‧‧‧Specialization Department

502‧‧‧ Specimen series

503‧‧‧Memory Department

506‧‧‧ benchmark specimen series

507‧‧‧Object specimen series

509‧‧‧Revised object specimen series

601‧‧‧Revision calculation unit

602‧‧‧ Correction

603‧‧‧Revision Department

701‧‧‧Control signal

702‧‧‧Time Synchronization Department

703‧‧‧Operational State Estimation Department

704‧‧‧Inferred motion speed series

[Fig. 1] is a block diagram showing an abnormal sound diagnosis apparatus according to a first embodiment of the present invention; [Fig. 2] is a schematic diagram showing a sample series before and after correction by a correction unit of the first embodiment; [Fig. 3] is a schematic view showing a sampled series before and after correction by the correction unit of the second embodiment; [Fig. 4] showing the correction before and after correction by the correction unit of the third embodiment. A schematic diagram of the normalized series; [Fig. 5] is a schematic diagram showing a series of corrections before and after correction by the correction section of the fourth embodiment; [Fig. 6] shows an abnormal sound according to the fifth embodiment of the present invention. A block diagram of the diagnostic device; [Fig. 7] is a schematic diagram showing a sampled series before and after correction by the correction unit of the fifth embodiment; [Fig. 8] is from the acquisition of the measurement signal from the sound collector to A flowchart of processing until the learning mode or the diagnostic mode is judged; [Fig. 9] is a flowchart of processing in the learning mode; and [Fig. 10] is a processing flowchart in the case of the diagnostic mode.

[First Embodiment]

The present embodiment is a device for diagnosing an abnormal sound emitted from a machine to be inspected, and is actually assembled as a software on a personal computer (hereinafter referred to as a PC), and has a learning mode in which a waveform in a normal state is taken and a diagnosis mode in which a waveform is taken during a test. . The measurer sets a microphone, an acoustic sensor, an accelerometer, and the like on the inspection target machine, and connects the sound collector to the input terminal of the USB (Universal Serial Bus) interface of the PC, when performing the learning mode or the diagnosis. The operation at the time of the mode.

The object to be inspected, for example, a machine consisting of a plurality of running parts like a lift. In the case of an elevator, the sound collector is installed in or out of the cage, and the signal of the sound collector is taken in via the control cable to the PC placed in the machine room, and the machines of the elevator are diagnosed by moving up and down the cage. The sound of the operation.

Since the lift is composed of a plurality of parts, the sound of the sound collector is collected. The phonology is a mixture of these parts. Moreover, the sound frequency characteristics produced by each part are different. Moreover, the position of the sound collector moves with time due to the ride cage, and the time-frequency component differs depending on the time and time-frequency component. Moreover, the sound intensity generated by each component is different depending on various conditions such as temperature, air pressure, humidity, speed, acceleration, pressure, tension, and load, and different changes of each component are displayed. For example, the orbital sound generated when the rail and the rail are slid, the viscosity of the oil applied to the rail decreases as the temperature rises, and when the temperature rises, the friction becomes small and the sound pressure drops. Conversely, the sound produced by the cable coiled around the pulley does not appear to be temperature dependent. Further, the air-conditioning sound system increases in the number of rotations of the fan as the temperature rises, and it is seen that the strength tends to increase.

Therefore, the sound time-frequency distribution of the sound collector's sound collection is based on the change of the above-mentioned conditions. Since the time-frequency characteristic is different for each specific time and frequency, even if the machine is normal, the sound collector is collected according to the above various conditions. The sound time frequency also changes every specific time and frequency, and it is highly probable that the change component generated by the above-described different conditions is a component generated by the abnormal sound.

Fig. 1 is a block diagram showing an abnormal sound diagnosis apparatus according to a first embodiment of the present invention. In the first figure, a sound collector that is a microphone, an acoustic sensor, a vibration sensor, or the like, and a signal acquisition unit that samples the signal from the sound collector 1 into a digital signal and outputs the waveform data 3, 4 series The time-frequency analysis unit attaches a time window to the waveform data 3, and shifts the time window in the time direction, and analyzes the waveform data 3 according to the Fourier transform (hereinafter referred to as FFT) calculus, and outputs the intensity for time and frequency. The time-frequency distribution of the spectrum value is 5,501 is the specimenization unit, and the spectrum values of the respective frequency intensities are sampled and displayed at each time from the time-frequency distribution 5, and the respective frequencies of the obtained time series are output. Specimen series 502. The specimen series 502 is a time series of standard values for each time of a predetermined frequency.

Further, in the following description of the embodiment, the case where the time-frequency distribution 5 outputted by the time-frequency analysis unit 4 is used as the sound data will be described. Waveform data and other feature quantities obtained by analyzing the waveform data may be used as the sound data.

503 is a memory part of the specimen series 502 which is obtained from the sound data obtained at the time of learning, and 506 is a reference specimen series which is stored in the memory unit 503 during the learning, and which is a series of specimens of the frequencies at which the abnormality is calculated. In addition, the 507 series is obtained from the sound data acquired at the time of diagnosis, and is a target specimen series composed of a series of specimens of the respective frequencies at which the abnormality is calculated.

The 601 series correction amount calculation unit calculates the correction amount 602 and the 603 series correction unit with reference to the reference sample series 506 and the target sample series 507, and corrects the target sample series 507 based on the correction amount 602, and outputs the corrected target sample series 509.

The 15th degree of abnormality calculation unit calculates the degree of abnormality indicating the degree of possibility of abnormal noise generation by referring to the reference sample series 506 and the corrected target sample series 509 for each of the predetermined frequencies, and outputs the abnormality degree 16 and the 17-series determination unit. The probability of occurrence of an abnormal sound is determined based on the abnormality 16 of each of the predetermined frequencies, and the determination result 18 is output.

The operation will be described below with reference to the processing flowcharts of Figs. 8 to 10.

In the learning mode or the diagnostic mode, the waveform obtaining unit 2 obtains the measurement signal output from the sound collector 1 and amplifies it, and samples and converts the measurement signal into a 16-bit linear PCM with a sampling frequency of 48 kHz by pulse conversion (pulse code modulation). Waveform data 3 of the digital signal (step S1 of Fig. 8).

The time-frequency analysis unit 4 cuts out the frame of the waveform data 3 output from the waveform acquisition unit 2 at a time interval of 16 milliseconds at intervals of 16 milliseconds, and obtains a frequency spectrum from each frame based on the FFT calculation. The series y(t, f) is output as the time frequency distribution 5 (step S2 of Fig. 8). Here, t is an index corresponding to the time at which the displacement interval of the analysis window is shifted, and f is an index showing the frequency of the FFT calculation result. Further, the time t and the frequency f satisfy the relationship of 0≦t≦T, 0≦f≦F, respectively. Here, the number of frames in the time direction of the T-time frequency distribution 5, and F is an index indicating the frequency of the Nyquist frequency (N=quist frequency) corresponding to 1/2 of the sampling frequency fs of the waveform data 3 (F=fs) /2).

The specimenization unit 501 is based on the time-frequency distribution 5, and each of the predetermined frequencies has a center frequency of 0.5 kHz, 1 kHz, 2 kHz, 4 kHz, and 8 kHz, and is included in the five frequency bands formed by the frequency bands of the single-tone width. The frequency components in the five frequency bands are in units of 8 frames, and the sum of the spectral values is obtained, and the sample values of the predetermined time in units of 8 frames (256 milliseconds) are obtained, and the sample series 502 is output (Fig. 8 Step S3). Now, when the standard value in the specimen series 502 of each frequency and time is assumed to be Y(n, b), Y(n, b) is calculated by the equation (1-1).

Here, the time index of the n-series standardization series, the natural number of the range of 1~N (however, N is the upper limit of the time range, the remainder is rounded off by N=T/8), the b-system frequency index, 1~B The natural number of the range (the number of B-line bands, B = 5 in this embodiment). Further, in the time-frequency distribution y(t, f), Ω(n, b) indicates a set of time and frequency groups (t, f) to be the target of the summation for the standardization.

When the specimen series 502 is acquired by the specimen processing unit 501, the abnormal sound diagnostic apparatus determines whether it is the learning mode or the diagnostic mode (step S4 of Fig. 8).

In the learning mode, for each frequency (step S201 of Fig. 9), the memory unit 503 stores the specimen series 502 as the reference specimen series 506 (step S202 of Fig. 9).

Next, the operation of the diagnosis processing in the diagnosis mode will be described.

In the case of the diagnosis mode, for each frequency (step S301 of Fig. 10), the specimen series 502 is used as the target sample series 507 for diagnosis (step S302 of Fig. 10).

The correction amount calculation unit 601 calculates the correction amount 602 based on the reference specimen series 506 stored in the storage unit 503 and the target specimen series 507 which the specimenization unit 501 has standardized for each frequency (step S303 in Fig. 10). The details are calculated by the next steps "A1-1" to "A1-2".

"A1-1" finds the specimen series D(n, b) of the difference between the target specimen series Y ₁ (n, b) of the frequency b and the reference specimen series Y ₀ (n, b) of the frequency b. (Formula (2-1))

"A1-2" finds the average of the time n of the difference sample series D(n, b), and outputs the correction amount H ₁ (b). (Formula (2-2))

Further, the equation (2-2) finds the average of the difference series, and since the right side of the equation (2-2) can be deformed as the right side of the equation (2-3), in order to obtain the correction amount H ₁ (b), each is obtained. After averaging the specimen series, the difference between the averages is obtained, and the correction amount H ₁ (b) may be calculated.

The correction unit 603 obtains the corrected target sample series 509 by subtracting the correction amount H ₁ (b) from the target specimen series 507 (step S304 in Fig. 10). When the object specimen series of the object whose frequency b is corrected is assumed to be Y ₁₁ (n, b), the correction amount is subtracted from the target specimen series Y ₁ (n, b) of the frequency b as in the equation (2-4). H ₁ (b), calculate Y ₁₁ (n, b).

[Number 4] Y ₁₁ (n, b) = Y ₁ (n, b) - H ₁ (b) ... (2-4)

Here, Y ₁₁ (n, b) is the corrected target sample series 509.

Fig. 2 is a schematic view showing a sample series before and after correction by the correction unit 603.

(A) shows the relationship between the reference specimen series before correction and the target specimen series. At this time, since the temperature rises, the specimen value of the target specimen series becomes substantially the same as the reference specimen series at the same time. The arrow indicates the direction of correction based on the correction amount.

(B) shows the relationship between the corrected reference specimen series and the target specimen series. It is understood that the correction is to make the average of the difference series of the corrected reference specimen series and the target specimen series close to zero.

The abnormality calculating unit 15 inputs the reference sample series 506 and the corrected target sample series 509 for each frequency, calculates the difference between the two, and outputs the abnormality 16 (step S305 in Fig. 10). The details are, for each frequency b, the degree of abnormality of the difference between the reference specimen series Y ₀ (n, b) and the corrected target specimen series Y ₁₁ (n, b) is calculated as the square mean of the difference between the specimen series (sample) When the series is regarded as a vector of the N-dimensional space, it has the same value as the Euclidean distance (Equation (3-1)).

Here, in the above formula, a(b) is the degree of abnormality of the frequency b. Also, the range of n in the sum is the full time (1≦n≦N).

The determination unit 17 determines whether or not there is a possibility of generating an abnormal sound by comparing the total degree of abnormality calculated based on the abnormality 16 of each frequency output by the abnormality calculating unit 15 with the predetermined threshold value, and outputs the result as the determination result 18 (10th) Step S307) of the figure. The details are as follows: First, the normalized abnormality â(b) (Expression (3-2)) of the abnormality a(b) at each frequency of the dispersion normalization is obtained for each frequency. Next, assume that the maximum value of the frequency with respect to the normalized abnormality is the total agglomeration a* (formula (3-3)). Finally, the total degree of abnormality a* calculated by the equation (3-3) and the critical value are compared, and when the total degree of abnormality is equal to or greater than the critical value, it is determined that there is a possibility that an abnormal sound is generated, and "abnormal" is output as the determination result. . When the total degree of abnormality is less than the critical value, it is determined that the probability of occurrence of an abnormal sound is low, and "normal" is output as the determination result 18.

Here, a(b) is the degree of abnormality of the frequency b, and â(b) is the normalized abnormality of the degree of abnormality a(b) of the normalized frequency b of the dispersion δ(b), δ(b) is the frequency b of Dispersion of abnormalities. For the dispersion of the degree of abnormality, for example, the sample of the degree of abnormality is obtained from the complex learning sound data, and the standard deviation of the sample of these abnormalities can be obtained as the dispersion of the degree of abnormality.

As described above, according to the first embodiment, between the learning time and the diagnosis time, the specimen series changes even along the time axis due to changes in conditions such as temperature, and the respective frequencies are independent, in order to make the average difference between the two different. 0, after the correction of the specimen series at the time of diagnosis, the abnormality of each frequency is obtained, and the erroneous determination due to the change of the specimen value which is not necessarily the same in each frequency may be reduced between the learning time and the diagnosis time. Sexual effect.

In addition, in the above description, the target specimen series is corrected based on the average of the difference series between the target specimen series and the reference specimen series. However, according to the average of the difference series between the reference specimen series and the target specimen series, the same is true for the correction of the reference specimen series. Effect.

In addition, based on the average of the difference series between the reference specimen series and the target specimen series, the average of each half of the series of the difference between the reference specimen series and the target specimen series, or the amount according to the predetermined ratio, is corrected to make the difference therebetween. The average is close to 0, and of course the same effect is achieved.

[Second embodiment]

In the present embodiment, if the sound of the inspection target device at the time of diagnosis is normal, the same sample value as the normal reference sample series is obtained, and if the sound of the inspection target device is abnormal at the time of diagnosis, Due to the component generated by the abnormal sound, since the specimen value is increased, the series of the restricted subject specimen is corrected more than the reference specimen series.

Only points different from the first embodiment will be described.

The operations of the different point correction amount calculation unit 601, the correction amount 602, and the correction unit 603 are different. The operation will be described below using the first, third and tenth figures.

The correction amount calculation unit 601 calculates the correction amount 602 based on the reference sample series 506 and the target specimen series 507 for each frequency (step S303 in FIG. 10). The amount of correction for each time n of the frequency b is calculated in the next step "A2-1" to "A2-2".

"A2-1" finds the specimen series D(n, b) of the difference between the target specimen series Y ₁ (n, b) of the frequency b and the reference specimen series Y ₀ (n, b) of the frequency b. (Formula (4-1))

"A2-2" finds the minimum value of the time n of the difference sample series D(n, b), and outputs it as the correction amount H ₂ (b). (Formula (4-2))

The correction unit 603 subtracts the correction amount H ₂ (b) from the target specimen series 507, and obtains the corrected target specimen series 509 (step S304 in Fig. 10). The detail is assumed to be Y ₁₂ (n, b) when the target specimen series is corrected, as in the equation (4-3), Y ₁₂ (n) is calculated by subtracting H ₂ (b) from Y ₁ (n, b) , b).

[8] Y ₁₂ (n, b) = Y ₁ (n, b) - H ₂ (b) ... (4-3)

Fig. 3 is a schematic view showing a sampled series before and after correction by the correction unit 603 of the second embodiment.

(A) shows the relationship between the reference specimen series before correction and the target specimen series. At this time, as the temperature rises, the specimen value of the target specimen series becomes substantially the same as the reference specimen series at the same time, and the specimen value becomes larger in the latter half of the time due to the occurrence of an abnormal sound. The arrow indicates the correction direction generated based on the correction amount H ₂ (b). Near the time of this arrow, the difference between the target specimen series and the reference specimen series becomes the minimum value, and it is corrected in the direction of the arrow.

(B) shows the relationship between the reference specimen series and the corrected target specimen series. The correction is corrected in order to make the minimum value of the difference series between the reference specimen series and the corrected target specimen series close to zero. The results of this correction are understood, but the components of the abnormal sound are added and buried in the reference specimen series, and the relative relationship between the abnormal sound components is maintained for the reference specimen series.

As described above, according to the second embodiment, with respect to each frequency, between the learning time and the diagnosis time, the specimen series also changes along the time axis due to a temperature change or the like, and when the abnormal sound components are unevenly overlapped, two are obtained. In order to make the minimum value close to 0, after correction, between the learning time and the specimen series at the time of diagnosis, the abnormality is obtained by the difference between the two, and between the diagnosis time and the learning time. In each frequency, the possibility of erroneous determination of the change in the specimen value due to temporal changes due to changes in temperature or the like is reduced, and the excessive correction of the abnormal sound component is prevented, and the effect of maintaining or improving the accuracy of the abnormal sound is obtained.

[Third embodiment]

In the present embodiment, the target specimen series is the same as the reference specimen series taken at the normal time. Although only the portion of the abnormal sound component is generated, it can be assumed that the energy rises, but actually, since the operation sound of the inspection target machine is noise, the band energy The instantaneous value keeps swaying. Therefore, the specimen value also has a sway for each measurement, because considering the sway of each measurement of this specimen value, instead of the minimum value, by using the q quantile obtained from the difference distribution between the specimen series, the effect of mitigating the sway is performed. After the correction of the specimen series, the abnormality between the specimen columns is determined.

The operations of the different point correction amount calculation unit 601, the correction amount 602, and the correction unit 603 are different. The operation will be described below using the first, fourth and tenth figures.

The correction amount calculation unit 601 calculates the correction amount 602 based on the reference specimen series 506 and the target specimen series 507 of each frequency band (step S303 in Fig. 10). The detail is the correction amount of each time n of the frequency b, and is calculated by the next step "A3-1" to "A3-2".

"A3-1" finds the specimen series D(n, b) of the difference between the target specimen series Y ₁ (n, b) of the frequency b and the reference specimen series Y ₀ (n, b) of the frequency b. (Formula (5-1))

In "A3-2", the distribution of the difference specimen series D(n, b) is obtained, and the q-quantile of the distribution is obtained, and the output is used as the correction amount H ₃ (b). (Formula (5-2))

The q of the q-quant, for example, can be assumed to be q = 0.25. Further, if q = 0, it is equivalent to the use minimum value as in the second embodiment. When q = 0.5, the central value of the distribution is obtained, and if the shape of the distribution is centered on the average, a substantially average close value is obtained, and the effect can be obtained as in the first embodiment.

[9] D(n,b)=Y ₁ (n,b)-Y ₀ (n,b) (5-1) H ₃ (b)=quantile{D(n,b );q}......(5-2)

Here, the quantitile (quartile) {X; q} displays the calculation of the q-quantile of the distribution of the specimens included in the specimen series X.

The correction unit 603 subtracts the correction amount H ₃ (b) from the target specimen series 507, and obtains the corrected target specimen series 509 (step 304 of Fig. 10). The details are as follows when the object specimen series after the correction is Y ₁₃ (n, b), as in equation (5-3), Y ₁₃ (n is calculated by subtracting H ₃ (b) from Y ₁ (n, b). b).

[Number 10] Y ₁₃ (n, b) = Y ₁ (n, b) - H ₃ (b) ... (5-3)

Fig. 4 is a schematic view showing a series of specimens before and after correction by the correction unit 603 of the third embodiment.

(A) shows the relationship between the reference specimen series before correction and the target specimen series. At this time, as the temperature rises, the specimen value of the target specimen series becomes substantially the same as the reference specimen series at the same time, and the specimen value becomes larger in the latter half of the time due to the occurrence of an abnormal sound. Further, the specimen values C and D of the graph show that a part of the measurement is extremely small due to the sway of each measurement. The difference between the specimen values C and D and the specimen value of the reference specimen series at the corresponding position is the deviation value of the lower side in the distribution, and the result of the q quantile is used instead of the minimum value, and the appropriate correction amount indicated by the arrow is obtained. H ₃ (b).

(B) shows the relationship between the corrected reference specimen series and the target specimen series. The correction makes the q-quantile of the difference series of the corrected reference specimen series and the target specimen series close to zero. I understand the result of this correction, but I am burying the component of the abnormal sound and burying it in the reference specimen series, and correcting the deviation value of the difference series generated by the sway of each measurement, and maintaining the appropriate abnormal sound component for the reference specimen series. Relative relationship.

As described above, according to the third embodiment, each frequency, between the learning time and the diagnosis time, due to the temperature change or the like, the specimen series also changes along the time axis, and the abnormal sound components are unevenly overlapped, according to each time. The measured sway, even if a part of the sample value in the difference series becomes a disengagement value in the distribution, the q-quantile of the distribution of the difference between the two is obtained, and the correction makes the quantile close to 0, at the time of learning and diagnosis. The difference between the specimen series is based on the difference between the two, and between the diagnosis and the learning time, the frequency is reduced by the temperature. In addition to the possibility of erroneous determination of the change in the value of the specimen, such as the change in the degree of time, the excessive correction of the abnormal sound component is prevented, and the direction in which the difference is extremely widened due to the sway of each measurement is not corrected, and the accuracy of the abnormal sound is improved. Effect.

[Fourth embodiment]

In the present embodiment, in order to alleviate the influence of each measurement sway, after smoothing at least one of the specimen series, the difference series is obtained, and the correction amount is determined based on the minimum value of the difference series or the q-quantile of the distribution of the sample values of the difference series.

The operations of the different point correction amount calculation unit 601, the correction amount 602, and the correction unit 603 are different. The operation will be described below using the first, fifth and tenth figures.

The correction amount calculation unit 601 calculates the correction amount 602 based on the reference specimen series 506 and the target specimen series 507 of each frequency band (step S303 in Fig. 10). The details are the correction amounts of the respective times n of the frequency b, which are calculated in the next steps "A4-1" to "A4-5".

"A4-1" smoothes the target specimen series Y ₁ (n, b) of the frequency b, and obtains the smoothed target specimen series Y ₁ ~ (n, b) (formula (6-1)).

In "A4-2", the reference specimen series Y ₀ (n, b) of the frequency b is smoothed, and the smoothed reference specimen series Y ₀ ~ (n, b) (formula (6-2)) is obtained.

"A4-3" finds the specimen type D~(n of the difference between the smoothed target specimen series Y ₁ ~ (n, b) of the frequency b and the smoothed reference specimen series Y ₀ ~ (n, b) of the frequency b , b) (Formula (6-3)).

In "A4-4", the correction amount H ₄ (b) (formula (6-4)) is obtained and output based on the difference specimen type D~(n, b).

In "A4-5", the distribution of the sample D~(n, b) of the difference is obtained, and the q-quantile of the distribution (formula (6-5)) is obtained, and is output as the correction amount H ₅ (b).

The q of the q-quant, for example, can be assumed to be q = 0.25. Further, if q = 0, it is equivalent to the use minimum value as in the second embodiment. When q = 0.5, the central value of the distribution is obtained, and if the shape of the distribution is centered on the average, a substantially average close value is obtained, and the effect can be obtained as in the first embodiment.

In this smooth {X}, the calculation of the average smoothed specimen series X is performed in accordance with the time direction, and quantim{X;q} is a calculation for calculating the q-quantile of the distribution of the specimens contained in the specimen series X. Here, the time window width of the smoothed moving average can be assumed to be, for example, 1 second.

The correction unit 603 subtracts the correction amount H ₄ (b) or the correction amount H ₅ (b) from the target specimen series 507, and obtains the corrected target specimen series 509 (step S304). The details are the corrected object specimen series assumed to be Y ₁₄ (b, n) or Y ₁₅ (b, n), as in equation (6-6) or equation (6-7), by from Y ₁ (n, b) Subtract H ₄ (b) or H ₅ (b) and calculate Y ₁₄ (n, b) or Y ₁₅ (n, b).

[Number 12] Y ₁₄ (n, b) = Y ₁ (n, b) - H ₄ (b) (6-6) Y ₁₅ (n, b) = Y ₁ (n, b )-H ₅ (b)......(6-7)

Fig. 5 is a view showing the complement generated by the correcting unit 603 of the fourth embodiment. The pattern diagram of the specimen series before and after.

<A> shows the relationship between the target specimen series before correction and the reference specimen series. At this time, as the temperature rises, the specimen value of the target specimen series becomes substantially the same as the reference specimen series at the same time, and the specimen value becomes larger in the latter half of the time due to the occurrence of an abnormal sound. However, due to the sway of each measurement, the time is swaying slowly.

<B> shows the relationship between the reference specimen series after smoothing and the target specimen series. Due to the smoothing, the sway of the small embossing is removed, and the change is indicated globally. Here, the arrow displays the correction amount and the correction amount at the time of indicating the minimum value of the difference between the smoothed target specimen series and the reference specimen series.

<C> shows the relationship between the corrected target specimen series and the reference specimen series. Correction makes the difference between the target series and the reference specimen series smoothed after correction (B) close to zero. The result of this correction is understood, except that the component of the abnormal sound is added, and the correction of the difference value of the difference series caused by the swing of each measurement is performed, and the relative relationship of the abnormal sound component is appropriately maintained for the reference specimen series.

As described above, according to the fourth embodiment, each frequency, between the learning time and the diagnosis time, the specimen series also changes along the time axis due to a temperature change or the like, and the abnormal sound components are unevenly overlapped, even each time. The intense sway of the temporal change of the measurement is also corrected by using the minimum value or the q-quantile of the smoothed difference series, and the abnormality is obtained between the learning time and the specimen series at the time of diagnosis. Between the diagnosis time and the learning time, each frequency is reduced by the possibility of erroneous determination of the change in the specimen value due to the change in temperature, etc., while preventing the excessive correction of the abnormal sound component, and not correcting The direction in which the difference is extremely widened in the sway of each measurement has an effect of improving the determination accuracy of the abnormal sound.

[Fifth Embodiment]

In the present embodiment, when the intensity of the sound data is synchronized with the operation of the device to be inspected, a device for acquiring or estimating or learning the operation state of the device to be inspected at each time is provided, and the correction series of the target specimen series or the reference specimen series are set. The operating state of the above-mentioned inspection target machine changes according to each time.

Fig. 6 is a configuration diagram of the present embodiment. In the figure, 701 is a control signal for controlling the operation of the inspection target device, and 703 is an operation state estimation unit that estimates the operation state of the inspection target device in synchronization with the waveform acquisition unit 2, and 702 is a control signal 701 input by the inspection target device. The signal for synchronizing the operation and time of the inspection target device to the time synchronization unit of the waveform acquisition unit 2 and the operation state estimation unit 703 is output, and the machine operation speed of the inspection target medium 502 estimated by the operation state estimation unit 703 is estimated. Inferred value series of inferred action speed series.

Further, the 601 series correction amount calculation unit inputs the reference sample series 506, the target sample series 507, and the estimated operation speed series 704, and outputs the correction amount 602.

The correction amount calculation unit 601 calculates the correction amount 602 based on the reference specimen series 506, the target specimen series 507, and the estimated operation speed series 704 (step S303 in Fig. 10). The details of the correction for each time n of the frequency b are calculated in the next step "A6-1" to "A6-4".

"A6-1" finds the specimen series D(n, b) of the difference between the target specimen series Y ₁ (n, b) of the frequency b and the reference specimen series Y ₀ (n, b) of the frequency b (Expression (7- 1)).

"A6-2" infers the motion speed system by normalizing the maximum value of n In the column V(n), the load coefficient series W(n) (formula (7-2)) is obtained.

In "A6-3", the differential sample series D(n, b) is multiplied by the load coefficient series W(n), and the load difference series DW(n, b) is obtained (formula (7-3)).

In "A6-3", the average of the load difference series DW(n, b) is obtained, and the temporary correction amount H ₆ (b) (formula (7-4)) is obtained.

In "A6-4", the correction amount series H ₇ (n, b) (formula (7-5)) is obtained by multiplying the temporary correction amount H ₆ (b) by the load factor series W(n).

Here, Y ₀ (n, b) is the specimen value of the reference specimen series at time n of the frequency b, and Y ₁ (n, b) is the specimen value of the target specimen series at the time n of the frequency b, V(n) The value of the estimated action speed series of time n, W(n) is the value of the load factor series of time n, H ₆ (b) is the temporary correction amount, and H ₇ (n, b) is the time n of the frequency b The value of the correction series.

The correction unit 603 obtains the corrected target sample series 509 by subtracting the correction amount H ₇ (n, b) from the target specimen series 507 (step S304 in Fig. 10). When the detail is the object specimen series of the corrected object is Y ₁₇ (n, b), as in the equation (7-6), the correction amount series H ₇ (n, b) is subtracted from Y ₁ (n, b), Calculate Y ₁₇ (n, b).

[14] Y ₁₇ (n, b) = Y ₁ (n, b) - H ₇ (n, b) (7-6)

Here, Y ₁₇ (n, b) is a series of target specimens that have been corrected.

Fig. 7 is a schematic view showing a sampled series before and after correction by the correction unit 603.

(A) shows the relationship between the reference specimen series before correction and the target specimen series. Also, the lower part shows the operating speed of the machine. At this time, the specimen value of the target specimen series becomes smaller for the reference specimen series due to the temperature rise. Further, the mode of the reduction is roughly proportional to the speed of the operation, and the interval of T1 and T3 with a small operation speed is small due to the change in temperature. Further, in the section of T2 where the operation speed is the largest, the change in temperature becomes large. The three arrows indicate the average size and correction direction of the correction amount series in each of the sections T1, T2, and T3.

(B) shows the relationship between the corrected reference specimen series and the target specimen series. It is understood that the correction is to make the average of the difference series of the corrected reference specimen series and the target specimen series close to zero.

As described above, according to the fifth embodiment, each frequency is between the learning time and the diagnosis time, and the specimen series obtains the difference between the load of the load by the action of the machine even when the motion of the matching machine changes due to the change in temperature. The average correction amount series generated by the time smoothing is corrected by the correction amount series. The difference between the learning time and the specimen series at the time of diagnosis is determined by the difference between the two, at the time of diagnosis and learning. In addition, each frequency also has an effect of reducing the possibility of erroneous determination of a change in the specimen value due to a change in the temperature of the machine.

[Industry use possibility]

The abnormal sound diagnostic apparatus of the present invention is a machine that uses a large change in conditions, and for example, it is possible to use a detecting device that detects an abnormal state in an elevator.

1‧‧‧concentrator

2‧‧‧ Waveform acquisition department

3‧‧‧ Waveform data

4‧‧‧Time Frequency Analysis Department

5‧‧‧Time frequency distribution

15‧‧‧Abnormality calculation department

16‧‧‧ anomalies

17‧‧‧Decision Department

18‧‧‧Results

501‧‧‧Specialization Department

502‧‧‧ Specimen series

503‧‧‧Memory Department

506‧‧‧ benchmark specimen series

507‧‧‧Object specimen series

509‧‧‧Revised object specimen series

601‧‧‧Revision calculation unit

602‧‧‧ Correction

603‧‧‧Revision Department

Claims (7)

An abnormal sound diagnosis device that acquires a sound generated by a device to be inspected at the time of learning and at the time of diagnosis, compares the sound at the time of learning and diagnosis, and diagnoses a sound abnormality, and is characterized in that the sound data acquisition device and the inspection target device are included. The operation is synchronized, and the sound data generated by the inspection target device is obtained; the analysis device obtains the specimen series composed of the intensity of each time based on the sound data; the memory device, the specimen series during the memory learning, as the reference specimen series; the correction device, According to the correction amount or correction amount series estimated by the target specimen series of the specimen series at the time of diagnosis and the above-mentioned reference specimen series, at least one of the target specimen series or the reference specimen series is corrected; and the determination device compares the corrected target specimen series or The reference specimen series and the corresponding reference specimen series or the target specimen series are calculated, and the calculated abnormality is compared with a predetermined critical value, and the determination abnormality is output; wherein the correction device is configured to obtain the target specimen series and Benchmark specimen series at various times Isobutyl, series of specimens to be corrected so that the two series of samples and the reference average difference close to zero. An abnormal sound diagnosis device that acquires a sound generated by a device to be inspected during learning and at the time of diagnosis, compares a sound at the time of learning and diagnosis, and diagnoses a sound abnormality, and is characterized by: The sound data acquisition device acquires the sound data generated by the inspection target device in synchronization with the operation of the inspection target device, and the analysis device obtains the specimen series composed of the intensity of each time based on the sound data; the memory device stores the specimen at the time of learning The series is used as a reference specimen series; the correction device corrects at least one of the target specimen series or the reference specimen series based on the correction amount or the correction amount series estimated from the target specimen series of the specimen series at the time of diagnosis and the reference specimen series; and the determination device And comparing the corrected target specimen series or the reference specimen series with the corresponding reference specimen series or the target specimen series, calculating the abnormality degree, comparing the calculated abnormality degree with a predetermined critical value, and outputting the determination abnormality degree; wherein the above correction The configuration of the device is to obtain the difference between the target specimen series and the reference specimen series at each time, and to correct the target specimen series and the reference specimen series so that the minimum value of the difference is close to zero. An abnormal sound diagnosis device that acquires a sound generated by a device to be inspected at the time of learning and at the time of diagnosis, compares the sound at the time of learning and diagnosis, and diagnoses a sound abnormality, and is characterized in that the sound data acquisition device and the inspection target device are included. Synchronizing the movements to obtain the sound data generated by the inspection target machine; the analysis device obtains the specimen series composed of the intensity of each time based on the above-mentioned sound data; A memory device, a series of specimens for memory learning, as a reference specimen series; a correction device that corrects a target specimen series or a reference specimen based on a correction amount or a correction amount series estimated from a target specimen series of the specimen series at the time of diagnosis and the above-described reference specimen series. And at least one of the series; and the determination device compares the corrected target specimen series or the reference specimen series with the corresponding reference specimen series or the target specimen series, calculates the abnormality, and compares the calculated abnormality with the predetermined critical value. The determination of the degree of abnormality is as follows: wherein the correction means is configured to obtain a difference between the target specimen series and the reference specimen series at each time, and to correct the q-quantile of the distribution of the difference between the target specimen series and the reference specimen series to be close to zero. An abnormal sound diagnosis device that acquires a sound generated by a device to be inspected at the time of learning and at the time of diagnosis, compares the sound at the time of learning and diagnosis, and diagnoses a sound abnormality, and is characterized in that the sound data acquisition device and the inspection target device are included. The operation is synchronized, and the sound data generated by the inspection target device is obtained; the analysis device obtains the specimen series composed of the intensity of each time based on the sound data; the memory device, the specimen series during the memory learning, as the reference specimen series; the correction device, Correcting at least one of the target specimen series or the reference specimen series based on the correction amount or correction amount series inferred from the target specimen series of the specimen series at the time of diagnosis and the above-mentioned reference specimen series; The determination device compares the corrected target specimen series or the reference specimen series with the corresponding reference specimen series or the target specimen series, calculates an abnormality, and compares the calculated abnormality with a predetermined critical value, and outputs a determination abnormality; The configuration of the correction device is one of the smoothing target specimen series and the reference specimen series, and the difference between the smoothed specimen series and the other specimen series is obtained, and the average or minimum value or the q quantile of the difference between the two is obtained. The object specimen series or the reference specimen series is corrected in the direction close to 0. An abnormal sound diagnosis device that acquires a sound generated by a device to be inspected at the time of learning and at the time of diagnosis, compares the sound at the time of learning and diagnosis, and diagnoses a sound abnormality, and is characterized in that the sound data acquisition device and the inspection target device are included. The operation is synchronized, and the sound data generated by the inspection target device is obtained; the analysis device obtains the specimen series composed of the intensity of each time based on the sound data; the memory device, the specimen series during the memory learning, as the reference specimen series; the correction device, According to the correction amount or correction amount series estimated by the target specimen series of the specimen series at the time of diagnosis and the above-mentioned reference specimen series, at least one of the target specimen series or the reference specimen series is corrected; and the determination device compares the corrected target specimen series or The reference specimen series and the corresponding reference specimen series or target specimen series are calculated, and the calculated abnormality is compared with the predetermined critical value, and the output judgment is different. The configuration of the correction device is configured such that the device that acquires or estimates or learns the operating state of the device is provided at each time when the intensity of the sound data is synchronized with the operation of the device to be inspected, and the device operates according to each time. The state change object specimen series or the correction series of the reference specimen series. An abnormal sound diagnosis device that acquires a sound generated by a device to be inspected at the time of learning and at the time of diagnosis, compares the sound at the time of learning and diagnosis, and diagnoses a sound abnormality, and is characterized in that the sound data acquisition device and the inspection target device are included. The operation is synchronized, and the sound data generated by the inspection target device is obtained; the analysis device obtains the specimen series composed of the intensity of each time based on the sound data; the memory device, the specimen series during the memory learning, as the reference specimen series; the correction device, According to the correction amount or correction amount series estimated by the target specimen series of the specimen series at the time of diagnosis and the above-mentioned reference specimen series, at least one of the target specimen series or the reference specimen series is corrected; and the determination device compares the corrected target specimen series or The reference specimen series and the corresponding reference specimen series or the target specimen series are calculated, and the calculated abnormality is compared with a predetermined critical value, and the determination abnormality is output; wherein the determination device is configured to calculate the abnormality According to the above corrected object This series or The standard specimen series and the corresponding reference specimen series or the target specimen series are used to determine the degree of abnormality, and the degree of abnormality of the dispersion normalization using the abnormality degree is used. The abnormal sound diagnosis device according to any one of the first to fifth aspect, wherein the determination device is configured to calculate a determination degree of abnormality based on the corrected target sample series or the reference sample series and the corresponding For the standard specimen series or the target specimen series, the degree of abnormality is obtained, and the degree of abnormality of the dispersion normalization using the abnormality is used.