JP5930789B2 - Abnormal sound diagnosis device - Google Patents

Description

  The present invention relates to an apparatus that collects sound generated from a device to be inspected and determines the possibility of occurrence of abnormal sound of the device being operated by time-frequency analysis of the collected sound.

  Regarding the abnormal sound diagnosis apparatus, an abnormal sound diagnosis apparatus as shown in Patent Document 1 is known. The abnormal sound diagnosis apparatus disclosed in Patent Document 1 stores sound data of a device to be inspected that has been heard by a person in advance as a normal sound in a storage unit as a reference value, and is stored in advance by the storage unit. The degree of coincidence between the reference value in the device and the measurement data of the sound generated from the inspection target device measured by the measuring unit is calculated by the processing unit, and is calculated by the processing unit. The determination means determines the quality of the abnormal sound of the inspection target device based on the degree of coincidence of the results.

  Further, the processing means for calculating the degree of coincidence 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 receives the amplitude value at each frequency of the reference value as an input, and the same amplitude value The straight line of the proportional expression with the output as the reference is set as the reference set straight line, the deviation from the set straight line of the amplitude value at each frequency of the measurement data is calculated by the least square method, and the result is used as an index of the degree of coincidence. is there.

  Furthermore, the means for digitizing the deviation from the setting line performs correction assuming that the amplitude value at each frequency is on the setting line when the amplitude value at each frequency to be measured falls below the setting line, It is intended to prevent a mismatch in the degree of coincidence with the sound data of the reference value due to zero or low amplitude due to a malfunction of the device.

JP 2005-283227 A

  However, the conventional abnormal sound diagnosis device has a change in the time-frequency characteristics of the sound emitted by the device to be inspected, particularly at a specific time or frequency, due to various conditions such as temperature, atmospheric pressure, humidity, speed, acceleration, pressure, tension, and load. The difference in the time and frequency characteristics of the reference sound data and the time and frequency characteristics of the input sound data at the time of diagnosis are mistakenly regarded as abnormal. There was a problem of detecting.

The abnormal sound diagnosis apparatus according to the present invention is:
An abnormal sound diagnosis device that obtains sound generated from 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 at the time of diagnosis, and diagnoses sound abnormality,
Sound data acquisition means for acquiring sound data generated from the inspection target device in synchronization with the operation of the inspection target device;
From the sound data, for each of a plurality of frequencies, analysis means for obtaining a sample series consisting of the intensity of each time,
Storage means for storing a sample sequence at the time of learning as a reference sample sequence;
A target sample series relating to each of the plurality of frequencies based on a correction amount or a correction amount series independently estimated for each of the plurality of frequencies based on the target sample series that is a sample series at the time of diagnosis and the reference sample series Or correction means for correcting at least one of the reference sample series;
For each of the plurality of frequencies, the degree of abnormality is calculated by comparing the corrected target sample series or reference sample series with the corresponding reference sample series or target sample series, and calculated for each of the plurality of frequencies . And a determination unit that calculates a total abnormality degree from the abnormality degree, compares the total abnormality degree with a predetermined threshold value, and outputs a determination abnormality degree.

  According to the abnormal sound diagnosis apparatus of the present invention, the environmental conditions of the inspection target device, such as temperature, atmospheric pressure, and humidity, and the speed, acceleration, and pressure, which may change the time-frequency characteristics of the sound emitted from the inspection target device. Because it is configured to take into account differences in changes in the time-frequency characteristics of sound data, especially at specific times and frequencies, due to various conditions of the operating conditions of the device under test such as tension, load, etc. There is an effect of reducing the possibility of erroneously detecting a change in a specific time or frequency due to a difference as an abnormality.

It is a block block diagram which shows the abnormal sound diagnostic apparatus in Embodiment 1 of this invention. 6 is a schematic diagram illustrating a sampling sequence before and after correction by the correction unit according to Embodiment 1. FIG. 6 is a schematic diagram illustrating a sampling sequence before and after correction by a correction unit according to Embodiment 2. FIG. FIG. 10 is a schematic diagram illustrating a sampling sequence before and after correction by a correction unit according to the third embodiment. FIG. 10 is a schematic diagram illustrating a sampling sequence before and after correction by a correction unit according to the fourth embodiment. It is a block block diagram which shows the abnormal sound diagnostic apparatus in Embodiment 5 of this invention. FIG. 10 is a schematic diagram illustrating a sampling sequence before and after correction by a correction unit according to the fifth embodiment. It is a flowchart of a process until it judges from the measurement signal acquisition from a sound collector to learning mode or diagnostic mode. It is a flowchart of the process at the time of learning mode. It is a flowchart of the process at the time of diagnostic mode.

Embodiment 1 FIG.
The present embodiment is implemented as software on a personal computer (hereinafter referred to as a PC) as a device for diagnosing abnormal sounds generated by a device to be inspected, and includes a learning mode for capturing a normal waveform and a waveform for a test. Has a diagnostic mode to capture. The measurer installs a sound collector such as a microphone, acoustic sensor, acceleration sensor, etc. on the device to be inspected, and connects the sound collector to the input terminal of the USB (Universal Serial Bus) interface of the PC. Do the hour operation.

  The inspection target device is, for example, a device including a plurality of operating parts such as an elevator. In the case of an elevator, the sound collector is installed in or out of the elevator car, and the signal from the sound collector is taken into the PC placed in the machine room via the control cable, and the car is reciprocated up and down. By driving, the operation sound of each device of the elevator is diagnosed.

  Since the elevator is composed of a plurality of parts, the sound collected by the sound collector is a mixed sound from these parts. Moreover, the frequency characteristics of the sound generated from each component are different. In addition, since the position of the sound collector moves with time, the time frequency component of the time frequency component varies depending on the time. Furthermore, the intensity of sound generated from each component shows different changes for each component due to differences in various conditions such as temperature, atmospheric pressure, humidity, speed, acceleration, pressure, tension, and load. For example, the rail noise generated when the rail and the guide shoe slide slides, and the viscosity of oil applied to the rail decreases as the temperature rises. Therefore, when the temperature rises, the friction decreases and the sound pressure decreases. On the contrary, the sound generated when the rope winds around the sheave does not show much correlation with temperature. In addition, the air-conditioning noise tends to increase in intensity as the fan speed increases with increasing temperature.

  In this way, the time frequency distribution of the sound collected by the sound collector is different in each time and frequency in the time frequency characteristics due to changes in the conditions, so that the equipment is normal. However, due to the difference in the various conditions, the time-frequency characteristic of the sound collected by the sound collector changes for each specific time and frequency, and the change component due to the difference in the various conditions is erroneously determined as a component due to abnormal sound. Is more likely to do.

FIG. 1 is a block configuration diagram showing an abnormal sound diagnosis apparatus according to Embodiment 1 of the present invention.
In FIG. 1, 1 is a sound collector such as a microphone, an acoustic sensor, or a vibration sensor, 2 is a waveform acquisition unit that samples a signal from the sound collector 1 and converts it into a digital signal and outputs waveform data 3, and 4 is a waveform. A time frequency distribution 5 comprising spectral values indicating time and intensity with respect to time and frequency is obtained by performing time-frequency analysis on the waveform data 3 by fast Fourier transform (hereinafter referred to as FFT) while shifting the time window in the time direction by multiplying the data 3 with a time window. A time frequency analysis unit 501 for outputting a sampling unit for outputting a sample sequence 502 of each frequency which is a time series obtained by sampling a spectrum value indicating the intensity of each predetermined frequency from the time frequency distribution 5 at each time. is there. The sample series 502 is a time series composed of sample values at each time for predetermined frequencies.
In the following description of the embodiment, a case will be described in which the time frequency distribution 5 output from the time frequency analysis unit 4 is sound data. The sound data may be waveform data or other feature quantities obtained by analyzing the waveform data.

  Reference numeral 503 denotes a storage unit that stores a sample series 502 obtained from sound data acquired at the time of learning. Reference numeral 506 denotes a reference that is stored in the storage unit 503 at the time of learning and includes a sample series of each frequency that serves as a reference when calculating the degree of abnormality. A sample sequence 507 is obtained from sound data acquired at the time of diagnosis, and is a target sample sequence including a sample sequence of each frequency that is a target when calculating the degree of abnormality.

  Reference numeral 601 refers to the reference sample series 506 and the target sample series 507, and a correction amount calculation unit that calculates the correction amount 602. Reference numeral 603 corrects the target sample series 507 based on the correction amount 602, and the corrected target sample series 509 is corrected. Is a correction unit that outputs.

  15 is an abnormality degree calculation unit that calculates the degree of abnormality indicating the degree of possibility of abnormal sound generation and outputs the degree of abnormality 16 with reference to the reference sample series 506 and the corrected target sample series 509 for each predetermined frequency. , 17 is a determination unit that determines the possibility of occurrence of abnormal sound based on the degree of abnormality 16 for each predetermined frequency and outputs a determination result 18.

The operation will be described below with reference to the processing flowcharts of FIGS.
In the learning mode or the diagnostic mode, the waveform acquisition unit 2 acquires a measurement signal output from the sound collector 1, amplifies and AD-converts the sampled signal to obtain a 16-bit linear PCM (pulse code modulation) having a sampling frequency of 48 kHz. The measurement signal is converted into the waveform data 3 of the digital signal (step S1 in FIG. 8).
The time frequency analysis unit 4 extracts a frame from the waveform data 3 output by the waveform acquisition unit 2 while shifting the time window of 1024 points in the time direction at intervals of 16 milliseconds, and performs an FFT operation on each frame. A frequency spectrum series y (t, f) is obtained and output as a time-frequency distribution 5 (step S2 in FIG. 8). Here, t is an index of time corresponding to the shift interval for shifting the analysis window, and f is an index indicating the frequency of the result of the FFT operation. The time t and the frequency f satisfy the relations 0 ≦ t ≦ T and 0 ≦ f ≦ F, respectively. Here, T is the number of frames in the time direction of the time frequency distribution 5, and F is an index indicating a frequency corresponding to the Nyquist frequency that is ½ of the sampling frequency fs of the waveform data 3 (F = fs / 2).

  From the time-frequency distribution 5, the sampling unit 501 applies these five frequency bands each having a band of one octave width with 0.5 kHz, 1 kHz, 2 kHz, 4 kHz, and 8 kHz as the center frequencies. The sum of the spectral values is obtained with 8 frames as a unit of frequency components included in one frequency band, the sample values for each predetermined time with 8 frames (256 ms) as a unit are obtained, and a sample sequence 502 is output (FIG. 8 step S3). Now, assuming that the sample value in the sample sequence 502 at each frequency and each time is Y (n, b), Y (n, b) is calculated by the equation (1-1).

  Here, n is a time index of a sampled sequence, a natural number ranging from 1 to N (where N is the upper limit of the time range, N = T / 8 and the remainder is rounded down), and b is a frequency index. Is a natural number in the range of -B (B is the number of frequency bands, and B = 5 in the present embodiment). Further, Ω (n, b) represents a set of time and frequency sets (t, f) to be summed for sampling in the time-frequency distribution y (t, f).

  When the sample series 502 is acquired by the sampling unit 501, the abnormal sound diagnosis apparatus determines whether it is in the learning mode or the diagnosis mode (step S4 in FIG. 8).

  In the learning mode, the storage unit 503 stores the sample series 502 as the reference sample series 506 for each frequency (step S201 in FIG. 9) (step 202 in FIG. 9).

Next, the operation of the diagnosis process in the diagnosis mode will be described.
In the diagnosis mode, for each frequency (step S301 in FIG. 10), the sample series 502 is set as a diagnosis target specimen series 507 (step S302 in FIG. 10).

The correction amount calculation unit 601 calculates a correction amount 602 for each frequency from the reference sample series 506 stored in the storage unit 503 and the target sample series 507 sampled by the sampling unit 501 (FIG. 10). Step S303). Specifically, the correction amount for each time n of each frequency b is calculated by the following procedures << A1-1 >> to << A1-2 >>.
<< A1-1 >> A sample sequence D (n, b) of a difference between the target sample sequence Y ₁ (n, b) at the frequency b and the reference sample sequence Y ₀ (n, b) at the frequency b is obtained. (Formula (2-1))
<< A1-2 >> The average of the difference sample series D (n, b) with respect to time n is obtained and output as the correction amount H ₁ (b). (Formula (2-2))

Equation (2-2) calculates the average of the difference series. However, since the right side of equation (2-2) can be transformed like the right side of equation (2-3), the correction amount H ₁ (b) Therefore, after obtaining the average of the respective sample series, the difference between the respective averages may be obtained to calculate the correction amount H ₁ (b).

The correction unit 603 obtains the corrected target sample series 509 by subtracting the correction amount H ₁ (b) from the target sample series 507 (step S304 in FIG. 10). Specifically, if the target sample series after the correction of the frequency b is Y ₁₁ (b, n), Y ₁₁ (b, n) is the target sample series Y of the frequency b as shown in Expression (2-4). It is calculated by subtracting the correction amount H ₁ (b) from ₁ (n, b).

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

FIG. 2 is a schematic diagram illustrating a sampling sequence before and after correction by the correction unit 603.
(A) shows the relationship between the reference sample series before correction and the target sample series. In this case, since the temperature has risen, the sample values of the target sample series are almost uniformly smaller than the reference sample series over the entire time. The arrow indicates the correction direction according to the correction amount.
(B) shows the relationship between the corrected reference sample series and the target sample series. It can be seen that after the correction, the average of the difference series between the reference sample series and the target sample series is corrected so as to approach zero.

The abnormality degree calculation 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 difference as the degree of abnormality 16 (step S305 in FIG. 10). Specifically, for each frequency b, the degree of abnormality of the difference between the reference sample series Y ₀ (n, b) and the corrected target sample series Y ₁₁ (n, b) is expressed as the square of the difference between the sample series. It is calculated as an average value (equivalent to the Euclidean distance when the sample series is regarded as a vector in the N-dimensional space) (formula (3-1)).

  Here, in the above equation, a (b) is the degree of abnormality of the frequency b. Further, the range of n in the sum is the total time (1 ≦ n ≦ N).

  The determination unit 17 compares the total abnormality degree calculated from the abnormality degree 16 of each frequency output by the abnormality degree calculation part 15 with a predetermined threshold value to determine whether or not there is a possibility that an abnormal sound is generated. Determination is made and output as the determination result 18 (step S307 in FIG. 10). Specifically, first, a normalized abnormality degree a ^ (b) obtained by normalizing the abnormality degree a (b) of each frequency by its variance is obtained for each frequency (formula (3-2)). Next, the maximum value regarding the frequency of the normalized abnormality degree is set as the total abnormality degree a * (formula (3-3)). Finally, the total abnormal degree a * calculated by the equation (3-3) is compared with a threshold value, and when the total abnormal degree is equal to or higher than the threshold value, it is determined that there is a possibility that an abnormal sound is generated. “Abnormal” is output as the determination result 18. When the total abnormality level is less than the threshold value, it is determined that there is a low possibility that abnormal noise has occurred, and “normal” is output as the determination result 18.

  Here, a (b) is the abnormality degree of the frequency b, a ^ (b) is the normalized abnormality degree obtained by normalizing the abnormality degree a (b) of the frequency b by the variance δ (b), and δ (b) is The variance of the degree of anomaly at frequency b. For example, the variance of the degree of abnormality can be obtained by obtaining a sample of the degree of abnormality from a plurality of learning sound data, and obtaining the standard deviation of the samples of the degree of abnormality as the variance of the degree of abnormality.

  As described above, according to the first embodiment, even when the sample sequence changes uniformly along the time axis due to changes in conditions such as temperature between learning and diagnosis, for each frequency, Independently, after correcting the sample series at the time of diagnosis so that the average of the difference between the two approaches 0, the degree of abnormality of each frequency is obtained, so between the time of diagnosis and the time of learning, This has the effect of reducing the possibility of erroneous determination due to changes in sample values that are not necessarily the same at each frequency.

In the above description, the target sample series is corrected based on the average of the difference series between the target sample series and the reference sample series, but based on the average of the difference series between the reference sample series and the target sample series, It goes without saying that the same effect can be obtained even if the reference sample series side is corrected.
Furthermore, based on the average of the difference series of the reference sample series and the target sample series, each of the reference sample series and the target sample series is divided by half the average of the difference series or by an amount based on a predetermined ratio. It goes without saying that the same effect can be obtained even if the average of the differences is corrected to approach zero.

Embodiment 2. FIG.
In this embodiment, if the sound of the inspection target device at the time of diagnosis is normal, the target sample sequence takes the same sample value as the reference sample sequence at the time of normal, while the sound of the inspection target device at the time of diagnosis is abnormal. If there is, the sample value is considered to increase due to the component due to the abnormal sound, and therefore, the correction is performed with a constraint that the target sample series is higher than the reference sample series.

Only differences from the first embodiment will be described.
The difference is that the operations of the correction amount calculation unit 601, the correction amount 602, and the correction unit 603 are different.
Hereinafter, the operation will be described with reference to FIG. 1, FIG. 3, and FIG.

The correction amount calculation unit 601 calculates the correction amount 602 for each frequency from the reference sample series 506 and the target sample series 507 (step S303 in FIG. 10). Specifically, the correction amount for each time n of the frequency b is calculated by the following procedures << A2-1 >> to << A2-2 >>.
<< A2-1 >> A sample sequence D (n, b) of a difference between the target sample sequence Y ₁ (n, b) at the frequency b and the reference sample sequence Y ₀ (n, b) at the frequency b is obtained. (Formula (4-1))
<< A2-2 >> The minimum value for the time n of the difference sample series D (n, b) is obtained and output as the correction amount H ₂ (b). (Formula (4-2))

The correction unit 603 obtains the corrected target sample series 509 by subtracting the correction amount H ₂ (b) from the target sample series 507 (step S304 in FIG. 10). Specifically, if the target sample series after correction is Y ₁₂ (b, n), Y ₁₂ (b, n) is calculated from Y ₁ (b, n) to H ₂ as shown in Equation (4-3). Calculated by subtracting (b).

FIG. 3 is a schematic diagram illustrating sampling sequences before and after correction by the correction unit 603 according to the second embodiment.
(A) shows the relationship between the reference sample series before correction and the target sample series. In this case, since the temperature has increased, the sample value of the target sample sequence becomes almost uniformly smaller than the reference sample sequence over the entire time, and abnormal sound is generated in the latter half of the time, so the sample value is It is getting bigger. The arrow indicates the direction of correction by the correction amount H ₂ (b). Near the time indicated by the arrow, the difference between the sample values of the target sample series and the reference sample series is a minimum value, so that correction is performed in the direction of the arrow.
(B) shows the relationship between the reference sample series and the corrected target sample series. The correction is performed so that the minimum value of the difference series between the reference sample series and the corrected target sample series approaches zero. As a result of this correction, it can be seen that the relative relationship of the abnormal sound component is maintained with respect to the reference sample sequence without excessively correcting the abnormal sound component and being buried in the reference sample sequence.

  As described above, according to the second embodiment, for each frequency, the sample sequence changes uniformly along the time axis due to a change in temperature between the time of learning and the time of diagnosis. Even when non-uniform superimposition is performed, the minimum value of the difference between the two is obtained, corrected so that the minimum value approaches 0, and the degree of abnormality is determined from the difference between the two at the time of learning and the sample series at the time of diagnosis. Therefore, it is possible to reduce the possibility of misjudgment caused by changes in the sample value over time due to temperature changes, etc. This has the effect of preventing excessive correction of sound components and maintaining or improving the accuracy of abnormal sound determination.

Embodiment 3 FIG.
In this embodiment, it can be assumed that the target sample series is the same as the reference sample series collected at normal time, or that the power increases by the amount of abnormal sound components. Therefore, the instantaneous value of the band power is constantly fluctuating. For this reason, the sample values also have fluctuations at each measurement time, and in order to consider fluctuations at each measurement time of the sample values, instead of the minimum value, the q quantile obtained from the distribution of the differences between the sample sequences. By using the number, after correcting the sample sequence with constraints that relaxes the influence of fluctuation, the abnormality between the sample sequences is determined.

The difference is that the operations of the correction amount calculation unit 601, the correction amount 602, and the correction unit 603 are different.
Hereinafter, the operation will be described with reference to FIG. 1, FIG. 4, and FIG.

The correction amount calculation unit 601 calculates the correction amount 602 from the reference sample series 506 and the target sample series 507 in each frequency band (step S303 in FIG. 10). Specifically, the correction amount for each time n of the frequency b is calculated by the following procedures << A3-1 >> to << A3-2 >>.
<< A3-1 >> A sample sequence D (n, b) of a difference between the target sample sequence Y ₁ (n, b) at the frequency b and the reference sample sequence Y ₀ (n, b) at the frequency b is obtained. (Formula (5-1))
<< A3-2 >> The distribution of the difference sample series D (n, b) is obtained, the q quantile of the distribution is obtained, and output as the correction amount H ₃ (b). (Formula (5-2))
The q quantile q can be, for example, q = 0.25. If q = 0, it corresponds to using the same minimum value as in the second embodiment. If q = 0.5, it becomes the median value of the distribution, and if the distribution shape is symmetrical about the average, a value close to the average is obtained, and the same effect as in the first embodiment can be obtained. .

  Here, quantile {X; q} represents an operation for obtaining the q quantile of the distribution formed by the samples included in the sample series X.

The correction unit 603 obtains the corrected target sample series 509 by subtracting the correction amount H ₃ (b) from the target sample series 507 (step S304 in FIG. 10). Specifically, if the target sample series after correction is Y ₁₃ (b, n), Y ₁₃ (b, n) is calculated from Y ₁ (b, n) to H ₃ as shown in Equation (5-3). Calculated by subtracting (b).

FIG. 4 is a schematic diagram illustrating sampling sequences before and after correction by the correction unit 603 according to the third embodiment.
(A) shows the relationship between the reference sample series before correction and the target sample series. In this case, since the temperature has increased, the sample value of the target sample sequence becomes almost uniformly smaller than the reference sample sequence over the entire time, and abnormal sound is generated in the latter half of the time, so the sample value is It is getting bigger. Also, the sample values C and D in the figure indicate that some of the sample values C and D are extremely small due to fluctuations at each measurement. Since the difference between the sample values C and D and the sample value of the reference sample series at the corresponding position becomes the lower outlier in the distribution, the result of using the q quantile instead of the minimum value is The appropriate correction amount H ₃ (b) shown is obtained.
(B) shows the relationship between the corrected reference sample series and the target sample series. After the correction, the q quantile of the difference series between the reference sample series and the target sample series is corrected so as to approach zero. As a result of this correction, the abnormal sound component is not excessively corrected and buried in the reference sample series, and correction is performed to exclude the outliers of the difference series due to fluctuations at each measurement. It can be seen that the relative relationship of the components is properly maintained.

  As described above, according to the third embodiment, for each frequency, the sample series changes uniformly along the time axis due to a change in temperature between the time of learning and the time of diagnosis. In addition to non-uniform superposition, even if some sample values in the difference series become outliers in the distribution due to fluctuations at each measurement, the q quantile of the difference distribution between the two is obtained. After the correction is made so that the order approaches 0, the degree of abnormality is obtained from the difference between the sample series at the time of learning and at the time of diagnosis. In addition, it reduces the possibility of misjudgment caused by temporally uniform sample value changes due to temperature changes, etc., and prevents excessive correction of abnormal sound components. So that the difference is not corrected in an increasing direction, There is an effect of improving.

Embodiment 4 FIG.
In this embodiment, in order to alleviate the influence of fluctuation for each measurement, at least one of the sample series is smoothed, a difference series is obtained, and the minimum value of the difference series or the q of the distribution of the sample values of the difference series is obtained. The correction amount is determined by the quantile.

The difference is that the operations of the correction amount calculation unit 601, the correction amount 602, and the correction unit 603 are different.
Hereinafter, the operation will be described with reference to FIG. 1, FIG. 5, and FIG.

The correction amount calculation unit 601 calculates the correction amount 602 from the reference sample series 506 and the target sample series 507 for each frequency (step S303 in FIG. 10). Specifically, the correction amount for each time n of the frequency b is calculated by the following procedures << A4-1 >> to << A4-5 >>.
"A4-1" frequency b of the target sample sequence Y ₁ (n, b) the seek smoothed by smoothing the target specimen sequence Y ₁ ~ (n, b) (Formula (6-1))).
"A4-2" obtaining a reference sample sequence Y ₀ (n, b) of the frequency b of the smoothed reference sample sequence Y ₀ is smoothed ~ (n, b) (Formula (6-2))).
<< A4-3 >> The sample series D to (D) of the difference between the target sample series Y ₁ to (n, b) smoothed at the frequency b and the reference sample series Y ₀ to (n, b) smoothed to the frequency b n, b) is obtained (formula (6-3)).
<< A4-4 >> The correction amount H ₄ (b) is obtained from the difference sample series D to (n, b) (formula (6-4)) and output.
<< A4-5 >> The distribution of the difference sample series D to (n, b) is obtained, the q quantile of the distribution is obtained (formula (6-5)), and output as the correction amount H ₅ (b).
The q quantile q can be, for example, q = 0.25. If q = 0, it corresponds to using the same minimum value as in the second embodiment. If q = 0.5, it becomes the median value of the distribution, and if the distribution shape is symmetrical about the average, a value close to the average is obtained, and the same effect as in the first embodiment can be obtained. .

  Here, smooth {X} is an operation for smoothing the sample sequence X by moving average in the time direction, and quantile {X; q} is a q quantile of the distribution formed by the samples included in the sample sequence X. Represents an operation. Here, the width of the smoothing moving average time window can be set to 1 second, for example.

The correction unit 603 obtains the corrected target sample series 509 by subtracting the correction amount H ₄ (b) or the correction amount H ₅ (b) from the target sample series 507 (step S304). Specifically, if the target sample series after correction is Y ₁₄ (b, n) or Y ₁₅ (b, n), Y ₁₄ (n, b) or Y ₁₅ (n, b) is expressed by the equation (6- It is calculated by subtracting H ₄ (b) or H ₅ (b) from Y ₁ (n, b) as in 6) or equation (6-7).

FIG. 5 is a schematic diagram illustrating sampling sequences before and after correction by the correction unit 603 according to the fourth embodiment.
(A) shows the relationship between the target sample series before correction and the reference sample series. In this case, since the temperature has increased, the sample value of the target sample sequence becomes almost uniformly smaller than the reference sample sequence over the entire time, and abnormal sound is generated in the latter half of the time, so the sample value is It is getting bigger. However, it fluctuates constantly in time due to fluctuations at every measurement.
(B) shows the relationship between the reference sample series and the target sample series after smoothing both. By smoothing, every small fluctuation is removed and a global change in time is represented. Here, the arrows indicate the correction amount and the magnitude of the correction at the time indicating the minimum value of the difference sequence between the target sample sequence and the reference sample sequence after smoothing.
(C) shows the relationship between the corrected target sample series and the reference sample series. After the correction, the minimum value of the difference series between the target sample series and the reference sample series after smoothing in (B) is corrected so as to approach zero. As a result of this correction, the abnormal sound component is not excessively corrected, and the correction is performed by removing the outlier of the difference series due to fluctuations at each measurement, and the relative component of the abnormal sound is relative to the reference sample series. It can be seen that the target relationship is properly maintained.

  As described above, according to the fourth embodiment, for each frequency, the sample series changes uniformly along the time axis due to a change in temperature between the time of learning and the time of diagnosis. In addition to non-uniform superposition, even if there are fluctuations that vary greatly with time for each measurement, after correction using the minimum value of the difference series after smoothing or the q quantile, correction is performed at the time of learning and diagnosis. Because it is configured so that the degree of abnormality is obtained from the difference between the two sample sequences, it is caused by a change in the sample value over time due to a temperature change etc. for each frequency between diagnosis and learning. In addition to reducing the possibility of misjudgment, the excessive noise component is prevented from being corrected excessively, and fluctuations at each measurement are prevented from being corrected in the direction in which the difference is greatly expanded, thereby improving the accuracy of abnormal sound determination. There is an effect of improving.

Embodiment 5 FIG.
The present embodiment provides means for acquiring, estimating or learning the operating state of the inspection target device at each time when the intensity of sound data is sampled in synchronization with the operation of the inspection target device. The correction amount series of the reference sample series is changed according to the operating state of the inspection target device at 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, 703 is an operation state estimation unit for estimating the operation state of the inspection target device in synchronization with the waveform acquisition unit 2, and 702 is a control input from the inspection target device. A time synchronization unit that processes the signal 701 and outputs a signal for synchronizing the operation and time of the inspection target device to the waveform acquisition unit 2 and the operation state estimation unit 703, and 704 is a sample series estimated by the operation state estimation unit 703 This is an estimated operation speed sequence which is a sequence of estimated values of the operation speed of the inspection target device at each time of 502.
Reference numeral 601 denotes a correction amount calculation unit that inputs the reference sample sequence 506, the target sample sequence 507, and the estimated operation speed sequence 704, and outputs a correction amount sequence 602.

The correction amount calculation unit 601 calculates the correction amount sequence 602 from the reference sample sequence 506, the target sample sequence 507, and the estimated operation speed sequence 704 (step S303 in FIG. 10). Specifically, the correction amount for each time n of the frequency b is calculated by the following procedures << A6-1 >> to << A6-4 >>.
<< A6-1 >> A sample sequence D (n, b) of a difference between the target sample sequence Y ₁ (n, b) at the frequency b and the reference sample sequence Y ₀ (n, b) at the frequency b is obtained (formula (7− 1)).
<< A6-2 >> A load coefficient series W (n) is obtained by normalizing the estimated operation speed series V (n) with the maximum value for n (formula (7-2)).
<< A6-3 >> The difference sample series D (n, b) is multiplied by the load coefficient series W (n) to obtain the load difference series DW (n, b) (formula (7-3)).
<< A6-3 >> An average of the load difference series DW (n, b) is obtained to obtain a provisional correction amount H ₆ (b) (formula (7-4)).
<< A6-4 >> The correction amount series H ₇ (n, b) is obtained by multiplying the temporary correction amount H ₆ (b) by the load coefficient series W (n) (formula (7-5)).

Where Y ₀ (n, b) is the sample value of the reference sample sequence at time n of frequency b, Y ₁ (n, b) is the sample value of the target sample sequence at time n of frequency b, and V (n) is The value of the estimated operating speed series at time n, W (n) is the value of the load coefficient series at time n, H ₆ (b) is a temporary correction amount, and H ₇ (n, b) is the correction of time n at frequency b. The value of the quantity series.

The correction unit 603 obtains the corrected target sample series 509 by subtracting the correction amount series H ₇ (n, b) from the target sample series 507 (step S304 in FIG. 10). Specifically, when the corrected target sample series is Y ₁₇ (n, b), Y ₁₇ (n, b) is corrected from Y ₁ (n, b) as shown in Equation (7-6). Calculated by subtracting the quantity sequence H ₇ (n, b).

Here, Y ₁₇ (n, b) is the target sample series after correction.

FIG. 7 is a schematic diagram showing a sample series before and after correction by the correction unit 603.
(A) shows the relationship between the reference sample series before correction and the target sample series. The operating speed of the equipment is shown at the bottom. In this case, since the temperature has risen, the sample value of the target sample series is smaller than the reference sample series over the entire time. Moreover, the way of decreasing is in a relation substantially proportional to the operating speed, and the change due to the temperature is small in the interval between T1 and T3 where the operating speed is low. Further, in the section of T2 where the operation speed is maximum, the change due to temperature is large. 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 sample series and the target sample series. It can be seen that after the correction, the average of the difference series between the reference sample series and the target sample series is corrected so as to approach zero.

  As described above, according to the fifth embodiment, even when the sample series is changed in accordance with the operation of the device due to a change in temperature between the time of learning and the time of diagnosis for each frequency, the load is applied by the operation of the device. After obtaining an average correction amount series by time smoothing of the load difference between the two and performing correction by this correction amount series, the degree of abnormality is obtained from the difference between the sample series at the time of learning and at the time of diagnosis. With this configuration, there is an effect of reducing the possibility of misjudgment due to a change in sample value depending on the operation of the device due to a change in temperature for each frequency between diagnosis and learning.

  The abnormal sound diagnosis apparatus of the present invention may be used as a detection apparatus that detects an abnormal state in a device whose use conditions are greatly changed, for example, an elevator.

  DESCRIPTION OF SYMBOLS 1; Sound collector, 2; Waveform acquisition part, 3; Waveform data, 4; Time frequency analysis part, 5; Time frequency distribution, 15: Abnormality calculation part, 16; Abnormality degree, 17; Result, 501; Sampling unit, 502; Sample series, 503; Storage unit, 506; Reference sample series, 507; Target sample series, 509; Target sample series after correction, 601; Correction amount calculation unit, 602; 603; Correction unit 701; Control signal 703; Operating state estimation unit 702; Time synchronization unit 704;

Claims (7)

An abnormal sound diagnosis device that obtains sound generated from 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 at the time of diagnosis, and diagnoses sound abnormality,
Sound data acquisition means for acquiring sound data generated from the inspection target device in synchronization with the operation of the inspection target device;
From the sound data, for each of a plurality of frequencies, analysis means for obtaining a sample series consisting of the intensity of each time,
Storage means for storing a sample sequence at the time of learning as a reference sample sequence for each of the plurality of frequencies ,
A target sample series relating to each of the plurality of frequencies based on a correction amount or a correction amount series independently estimated for each of the plurality of frequencies based on the target sample series that is a sample series at the time of diagnosis and the reference sample series Or correction means for correcting at least one of the reference sample series;
For each of the plurality of frequencies, the degree of abnormality is calculated by comparing the corrected target sample series or reference sample series with the corresponding reference sample series or target sample series, and calculated for each of the plurality of frequencies . An abnormal sound diagnosis apparatus comprising: a determination unit that calculates a total abnormality degree from an abnormality degree, compares the total abnormality degree with a predetermined threshold value, and outputs a determination abnormality degree. The correction means is
The difference between the target sample sequence and the reference sample sequence at each time is obtained independently for each of the plurality of frequencies, and the target sample sequence or the reference sample sequence is corrected so that the average of the difference between the two approaches zero. The abnormal sound diagnosis apparatus according to claim 1. The correction means is
The difference between the target sample series and the reference sample series at each time is obtained independently for each of the plurality of frequencies, and the target sample series or the reference sample series is corrected so that the minimum value of the difference between the two approaches zero. The abnormal sound diagnosis apparatus according to claim 1. The correction means is
The difference between the target sample series and the reference sample series at each time is obtained independently for each of the plurality of frequencies, and the target sample series or the reference sample series is corrected so that the q quantile of the difference distribution approaches zero. The abnormal sound diagnosis apparatus according to claim 1, wherein the abnormal sound diagnosis apparatus is configured. The correction means is
For each of the plurality of frequencies, either the target sample series or the reference sample series is smoothed, the difference between the smoothed sample series and the other sample series is obtained, and the average or minimum value of the difference between the two is obtained. The abnormal sound diagnosis apparatus according to claim 1, wherein the target sample series or the reference sample series is corrected in a direction in which the q quantile approaches zero. The correction means is
Means are provided for acquiring, estimating or learning the operating state of the device at each time when the intensity of sound data is sampled in synchronization with the operation of the device to be inspected , and independently for each of the plurality of frequencies The abnormal sound diagnosis apparatus according to claim 1, wherein the correction amount series of the series or the reference sample series is configured to change according to the operating state of the device at each time. The determination means is
Normalization independent degree of abnormality of variance for each of the plurality of frequency target specimen series or determined from the reference sample-sequence or target specimen sequences corresponding to the reference sample sequence after the correction when calculating the determination abnormality degree The abnormal sound diagnosis apparatus according to claim 1 , wherein the measured value is used as the total abnormality degree .

Images