WO2014016914A1

WO2014016914A1 - Abnormal noise detection system

Info

Publication number: WO2014016914A1
Application number: PCT/JP2012/068755
Authority: WO
Inventors: 今朝明峰村; 晋也湯田; 崇佐伯
Original assignee: 株式会社日立製作所
Priority date: 2012-07-25
Filing date: 2012-07-25
Publication date: 2014-01-30
Also published as: GB2520628A; CN104380063B; GB2520628B; GB201419256D0; JP5948418B2; CN104380063A; JPWO2014016914A1

Abstract

The purpose of the present invention is to detect, among abnormal noises emitted from a device, abnormal noises in which changes in amplitude intensity (sound pressure) do not occur or are negligible, and abnormal noises in which a phase change occurs. This abnormal noise detection system, which diagnoses abnormalities in a device being diagnosed using sound data extracted from the device as an input signal, comprises: a zero point magnitude calculation unit that calculates zero point magnitudes of the sound data for one cycle of the input signal; a zero point density distribution calculation unit that calculates a zero point density distribution from the zero point magnitudes; and a normal/abnormal diagnosis unit that determines whether the device being diagnosed is normal or abnormal by comparing the zero point density distribution of the input signal with the zero point density distribution of normal sound data.

Description

Abnormal sound detection system

The present invention relates to an abnormal sound detection system.

As a background technology in this technical field, device diagnosis is important for maintaining the operation rate of the device. At the time of diagnosis of a device, as a countermeasure when it is difficult to attach a sensor to a diagnosis target part, acoustic diagnosis that does not require contact with the diagnosis target part has attracted attention. Therefore, a diagnosis using sound has been conventionally used.

Patent Document 1 discloses an “acoustic diagnosis support apparatus that supports diagnosis of a sound source that generates abnormal noise at a predetermined period when an abnormality occurs”, and “makes it easy to grasp the abnormality”. It is described. For this purpose, “the sound collector records measurement data obtained by sampling the sound generated from the rotating machine at a predetermined frequency for a predetermined time longer than the rotation period. The diagnosis support apparatus acquires measurement data from the sound collector. The support apparatus extracts a reference data sequence for one cycle from the beginning of the first measurement data, extracts a comparison data sequence for one cycle while shifting the extraction position in the nth measurement data, The correlation degree of the comparison data series is calculated, the second measurement data is shifted using the extraction position with the largest correlation degree as the shift amount, and then added to the first measurement data ".

Further, Patent Document 2 discloses that “a method for diagnosing an abnormality occurring in the diagnostic object based on an acoustic signal generated from the diagnostic object, and the acoustic signal when the diagnostic object is in a normal state”. Based on the relationship between the sound pressure and the frequency or frequency band obtained by frequency analysis of the sound pressure of each of the frequencies or the frequency band and one or more other frequencies around the frequency or frequency band “Abnormality diagnosis method for obtaining a relative difference from a sound pressure corresponding to a frequency band as a relative sound pressure difference”.

Further, Patent Document 3 relates to a signal detection method for searching for and detecting a predetermined signal from a stored signal series or a signal similar to a part thereof, and can be applied to, for example, acoustic signal detection. Conventionally, with regard to the signal detection method, there is known a signal search method for the purpose of detecting a location similar to the target signal in the accumulated signal, but because only local pruning is used. However, when a large amount of accumulated signals are targeted, there is a disadvantage that it takes a long time to search.Based on the L1 distance, global grouping and local grouping are performed to make the search space efficient. By narrowing down to, there is an advantage that effective partial signal detection can be performed at high speed while maintaining the search system. "

As described above, the abnormal sound detection method, the feature amount, and the search for similar information emitted from the device have been worked on from the past.

JP 2012-93094 A JP 2005-257460 A International Publication No. 2006/009035 Pamphlet

However, in the conventional method, it is assumed that a change in abnormal sound appears in the amplitude intensity (sound pressure). Therefore, when no abnormality occurs in the amplitude intensity (sound pressure) and it appears as a phase change, for example, an abnormality with a small amplitude change such as an air inlet cannot be detected.

Patent Document 1 is an invention that, when an abnormality occurs, acquires a plurality of periodic data of acoustic data, synchronizes them, and then superimposes them to grasp the abnormality of the sound source that occurs periodically. For this reason, it is impossible to grasp an abnormality of a sound source that does not occur periodically, for example, an abnormal sound in which a phase change occurs. For this reason, it is impossible to detect an abnormality such as a small amplitude change, an air inlet clogging, or an initial abnormality.

Patent Document 2 defines an abnormality determination level difference upper limit threshold and an abnormality determination level difference lower limit threshold based on the relationship between the sound pressure and the frequency or frequency band that can be obtained by frequency analysis of an acoustic signal, and diagnoses In some cases, diagnosis is performed using a frequency component or its sound pressure. Therefore, it is not possible to detect an abnormality in which abnormal sound does not appear in the frequency band and sound pressure value (size).

Patent Document 3 is an invention for performing a search by making a histogram of feature amounts in order to perform a high-speed search using the L1 distance. Also, the color of the video is used as a feature amount. The L1 distance is defined as a distance based on a difference in distance based on the first power.

As described above, there is no description in the prior art literature regarding a method for detecting an abnormal sound in which there is no change in the frequency band and amplitude intensity (sound pressure) and a phase change occurs.

An object of the present invention is to detect an abnormal sound in which an amplitude intensity (sound pressure) does not change or is small among abnormal sounds emitted from a device and a phase change occurs.

Therefore, in the present invention, a diagnosis focusing on a zero point is performed as a feature quantity used for abnormality diagnosis. The zero point is a zero crossing in the waveform in the time domain, and is a point where the energy becomes zero on the frequency domain. The density distribution of the size of the zeros is a distribution corresponding to the phase change. This is because the interval of zero crossings in the time domain appears on the complex plane as the size of the zero, and on the complex plane, the number of zeros can be counted for each size. That is, when the phase change does not occur, the distribution is uniquely determined, and when the phase change occurs, the zero point density distribution changes.

From the above, by using the zero-point density distribution, abnormal sound with phase change is detected among abnormal sounds that cannot be detected by the conventional method and in which amplitude intensity (sound pressure) does not change. .

The present invention calculates, for example, the size of a zero point of sound data for one cycle of the input signal in an abnormal sound detection system that performs abnormality diagnosis of the device using sound data collected from the device to be diagnosed as an input signal. The diagnosis by comparing the zero point size calculation unit, the zero point density distribution calculation unit for calculating the zero point density distribution from the zero point size, the zero point density distribution of the input signal, and the zero point density distribution of normal sound data And a normal / abnormal diagnosis unit that determines whether the target device is normal or abnormal.

According to the present invention, it is possible to detect an abnormal sound with a small change in amplitude intensity and a phase change.

Issues, configurations, and effects other than those described above will be clarified by the following description of embodiments.

It is an example of a block diagram of an abnormal sound detection system. It is an example of a hardware configuration. It is an example of a system flowchart. It is an example of the process of 1 period calculation part. It is an example of a zero point density distribution data structure. It is an example of a zero point density distribution. It is an example of a zero point density distribution display. It is an example of a diagnosis of a motor sound.

Embodiments of the present invention will be described with reference to the drawings. In the drawings, the same or similar components are denoted by the same reference numerals and description thereof is omitted.

FIG. 1 is an example of a configuration diagram of an abnormal sound detection system. FIG. 2 is an example of a hardware configuration. As shown in FIG. 1, the abnormal sound detection system 1 includes a one-cycle calculation unit 101, a zero point calculation unit 102, a zero point size calculation unit 103, a zero point density distribution calculation unit 104, an accumulated signal collation unit 105, and a normal / abnormal diagnosis unit. 106 and a zero point density database 107. The abnormal sound detection system 1 shown in FIG. 1 corresponds to the data processing unit H02 shown in FIG. 2, and the processing is realized by, for example, a computer or a microcomputer. Further, the display 108 for displaying the result such as the density distribution shown in FIG. 1 corresponds to the output result display unit H03 in FIG.

The abnormal sound detection system 1 uses the input signal I1 as an input to the one-cycle calculation unit 101, and the one-cycle calculation unit 101 cuts out one cycle of the input signal I1 as a signal for one cycle of the input signal. Output to the zero point calculation unit 102. The input signal I1 is sound data collected from the device to be diagnosed. The input means for the input signal I1 is directly input from the microphone H01 shown in FIG. 2 to the data processing unit H02, and the data logger from the microphone H01. A case where the data is input to the data processing unit H02 via, for example, can be considered.

The zero point calculation unit 102 calculates a zero point from the signal corresponding to one cycle of the input signal calculated by the one period calculation unit 101, and outputs the calculated zero point to the zero point size calculation unit 103. The zero point is calculated by performing n-order polynomial approximation and solving the n-th order approximation. As a polynomial approximation method, a method of numerical calculation (reference document: numerical calculation method, written by Hideyo Nagashima) such as a method using power series expansion and a method using Lagrange interpolation is known.

The zero point size calculation unit 103 calculates the size of the zero point from the zero point calculated by the zero point calculation unit 102. Zeros are defined as complex numbers. Therefore, the magnitude is calculated by taking the absolute value of the complex number. The zero size calculated by the zero size calculator 103 is input to the zero density distribution calculator 104.

The zero density distribution calculation unit 104 calculates the zero density distribution from the input zero size data. Specifically, the ratio of the number of zeros having a specific zero point size to the total number of zeros is calculated. Zeros are calculated for the number of approximate orders. For example, when approximated to a sixth-order function, a maximum of six zeros are calculated. In addition, there are six periods of waveforms, and when approximated to a sixth-order function, the total number of zeros is 36. The zero point density distribution of the input signal I1 calculated by the zero point density distribution calculation unit 104 is output to the accumulated signal collation unit 105.

Also, the abnormal sound detection system 1 performs the same calculation in the one cycle calculation unit 101, the zero point calculation unit 102, the zero point size calculation unit 103, and the zero point density distribution calculation unit 104 even for the accumulated signal I2. Here, the accumulation signal I2 is a normal sound collected when the device is in a normal state. The zero point density distribution of the accumulated signal I2 calculated by the accumulated signal I2 is recorded in the zero point density database 107.

The accumulated signal collating unit 105 compares an accumulated signal for comparison with the zero density distribution of the input signal I1 calculated from the input signal I1 from the zero density distribution database of the accumulated signal I2 constructed in the zero density database 107. The zero density distribution of I2 is read, and the zero density distribution of the input signal I1 and the zero density distribution of the accumulated signal I2 are output to the normal / abnormal diagnosis unit 106.

In the zero density database 107, the zero density distribution of the accumulation signal I2 is accumulated as normal data.

The normal / abnormal diagnosis unit 106 compares the zero point density distribution of the input signal I1 with the zero point density distribution of the accumulated signal I2, performs normal or abnormal diagnosis, and outputs a diagnosis result. As a comparison method of the zero density distribution of the input signal I1 and the zero density distribution of the accumulated signal I2, a method of comparing by calculating a correlation coefficient between the zero density distribution of the input signal I1 and the zero density distribution of the accumulated signal I2. Etc. are considered.

When comparing using the correlation coefficient, a correlation coefficient of 1.0 indicates that the correlation is strong, and that the accumulated signal I2 and the input signal I1 are the same, and the correlation coefficient When 0.0 is 0.0, it indicates that there is no correlation, and that the accumulated signal I2 and the input signal I1 are different. The correlation coefficient is already known as a method for statistically evaluating the degree of similarity between random variables. Note that the threshold value for distinguishing between normal and abnormal using the correlation coefficient needs to be determined according to the device to be diagnosed, the device to be detected, the progress of failure, and the like. For example, the normal / abnormal determination threshold is set to 0.8. If the correlation coefficient is equal to or higher than the normal / abnormal determination threshold, it is determined to be normal, and if the correlation coefficient is less than the normal / abnormal determination threshold, it is determined to be abnormal.

The display 108 displays a normal or abnormal determination result determined by the normal / abnormal diagnosis unit 106. The abnormal sound detection system 1 has a display control unit (not shown), and controls display on the display 108. The display 108 may also display other information together, as will be described later with reference to FIG.

FIG. 3 is an example of a system flowchart. In the one cycle calculation step F01, the input signal I1 or the accumulated signal I2, the rotation speed I10 of the diagnosis target device, and the device name I50 of the diagnosis target device are input to the one cycle calculation unit 101. The input method of the rotation speed I10 and the device name I50 may be a keyboard, online, or the like, and the input method is not limited. The one-cycle calculation unit 101 calculates and outputs a signal for one cycle of the input signal (or one cycle of the accumulated signal).

Based on the signal for one cycle of the input signal (or one cycle of the accumulated signal) calculated in the one cycle calculation step F01, the zero point calculation unit 102 calculates and outputs the zero point in the zero point calculation step F02. In the zero size calculation step F03, the zero size calculation unit 103 calculates the size of the zero.

In the zero point density distribution calculation step F04, the zero point density distribution calculation unit 104 calculates the zero point density distribution. The zero density distribution calculation unit 104 also receives a density distribution interval I60. The input method of the interval I60 may be input directly from the keyboard, online, or the like. The interval I60 is an interval for calculating the number of zero points for each density distribution calculation. For example, the interval I60 being 10 means that the size of the zero point is calculated in increments of 10.

In the accumulated signal collation step F05, the accumulated signal collation unit 105 sends the zero point density distribution of the input signal I1 calculated in the zero point density distribution calculation step F04, the rotation speed I10 of the diagnosis target device corresponding to the input signal I1, and the input signal. The diagnosis target device name I50 corresponding to I1 is input, the rotation speed I10 of the diagnosis target device corresponding to the input signal I1, and the zero point density of the accumulated signal I2 corresponding to the diagnosis target device name I50 corresponding to the input signal I1 The distribution is read from the zero density database 107. In the normal / abnormal diagnosis step F06, the normal / abnormal diagnosis unit 106 determines normal / abnormal.

FIG. 4 is an example of processing of the one-cycle calculation unit. The one-cycle calculation unit 101 calculates one cycle of the input signal I1 from the rotation speed I10 of the input signal I1, the recording time I20, and the sampling frequency I30, cuts out one cycle, and outputs I40 for one cycle of the input signal. . For example, when the rotational speed is 60 (Hz), the recording time is 20 (seconds), and the sampling frequency is 50000 (Hz), the rotational speed I10 is 60 (Hz), the recording time I20 is 20 (seconds), and the sampling frequency I30 is 50000 (Hz) is input. The input means may be data such as a keyboard and online data. In FIG. 4, the process in the case of the input signal I1 is shown as an example, but the process in the case of the accumulated signal I2 is the same.

FIG. 5 shows an example of the zero point density distribution data structure. The data configuration K01 of the zero point density database 107 shown in FIG. 5 includes a device name I50, a rotation speed I10, a zero point density distribution I70, and an interval I60. Regarding the rotational speed I10, either rpm, which is the rotational speed per minute, or Hz, which is the rotational speed per second, can be considered. In the case of rpm, which is the number of revolutions per minute, it is necessary to separately convert to Hz, which is the number of revolutions per second. For example, when 600 rpm is input, it is necessary to convert to 10 Hz. In the example of FIG. 5, the device X, the rotation speed 10 (Hz), and the zero point density distribution indicate that the interval of the zero point size is 10%, 20%, 40%, 20%, and 10% for every 10 respectively. ing.

FIG. 6 is an example of a zero point density distribution. In the zero density distribution M01 shown in FIG. 6, the horizontal axis indicates the size of the zero and the vertical axis indicates the zero density (%). The zero point density (%) is the ratio of the number of zeros having a specific zero size to the number of all zeros. For example, in FIG. 6, when the number of all zeros is 600, the number of zeros is 0 or more and less than 10 and 60, and similarly, the number of 10 or less and less than 20 is 120, 20 or more and 30. When the number is less than 240, the number between 30 and less than 40 is 120, and the number between 40 and less than 50 is 60, the density of zeros is less than 10 but less than 10 is 10 (%), less than 10 and less than 20 The density is 20 (%), the density 20 to 30 is 40 (%), the density 30 to 40 is 20 (%), and the density 40 to 50 is 10 (%). .

FIG. 7 is an example of a zero density distribution display. A zero density distribution display D01 shown in FIG. 7 is an example of a screen displayed on the display 108. The zero point density distribution M1 of the input signal I1 is indicated by a dotted line, and the zero point density distribution M2 of the accumulated signal I2 corresponding to the input signal I1 is indicated by a solid line. The horizontal axis indicates the size of the zeros and the vertical axis indicates the zero point density (%). It is. Further, FIG. 7 shows an example in which the index calculated by the normal / abnormal diagnosis unit 106 is also displayed. For example, when a normal sound and an abnormal sound are determined using a correlation coefficient, when the calculated correlation coefficient is 0.7 and the normal / abnormal determination threshold value is 0.8, in FIG. In this example, “correlation coefficient 0.7”, the normal / abnormal determination threshold is “correlation coefficient 0.8”, and the determination result is “abnormal”. These display controls are performed by a display control unit (not shown) of the abnormal sound detection system 1.

Fig. 8 shows an example of diagnosis of motor noise. FIG. 8 shows a case where the device to be diagnosed is a motor J01. A power source J02 for driving the motor is connected to the motor J01.

And, the microphone H01 is installed near the motor J01. The sound (sound data) recorded from the microphone H01 is input to the data logger J03. The recorded sound input to the data logger J03 is input as an input signal I1 to the data processing unit H02, diagnostic processing is performed, and the diagnostic result is displayed on the output result display unit H03.

A normal sound is recorded in advance as the accumulated signal I2, the zero density distribution of the accumulated signal I2 is calculated by the data processing unit H02, and recorded as zero point density distribution data in the zero density database 107.

As a method for collecting the accumulation signal I2, for example, there are a method of recording in a plurality of times every 60 seconds, and a method of continuously operating the device and accumulating signals. Similarly, for the input signal I1, a method of inputting a plurality of times or a method of performing a diagnosis by inputting a signal while continuously operating can be considered.

As described above, in the abnormality diagnosis, by using the zero point density distribution, the abnormal sound that cannot be detected by the conventional method and does not cause the amplitude intensity (sound pressure) change is accompanied by a phase change. Abnormal sound can be detected.

The embodiments of the present invention have been described above. However, the configurations described in the above embodiments are merely examples, and the present invention can be appropriately changed without departing from the technical idea.

DESCRIPTION OF SYMBOLS 1 Abnormal sound detection system 101 1 period calculation part 102 Zero point calculation part 103 Zero point size calculation part 104 Zero point density distribution calculation part 105 Accumulated signal collation part 106 Normal / abnormality diagnosis part 107 Zero point density database 108 Display H01 Microphone H02 Data processing part H03 Output result display section F01 1 cycle calculation step F02 Zero point calculation step F03 Zero point size calculation step F04 Zero point density distribution calculation step F05 Accumulated signal collation step F06 Normal / abnormal diagnosis step I1 Input signal I2 Accumulated signal I10 Number of revolutions I20 Recording time I30 Sampling time Frequency I40 One period of input signal I50 Device name I60 Interval I70 Zero density distribution K01 Zero density distribution data configuration M01 Zero density distribution M1 Zero density distribution M2 of input signal Zero of accumulated signal Degree distribution D01 zero density distribution display J01 motor J02 power J03 data logger

Claims

In an abnormal sound detection system that performs an abnormality diagnosis of the device using sound data collected from the device to be diagnosed as an input signal,
A zero size calculator for calculating the size of the zero of sound data for one cycle of the input signal;
A zero-point density distribution calculating unit for calculating a zero-point density distribution from the size of the zeros;
An abnormal sound comprising: a normality / abnormality diagnosis unit that compares the zero point density distribution of the input signal with the zero point density distribution of normal sound data to determine whether the device to be diagnosed is normal or abnormal Detection system.
The normal / abnormal diagnosis unit calculates a correlation coefficient between the zero-point density distribution of the input signal and the zero-point density distribution of the normal sound data, and compares the correlation coefficient with a normal / abnormal determination threshold value. The abnormal sound detection system according to claim 1, wherein the device to be diagnosed is determined to be normal or abnormal.
The abnormal sound detection system according to claim 1 or 2, further comprising a display control unit that displays a determination result of whether the diagnosis target device is normal or abnormal.
The abnormal sound detection system according to claim 3, wherein the display control unit displays a zero point density distribution of the input signal and a zero point density distribution of the normal sound data.