JP3641817B2 - Water leakage sound identification method by high accuracy frequency analysis - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a water leakage sound identification method, and more particularly to a water leakage sound identification method by a high-accuracy frequency analysis method.
[0002]
[Prior art]
Conventionally, as a water leakage sound identification method, there are a method using a pattern recognition method or a method using a high-accuracy frequency analysis method.
The pattern recognition method uses the same processing as the relative level vs. frequency characteristics of the detection sound recorded by the sensor installed for the detection knowledge of the leak location, and leaks sound, similar sound, obstacle sound, A typical standard pattern detection sound pattern and frequency band registered by knowing which one is a no-signal sound is identified with the factors of the standard pattern detection sound approximated to the detection sound. This is a method for detecting sound.
[0003]
Also, those using high-accuracy frequency analysis methods obtain level-to-frequency characteristics by performing high-accuracy frequency analysis, and from the moving average characteristics obtained by moving average with a predetermined frequency width, The relative level Li of the spectrum and the bandwidth BWi where the spectrum cuts the average level line (hereinafter, the relative level Li and the bandwidth BWi are collectively referred to as “feature”) are obtained, and are attached to a predetermined number above the bandwidth BWi. Thus, (ΣBWi, ΣLi) is given on the relative level vs. bandwidth coordinate plane from the sum of the respective feature quantities, and the relative level vs. bandwidth coordinate plane is roughly on the relative level vs. bandwidth coordinate plane. A data area, a similar sound data area, an obstacle sound data area, and a no-signal data area can be considered. For each measurement data, the detection sound is identified from the area where the coordinate point of (ΣBWi, ΣLi) exists. It is a method of.
[0004]
[Problems to be solved by the invention]
However, in the conventional water leakage sound identification method, those using the pattern recognition method require a typical standard pattern based on actual data, and further, the recognition process is complicated.
[0005]
Also, those using the high-precision frequency analysis method are not always clear because the overlapping part cannot be ignored in the leaked sound data area, similar sound data area, obstacle sound data area, and no signal data area on the level vs. bandwidth coordinate plane. It was not possible to classify them into two, and a sufficient correct identification probability could not be obtained, which was not technically satisfactory.
[0006]
[Means for Solving the Problems]
According to the present invention, a leakage sound identification method using a high-accuracy frequency analysis method receives a detection sound for each leakage sound detection knowledge, performs a high-accuracy frequency analysis on the received detection sound, calculates a spectrum in a frequency space, Is calculated with a predetermined frequency width, the relative distance between the spectrum and the moving average spectrum is calculated to enhance the spectrum, and the peak spectrum signal level and the peak frequency exceeding the reference value level are included. A predetermined number is extracted as a feature quantity in descending order of the frequency width with respect to the frequency width, and the first correction process and the second correction process are performed on the feature quantity based on two correction conditions having different preset correction ranks. Each is performed, and the leakage sound is identified from the detection sound based on each of the feature values after the first correction process and the second correction process.
[0007]
DETAILED DESCRIPTION OF THE INVENTION
FIG. 1 is a block diagram showing a configuration of a water leakage sound identification apparatus for carrying out a water leakage sound identification method according to an embodiment of the present invention.
In the figure, 1 is a sensor that detects a detection sound such as a leaking sound, 2 is a data recorder that receives the output signal S1 of the sensor 1, performs level conversion and recording of the signal, and 3 is detected by the sensor 1 from the data recorder 2. A frequency analysis circuit that receives the detected signal S2 of the detected sound and performs high-accuracy frequency analysis on the input signal, and a spectrum enhancement processing circuit that receives the output signal S3 of the frequency analysis circuit 3 and emphasizes the peak of the input signal. Reference numeral 5 denotes an integration circuit that receives the output signal S4 of the spectrum enhancement processing circuit 4 and integrates the input signal at predetermined time intervals. Reference numeral 6 denotes an output signal S5 of the integration circuit 5 that receives the low frequency range of the input signal. A low-frequency cutoff circuit 7 that receives the output signal S6 of the low-frequency cutoff circuit 6 is input to the peak signal level of the input signal and exceeds the reference value level. A feature quantity extraction circuit for extracting a characteristic parameter of the frequency band width including the frequency of click.
[0008]
Reference numeral 8 denotes an output signal S7 from the feature quantity extraction circuit 7, and a correction circuit A for performing correction processing on the input signal. Reference numeral 9 denotes an output signal S8 from the correction circuit A8, and the frequency width of the input signal. , And the peak signal level are added in the order of the frequency width, and based on the sum of the frequency bandwidths and the sum of the peak signal levels, the addition processing identification circuits A, 10 for identifying whether or not there is a water leak sound An output signal S7 from the feature quantity extraction circuit 7 is input, and correction circuits B and 11 that perform correction processing on the input signal receive an output signal S10 from the correction circuit A10. Addition processing identification circuits B and 12 that add signal levels in order of frequency width and identify whether or not there is a water leakage sound based on the sum of the frequency bandwidth and the sum of the peak signal levels are addition processing identification circuits A9 output signal 9 and the output signal S11 of the addition processing identification circuit B11 are respectively input, and based on the input signal, a comprehensive identification circuit for comprehensively identifying the leaking sound, 13 is input with the output signal S12 of the general identification circuit 12, This is a display for displaying the identification result based on the input signal.
[0009]
Next, the operation of this embodiment will be described.
First, the sensor 1 receives a detection sound such as a water leak sound and outputs the received detection sound S1 to the data recorder 2. The data recorder 2 amplifies or attenuates the input detection sound S1 to a required level suitable for recording, records it as the detection sound S2, and outputs it to the frequency analysis circuit 3.
[0010]
In the frequency analysis circuit 3, the frequency analysis width is sufficiently narrowed with respect to the detection sound S2, and a highly accurate spectrum analysis is performed to obtain a spectrum S3. The spectrum S3 is output to the spectrum enhancement processing circuit 4, and the spectrum is analyzed. The enhancement processing circuit 4 creates an enhanced spectrum S4 consisting only of sharp peaks of frequencies obtained by extracting only the spectrums of the individual peaks, regardless of the individual signal levels, to the integration processing circuit 5. Output.
[0011]
Here, the processing of the spectrum enhancement processing circuit 4 will be described in more detail with reference to FIGS. 2 is a diagram illustrating an example of a spectrum S3 output from the frequency analysis circuit 3, FIG. 3 is a diagram illustrating an example of a moving averaged spectrum, and FIG. 4 is a diagram illustrating an example of an output signal S4 of the spectrum enhancement processing circuit 4. is there.
First, in the spectrum emphasis processing circuit 4, the average level of the average section of a predetermined frequency width is applied to the spectrum S 3 output from the frequency analysis circuit 3 in the frequency band to be subjected to frequency analysis as shown in FIG. A moving average spectrum as shown in FIG. 3 is obtained as a trajectory of the average value of the average section by moving the average section as the moving average level of the center point of the average section.
[0012]
Then, the moving averaged spectrum is shifted by the reference value level VR and subtracted from the spectrum S3, whereby the relative value relative to the moving averaged spectrum as shown in FIG. 4 at the reference value level VR of the spectrum S3. As the level LR, an enhanced spectrum S4 consisting only of a sharp peak of the frequency is obtained, and the spectrum S4 is output from the spectrum enhancement processing circuit 4.
[0013]
In addition, the spectrum S4 output from the spectrum enhancement processing circuit 4 has frequency and level fluctuations, and these fluctuations are also important elements for identifying leaked sounds, and thus are emphasized by the spectrum enhancement processing circuit 4. The integration processing circuit 5 integrates the spectrum S4 at every integration time of a predetermined time, and outputs the spectrum S5 subjected to the integration processing to the low-frequency cutoff circuit 6.
[0014]
In the low-frequency cutoff circuit 6, main parts of non-leakage sounds such as similar sounds and obstacle sounds are concentrated in the low-frequency part of the spectrum, and an integration process is performed to remove the low-frequency part. The spectrum S5 is subjected to low-frequency cutoff processing, and the spectrum S6 is output to the feature quantity extraction circuit 7.
[0015]
As shown in FIG. 5, the feature amount extraction circuit 7 has a frequency width (hereinafter referred to as bandwidth) including the peak frequency fi and the relative level LR exceeding the reference value VR with respect to the spectrum S6 subjected to the low-frequency cutoff processing. BWi is extracted, the order of the size of the bandwidth BWi, the bandwidth BWi, and the relative level Li of the peak frequency fi corresponding to the bandwidth BWi (hereinafter referred to as the peak frequency fi, the size of the bandwidth BWi). And the relative amount Li of the peak frequency fi are collectively referred to as a feature amount), and feature amounts for a predetermined number of bandwidth BWi are extracted, and the correction circuit A8 and the correction are performed as a spectrum feature amount output S7. Output to circuit B10.
[0016]
Then, the correction circuit A8 performs correction processing A using the correction reference line A on the level vs. bandwidth coordinate plane as shown in FIG. It is output to the addition processing identification circuit A9 as a spectrum feature amount correction A output S8.
Further, the correction circuit B10 performs the correction process B using the correction reference line B on the level pair and bandwidth coordinate plane as shown in FIG. The output is output to the addition processing identification circuit B11 as the spectrum feature amount correction B output S10.
[0017]
Here, correction by the correction circuit A8 and the correction circuit B10 will be described.
FIG. 7 is an explanatory diagram for explaining correction by the correction circuit.
First, as shown in FIG. 7, it is known that the difference in the feature amount between the water leak sound data and the non-water leak sound data tends to be shown in the following (a) and (b).
[0018]
(A) In the case of leaked sound data, even if the bandwidth of the feature amount element is equal, the level is lower than that of the non-leakage sound data and is close to a trapezoidal shape. The level is high and the feature amount is close to a triangle.
(B) In the case of water leakage sound data, if the bandwidth BWi increases, the relative level increases in proportion to the increase in bandwidth.
[0019]
From the tendency of (a) and (b) as described above, the correction reference line A and the correction in which the relative level increases in proportion to the increase in bandwidth on the relative level vs. bandwidth coordinate plane as shown in FIG. A reference line B is created, and the relative level is corrected in the manner shown in (1) and (2) below (this process is hereinafter referred to as “correction process”).
In addition, as a method of creating the correction reference line A and the correction reference line B, a large amount of data is taken in advance, and from the tendency, the correction reference line A mainly has a water leakage sound in the region below the correction reference line A. The area between the correction reference line B and the correction reference line A is created so that water leak sound data and non-water leak sound data are mixed.
[0020]
(1) If the relative level of the feature amount does not reach the correction reference line, the relative level is unconditionally set to 0 dB, assuming that the feature amount is the leaked sound data.
(2) If the relative level of the feature amount is equal to or higher than the correction reference line, it is assumed that the feature amount of the non-leakage sound data is left as it is, and the correction process is not performed.
[0021]
Then, by performing this correction process, if the sums ΣBWi and ΣLi of the feature values are obtained from the data subjected to the correction process as described later, the water leakage sound tendency is large, that is, relative to the bandwidth. Leakage sound data with lower levels and more features to be corrected have a greater decrease in ΣLi compared to the data before the correction process, but on the other hand, the leaky sound tendency is small, that is, it corresponds to the bandwidth. Therefore, non-leakage sound data with a high relative level and many features that are not subject to correction have a smaller decrease in ΣLi compared to the data before correction processing. It will function in the direction of separation, and separation of leaking sound data and non-leakage sound data becomes easy, and as a result, improvement of the correct discrimination probability of leaking sound and non-leakage sound can be expected. .
[0022]
Further, in the addition process identification circuit A9, the corrected feature quantity as shown in FIG. 8 is added to the bandwidth BWi and the relative level Li of the feature quantity for each spectrum as shown in FIG. As shown in FIG. 10, the sum of the level ΣLi vs. the bandwidth sum ΣBWi is obtained from the relative position of ΣBWi and ΣLi with respect to the identification reference line 1 and the identification reference line 2 on the coordinate plane. Whether it is data, non-leakage data, or unknown data is identified and output to the overall identification circuit 12 as a leak / non-leakage identification output S9 of the spectrum after the correction process A.
[0023]
In addition, as a method of creating the identification reference line 1 and the identification reference line 2, a large amount of data is taken in advance, and from the tendency, the identification reference line 1 has an area under the identification reference line 1 mainly having a leak sound. The identification reference line 2 is created so that the area above the identification reference line 2 is mainly occupied by non-water leakage sound data.
[0024]
As described above, in the correction circuit A8 and the addition process identification circuit A9, the correction reference line A and the identification reference line 1 are set for an area where most of the data to be processed is water leakage data. By performing the correction process in the circuit A8, ΣBWi and ΣLi can easily identify the leakage data and the non-leakage data, and the identification reference line 2 is intended for an area where the majority is the non-leakage data. It is possible to increase the positive identification probability of the non-leakage data in the region exceeding the identification reference line 2.
[0025]
Further, in the addition processing identification circuit A11, as shown in FIG. 8, the corrected feature amount is added to the bandwidth BWi and the relative level Li of the feature amount for each spectrum as shown in FIG. And ΣLi, and as shown in FIG. 10, the data is obtained from the relational position of ΣBWi and ΣLi with respect to the identification reference line 3 and the identification reference line 4 on the coordinate plane, as shown in FIG. It is identified whether the data is leakage data, non-leakage data, or unknown data, and is output to the comprehensive identification circuit 12 as the leakage / non-leakage identification output S11 of the spectrum after the correction process B.
[0026]
In addition, as a method of creating the identification reference line 3 and the identification reference line 4, a large number of data is taken in advance, and from the tendency, the identification reference line 3 has mainly a water leakage sound in the region below the identification reference line 3. The identification reference line 4 is created so that the area above the identification reference line 4 is mainly occupied by non-water leakage sound data.
[0027]
As described above, in the correction circuit B10 and the addition process identification circuit B11, the correction reference line B is set between the correction reference line A and the correction reference line B as an area where water leakage data and non-water leakage data are mixed. Therefore, by performing the correction process in the correction circuit B11, the leaked data remaining from the correction process target of the correction circuit A9 is targeted, and the leaked data and the non-leaked data are preferentially mixed. Thus, both the leaked data and the non-leaked data can increase the probability of correct identification.
[0028]
Then, in the general identification circuit 12 to which the output signals S9 and S11 of the addition process identification circuit A9 and the addition process identification circuit B11 are input, the identification of the output signal S9 of the addition process identification circuit A9 is first performed for each detection sound data. When the result is data existing in the water leakage data area or the non-water leakage data area, the detection sound data is comprehensively identified as the water leakage data or the non-water leakage data.
[0029]
When the identification result of the output signal S9 of the addition process identification circuit A9 is data existing in the unknown data area, the addition process identification circuit is focused on the identification result of the output signal S11 of the addition process identification circuit B11. When the identification result of the output signal S11 of B11 is data existing in the leakage data area, the detection sound data is comprehensively identified as leakage data, and the identification result of the output signal S11 of the addition processing identification circuit B11 is non-leakage data. If the data is present in the area, the detection sound data is comprehensively identified as non-leakage data, and the identification result of the output signal S11 of the addition processing identification circuit B11 is data existing in the unknown data area. Is identified as unknown data.
[0030]
And the result identified comprehensively is output to a display as comprehensive identification output S12, and the display 13 displays the total identification output S12 for every detection sound data as an identification result.
[0031]
In this embodiment, correction processing is performed by changing the correction reference line used for the correction processing in the correction circuit A9 and the correction circuit B11, and the identification result based on the data after the correction processing by the correction circuit A9 and the correction circuit B11 is used. Since the comprehensive identification is performed in a duplex manner, it is possible to increase the probability of positive identification for both leaked data and non-leaked data.
[0032]
Although the data recorder 2 is used in this embodiment, the output of the sensor 1 may be input to the frequency analysis circuit 3 without using the data recorder 2. The low-frequency cutoff circuit 6 may be connected to the output side of the frequency analysis circuit 3 or the spectrum enhancement processing circuit 4.
[0033]
In this embodiment, the correction reference lines A and B and the identification reference lines 1, 2.3, and 4 are straight lines. However, any data such as a curve can be used as long as it takes a lot of data and is obtained from the tendency. Something like that.
[0034]
【The invention's effect】
As described above, according to the present invention, a detection sound is received for each leaked sound detection knowledge, the received detection sound is analyzed with high accuracy frequency, a spectrum in a frequency space is calculated, and the spectrum is set to a predetermined frequency width. In the moving average, the relative distance between the spectrum and the moving average spectrum is calculated to emphasize the spectrum, and the frequency level including the peak signal level and the peak frequency exceeding the reference value level of the emphasized spectrum A predetermined number is extracted as a feature quantity in descending order of frequency width, and a first correction process and a second correction process are performed on the feature quantity using two correction conditions having different preset correction ranks. Since the leakage sound is identified from the detection sound based on each of the correction amount and the feature amount after the second correction processing, the leakage data and the non-leakage Even when the over data are mixed, it has the effect that both the separation becomes extremely easy, and the positive identification probability of both the leakage data and non-leakage data can be greatly improved.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration of a water leakage sound identification device according to an embodiment.
FIG. 2 is a diagram illustrating an example of a spectrum S3 output from the frequency analysis circuit 3;
FIG. 3 is a diagram showing an example of a spectrum obtained by moving average.
4 is a diagram illustrating an example of an output signal S4 of the spectrum enhancement processing circuit 4. FIG.
FIG. 5 is an explanatory diagram for explaining feature amount extraction;
FIG. 6 is a diagram showing a relationship between a correction reference line and a data area on a relative level vs. bandwidth coordinate plane.
FIG. 7 is an explanatory diagram for explaining correction by a correction circuit;
FIG. 8 is a diagram illustrating a feature amount example and a level correction example;
FIG. 9 is an explanatory diagram for explaining a feature amount addition process;
FIG. 10 is an explanatory diagram for explaining water leakage sound identification in the addition processing identification circuit A;
FIG. 11 is an explanatory diagram for explaining the identification of a water leakage sound in the addition processing identification circuit B;
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Sensor 2 Data recorder 3 Frequency analysis circuit 4 Spectrum emphasis processing circuit 5 Integration circuit 6 Low frequency cutoff circuit 7 Feature amount extraction circuit 8 Correction circuit A
9 Addition processing identification circuit A
10 Correction circuit B
11 Addition processing identification circuit B
12 General identification circuit 13 Display

Claims (7)

Receive detection sound for different leak sound detection knowledge,
Analyzing the received detection sound with high precision frequency and calculating a spectrum on the frequency space,
From the spectrum, the peak signal level and the frequency width including the peak frequency exceeding the reference value level are extracted as feature amounts,
A first correction process and a second correction process are performed on the feature amount according to two correction conditions having different correction ranks set in advance,
A water leakage sound identification method using a high-accuracy frequency analysis method, wherein a water leakage sound is identified from the detection sound based on each of the feature values after the first correction processing and the second correction processing. Receive detection sound for different leak sound detection knowledge,
Analyzing the received detection sound with high precision frequency and calculating a spectrum on the frequency space,
Moving average the spectrum with a predetermined frequency width, calculating the relative distance between the spectrum and the moving average spectrum to enhance the spectrum,
Extracting a predetermined number as a feature quantity in descending order of the frequency width with respect to the frequency width including the frequency of the peak exceeding the signal level and reference value level of the peak of the emphasized spectrum,
A first correction process and a second correction process are performed on the feature amount according to two correction conditions having different correction ranks set in advance,
A water leakage sound identification method using a high-accuracy frequency analysis method, wherein a water leakage sound is identified from the detection sound based on each of the feature values after the first correction processing and the second correction processing. Receive detection sound for different leak sound detection knowledge,
Analyzing the received detection sound with high precision frequency and calculating a spectrum on the frequency space,
Moving average the spectrum with a predetermined frequency width, calculating the relative distance between the spectrum and the moving average spectrum to enhance the spectrum,
Integrating the enhanced spectrum at a given time;
Extracting a predetermined number as a feature quantity in descending order of the frequency width with respect to the frequency width including the frequency of the peak exceeding the signal level and reference value level of the peak of the integrated spectrum,
A first correction process and a second correction process are performed on the feature amount according to two correction conditions having different correction ranks set in advance,
A water leakage sound identification method using a high-accuracy frequency analysis method, wherein a water leakage sound is identified from the detection sound based on each of the feature values after the first correction processing and the second correction processing. Receive detection sound for different leak sound detection knowledge,
Analyzing the received detection sound with high precision frequency and calculating a spectrum on the frequency space,
Moving average the spectrum with a predetermined frequency width, calculating the relative distance between the spectrum and the moving average spectrum to enhance the spectrum,
Integrating the enhanced spectrum at a given time;
Extracting a predetermined number as a feature quantity in descending order of the frequency width with respect to the frequency width including the frequency of the peak exceeding the signal level and reference value level of the peak of the integrated spectrum,
A first correction process and a second correction process are performed on the feature amount according to two correction conditions having different correction ranks set in advance,
For each of the feature values after the first correction process and the second correction process, add the frequency width and peak signal level of the feature value, respectively.
The sum of the frequency width and the sum of the peak signal levels with respect to the added first correction processing and second corrected processing feature amounts, and the preset first correction processing and second post-correction features Water leakage by high-precision frequency analysis, characterized in that the leakage sound is identified from the detection sound based on the discrimination condition of the leakage sound and the non-leakage sound with respect to the sum of the frequency width of the quantity and the sum of the signal level of the peak Sound identification method. Receive detection sound for different leak sound detection knowledge,
Analyzing the received detection sound with high precision frequency and calculating a spectrum on the frequency space,
Moving average the spectrum with a predetermined frequency width, calculating the relative distance between the spectrum and the moving average spectrum to enhance the spectrum,
Integrating the enhanced spectrum at a given time;
Extracting a predetermined number as a feature quantity in descending order of the frequency width with respect to the frequency width including the frequency of the peak exceeding the signal level and reference value level of the peak of the integrated spectrum,
A first correction process and a second correction process are performed on the feature amount according to two correction conditions having different correction ranks set in advance,
For each of the feature values after the first correction process and the second correction process, add the frequency width and peak signal level of the feature value, respectively.
The sum of the frequency width and the peak signal level with respect to the added feature value after the first correction processing, and the sum of the frequency width and peak signal level of the feature amount after the first correction processing set in advance. Based on the condition for identifying the leaked sound / non-leakage sound for the sum of
When the leaked sound / non-leakage sound cannot be identified by the first leaked sound identification, the sum of the frequency width and the sum of the peak signal levels with respect to the added feature amount after the second correction processing is set in advance. Based on the water leak sound / non-water leak sound identification condition for the sum of the frequency width of the feature amount after the second correction process and the sum of the peak signal levels, the second leak sound is identified.
A water leakage sound identification method using a high-accuracy frequency analysis method, wherein a water leakage sound is identified from the detection sound based on identification results of the first water leakage sound identification and the second water leakage sound identification. For the two correction conditions with different correction ranks, a specific peak signal level with respect to the frequency width of the feature amount is set in advance with respect to the frequency width and peak signal level of the feature amount, and the set specification It is assumed that the signal level of the peak of the feature amount that is equal to or lower than the signal level of “0” is set to “0”, and two correction conditions differ in the signal level of the specific peak with respect to the frequency width of the feature amount. A method for identifying leaked sound by the high-precision frequency analysis method according to claim 1, 2, 3, 4, or 5. 7. The water leakage sound identification method according to claim 1, wherein a low frequency component of a spectrum is removed before extracting the feature amount.

Images