JP5783808B2 - Abnormal sound diagnosis device - Google Patents

Description

  The present invention relates to an apparatus for determining the possibility of occurrence of abnormal sound in a device being operated by time-frequency analysis of signals collected by a microphone (hereinafter referred to as a microphone) or a vibration sensor. In particular, the present invention relates to an abnormal sound diagnosis apparatus that diagnoses a wide variety of abnormal sounds generated during operation in a system composed of a plurality of devices.

  As a first technique related to an abnormal sound diagnosis apparatus for diagnosing a conventional abnormal sound, the time at which the intensity of the unsteady vibration becomes a set value or more from the time frequency distribution obtained by the time frequency analysis processing of vibration data from the determination object is obtained. Unsteady vibration data is extracted, and the possibility of occurrence of abnormal noise is determined based on the extracted unsteady vibration data (Patent Document 1). Calculating the time ratio of data that is equal to or greater than the vibration generation determination amplitude value obtained by the product of the maximum amplitude and the vibration generation threshold value in Japanese Patent Application Laid-Open No. 2003-259542 (Patent Document 2), an abnormal sound component from the time series spectrum There is a technique in which a region having a high intensity indicated by the contour lines is calculated and only a spectrum sequence including an abnormal sound component is extracted from this region (Patent Document 3).

  Further, as a second technique related to an abnormal sound diagnosis apparatus for diagnosing a conventional abnormal sound, attention is paid to specific known abnormal events, and dedicated processing means for confirming the occurrence of such abnormal events and when no known abnormal events occur. Perform general-purpose noise analysis, compare this analysis result with normal, detect unspecified unknown abnormal event, and generate processing procedure to detect change from normal state when unknown abnormal event is detected Then, there is what is given to the dedicated processing means (Patent Document 4).

Japanese Patent No. 3885297 Japanese Patent No. 4373350 Japanese Patent No. 4262878 JP-A-6-309580

Since the conventional first technique is configured to detect an abnormality based on various threshold values specialized for the appearance pattern of a specific abnormal sound component appearing in the frequency analysis result, each individual configuration has various bandwidths and There is a problem in that abnormal sound components having a duration cannot be diagnosed equally accurately. In addition, there is a problem in that there is no guarantee that an optimal diagnosis result is always obtained because a region where the intensity of the time frequency distribution is large is obtained from the observation data from the bottom up.
On the other hand, the second conventional technique has a problem that a diagnostic procedure for an unknown abnormal event needs to be registered in advance in the dedicated processing means in order to diagnose the unknown abnormal event.
The present invention has been made to solve the above problems, and it is not necessary to register a dedicated diagnostic procedure in the dedicated processing means, and there is a guarantee that an optimal diagnostic result can be obtained, which is included in the frequency analysis result. It is an object of the present invention to provide a method for detecting abnormal sound components having various bandwidths and durations with high accuracy.

The abnormal sound diagnosis apparatus according to the present invention is:
Waveform data acquisition means for capturing waveform data of sound or vibration generated by the device to be inspected;
Time frequency analysis of the waveform data, time frequency analysis means for obtaining a time frequency distribution with one axis as a time axis and the other axis as a frequency axis;
Prior knowledge specifying the shape of the appearance region in the time frequency distribution of the abnormal sound component generated from the device obtained by analyzing the inspection target device in advance, which is defined by the time axis of the time frequency distribution and the coordinate value of the frequency axis. A region extracting means for generating a plurality of regions based on the above and extracting a region including a variation component different from the steady state of the time frequency distribution;
A determination means for determining and outputting an abnormality based on the time-frequency distribution included in the extraction region ,
The region extraction means includes a region candidate generation unit that extracts a region including a variation component different from the steady state of the time frequency distribution as a region candidate, a time frequency distribution included in the region candidate, and a normal time frequency distribution. provided from the relationships and evaluation unit for determining the degree of condensation, it outputs the degree of condensation is greater region candidate extraction region.

According to abnormal sound diagnostic apparatus according to the present invention,
Dedicated diagnostic procedures for abnormal sound components with various bandwidths and durations appearing in frequency analysis results by providing means for extracting time frequency regions that are continuously formed from the time frequency distribution. There is no need to register the system, and the diagnosis can be performed with high accuracy.

It is a functional block block diagram which shows the abnormal sound diagnostic apparatus of this invention. It is a characteristic view of the time frequency distribution example in the case of scanning the sound from a plurality of devices. It is explanatory drawing of prior knowledge regarding the area | region of a time frequency distribution. It is a flowchart of the process in Embodiment 1 of this invention.

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 sound pressure generated by devices constituting a system to be inspected. It has a diagnostic mode for capturing waveforms during testing. The measurer installs a microphone or vibration sensor on the inspection target device, connects the microphone or vibration sensor to the input terminal of the USB (Universal Serial Bus) interface of the PC, and operates in the learning mode and diagnostic mode. Do.

As a system to be inspected, for example, a microphone is attached to an elevator car, and the microphone signal is taken into a PC placed in a machine room via a control cable, and the passenger car is reciprocated, so that Take the case of diagnosing the operating sound of each device as an example.
When an abnormal sound is generated from a specific device, for example, a top return car, there is an operating sound generated by a device other than the device that generates the abnormal sound, for example, a guide rail sliding sound. When the car approaches the top return car, for example, when the car approaches the top return car (when the car goes up, it is in the second half of the measurement section, and when the car goes down, the first half of the measurement section) The time frequency component of appears.
In addition, when abnormal sound is generated from the counterweight, the time frequency component of the abnormal sound appears in the center of the measurement section, which is a time zone in which the counterweight and the passenger car pass each other.

Further, the shape of the frequency spectrum of the abnormal sound varies depending on the device in which the abnormality occurs and the cause of the abnormality, and the occupied frequency range is various. In general, as described above, when operating sounds of a plurality of devices are diagnosed from a microphone that scans a device system, the time range and frequency range in which abnormal sound components appear in a measurement section are extremely complex and diverse.
FIG. 2 shows the time frequency distribution when an abnormal sound occurs in each elevator device with time on the horizontal axis and frequency on the vertical axis, and the intensity of the distribution at each time and each frequency is shown in shades. The dotted line indicates the intensity of the operating sound of the specific device before the occurrence of the abnormality with the microphone, and the solid line indicates the intensity of the operating sound of the device that has become abnormal. The alternate long and short dash line indicates the intensity of the synthesized sound of operating sounds from all devices including the device. (A) is an example when abnormal sound is generated from the top return wheel, and (B) is an example when abnormal sound is generated from the counterweight.

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 measurement signal output from a microphone or vibration sensor, 2 is a waveform that includes an amplifier, a low-pass filter circuit, and an AD converter, samples the measurement signal 1 and converts it into a digital signal, and outputs waveform data 3 The acquisition unit 4 applies a time window to the waveform data 3 and shifts the time window in the time direction to perform time-frequency analysis of the waveform data 3 by fast Fourier transform (hereinafter referred to as FFT), and from the spectrum value indicating the intensity with respect to time and frequency. It is a time frequency analysis part which outputs the time frequency distribution 5 which becomes.

  Reference numeral 6 denotes a normal time frequency distribution storage unit for storing the time frequency distribution 6a (not shown in the figure) when the time frequency distribution 5 is normal, and reference numeral 7 denotes a time frequency distribution 7a during the test of the time frequency distribution 5 (shown in the figure). (Not shown) time frequency distribution storage unit at the time of testing, 8 is a prior knowledge storage unit in which prior knowledge 8a (not shown) is stored as a table, 9 is based on prior knowledge 8a of the prior knowledge storage unit 8 A region candidate generation unit 11 that generates the predetermined region candidate 10 determined in this manner, and 11 is a normal time frequency distribution 6a of the normal time frequency distribution storage unit 6 and a test time frequency distribution storage unit 7 during the test for the region candidate 10. An evaluation unit that calculates and outputs the condensation degree 12 with reference to the time frequency distribution 7a, and 13 is an area that selects an optimum area candidate from the area candidates 10 based on the condensation degree 12 and outputs it as the extraction area 14. The output unit 15 calculates the degree of abnormality indicating the possibility of occurrence of abnormal sound from the time frequency distribution included in the extraction region 14 with reference to the normal time frequency distribution 6a and the diagnostic time frequency distribution 7a. An abnormal time calculation unit 17 for outputting is a determination unit that determines the possibility of occurrence of abnormal sound based on the degree of abnormality 16 and outputs a determination result 18.

The operation will be described below with reference to the flowchart of the processing in FIG.
In the learning mode or the diagnostic mode, the waveform acquisition unit 2 acquires, amplifies, and AD-converts the measurement signal 1 output from the microphone or vibration sensor, thereby obtaining a 16-bit linear PCM (pulse code modulation) with a sampling frequency of 32 kHz. Is converted into waveform data 3 of the digital signal (step S1).
The time frequency analysis unit 4 extracts a frame from the waveform data 3 output from the waveform acquisition unit 2 while shifting the time window of 1024 points in the time direction at intervals of 16 ms, and performs frequency calculation on each frame by FFT calculation. Is obtained as time-frequency distribution 5 (step S2). Here, t is a time at which a discrete value corresponding to the shift interval for shifting the analysis window is taken, and f is a frequency at which a discrete value corresponding to the frequency index as a result of the FFT operation is taken. The time t and the frequency f satisfy the relations 0 ≦ t ≦ T and 0 ≦ f ≦ F, respectively. Here, T is a time width in the time direction of the time frequency distribution 5, and F is a Nyquist frequency that is 1/2 of the sampling frequency fs of the waveform data 3 (F = fs / 2).

When the time frequency distribution 5 is calculated by the time frequency analysis unit 4, the abnormal sound diagnosis apparatus determines whether the time is the learning mode or the diagnosis mode (step S3).
In the learning mode, the time frequency distribution 5 is transferred to and stored in the normal time frequency distribution storage unit 6 as a normal time frequency distribution 6a (step S4). On the other hand, if the determination result in step S3 is in the diagnosis mode, the time frequency distribution 5 is transferred and stored in the diagnosis time frequency distribution storage unit 7 as the diagnosis time frequency distribution 7a (step S5).

Next, the operation of the diagnosis process in the diagnosis mode will be described.
The area candidate generation unit 9 generates an area candidate 10 based on the prior knowledge 8a (step S6). The prior knowledge 8a is knowledge for prescribing the shape of the appearance region in the time-frequency distribution of abnormal components generated from the devices constituting the diagnosis target system, and was obtained by analyzing the target in advance by the designer of this apparatus. It represents knowledge and is stored in the prior knowledge storage unit 8 in the form of a table as region candidates generated by the region candidate generation unit 9 of this apparatus. In this example, the entire time interval T is divided into n and the entire band F is divided into m with respect to the entire region of the time frequency distribution to obtain a lattice-like divided region, and a rectangle having an arbitrary lattice line as a side. Are generated and stored in the prior knowledge storage unit 8 as a table of prior knowledge 8a.

  FIG. 3 shows an example of a grid and a generated rectangle as prior knowledge 8a by A and B. The rectangular area A has an optimal shape for a short time frequency component of the middle and high frequency components in the latter half of the time interval. Further, the rectangular area B has an optimum shape for a case where a time frequency component having a long duration is generated in an intermediate frequency band in a time interval before the measurement time. Here, by increasing the number of divisions n and m, it is possible to express the boundary of the region in more detail. By the way, in the first 1/6 time section and the last 6/6 time section in the grid-like divided area, the operation speed of the inspection target is slower than the rated speed, so that the operation sound is not sufficiently generated. It is also possible to exclude it from the generation of region candidates. Further, in the above, for the entire region of the time frequency distribution, the entire time interval T is divided into n and the entire band F is divided into m to obtain a lattice-shaped divided region, and an arbitrary lattice line is set as an edge. Although an example of generating a rectangular region has been described, a region having an arbitrary shape by combining whether or not to select the above-described grid-like divided region as an optimal shape based on prior knowledge of the time-frequency component of the abnormal component May also be generated.

The evaluation unit 11 calculates the condensation degree E (R) 12 for the region candidate 10 (hereinafter, the region candidate is represented by R) (step S7).
The degree of condensation is E (R), y (t, f) for the time frequency distribution during the test, x (t, f) for the normal time frequency distribution, and R = [t1, t2, f1, f2] for the rectangular area. Then, the degree of condensation for these is obtained by the calculation shown in Equation 1. Here, t1, t2, f1, and f2 are a lower limit time, an upper limit time, a lower limit frequency, and an upper limit frequency of the rectangular region R, respectively. Further, the region candidate R other than the rectangle is obtained by the calculation shown in the more general expression 2 instead of the expression 1. Here, the symbol (t, f) ∈ R * means that the sum of the combinations of the discrete time t and the discrete frequency f included in the extraction region R * is taken.

  In the above equation, n is the number of spectral value samples included in the rectangular region of the time frequency distribution. Further, w (n) is a weighting factor corresponding to the number of samples n, and is, for example, the p-th root of the number of samples n (p is 2 for example). The number of samples n increases with the size of the region, and the weighting factor w (n) increases with the size of the region. Therefore, the degree of condensation E (R) for a small region is small, and the locality is small. It is used to mitigate the influence of existing outliers on the calculation results. Also, the function φ is a Box-Cox transformation (also referred to as a generalized logarithmic transformation) or a logarithmic transformation in order to nonlinearly transform the spectrum value and bring the distribution of the transformed value closer to a normal distribution. The Box-Cox transformation is expressed by Equation 3 and coincides with the logarithmic transformation when the parameter γ is γ = 0.

  The region extraction unit 13 examines the relationship between each region candidate and the degree of condensation E (R) with respect to each region candidate, and selects and outputs a region candidate having the largest value of the degree of condensation E (R) as the optimum extraction region. (Step S8). Each region candidate is {R1, R2,..., Rk}, each degree of condensation is {E (R1), E (R2),..., E (Rk)}, and the most suitable region candidate is R *. Is obtained by the calculation of Equation 4. Here, the natural number k is the number of region candidates.

  The degree-of-abnormality calculation unit 15 calculates the degree of abnormality from the spectrum values included in the respective optimum extraction regions R * of the normal time frequency distribution x (t, f) and the test time frequency distribution y (t, f). (Step S9). Now, when the extraction area R * is a rectangular area, R * = [t1, t2, f1, f2], and the degree of abnormality is a (R *), the degree of abnormality a (R *) is calculated by the calculation of Equation 5. This is the numerical value obtained. Here, t1, t2, f1, and f2 are as already defined.

In Equation 5 above, Ψ (x) is a nonlinear mapping function of the variable x, and the above-described Box-Cox transformation or the like can be used, for example. g (t) is a value obtained by dividing the cumulative value of the region R * in the frequency f direction by the number of unit frequencies, that is, a sample average regarding the frequency at the time t, and h (f) is the time t direction of the region R *. Is a value obtained by dividing the accumulated value by the number of unit times, that is, a sample average with respect to time at the frequency f. Further, g to (t) and h to (f) are values obtained by smoothing g (t) with respect to time t and h (f) with respect to frequency f, respectively. Smoothing is achieved, for example, by obtaining a moving average. Finally, the degree of abnormality a (R *) is either the maximum value of the time t from g to (t) after the moving average or the maximum value of the frequency from h to (f) after the moving average. Asking. A quantile that is a statistic may be used instead of the maximum value, and either value may be used as the degree of abnormality. Such an example is shown in a ₁ (R *), a ₂ (R *), a ₃ (R *), a ₄ (R *), a ₅ (R *), etc. in Formula 6. Here, quantile ({x}, α) represents the α quantile of the sequence {x}. If α is set to 1, it matches the maximum value max {x}. Α and β in Expression 6 may be close to 1, for example, 0.9.

Further, as another simpler method, the degree of abnormality a (R *) is calculated by mapping the normal time frequency distribution Ψ (x (t (t)) in the extraction region R * as indicated by a ₆ (R *) in Expression 7. , F)) and the average value of the map Ψ (y (t, f)) of the time frequency distribution during the test in the extraction region R *.

  The determination unit 17 compares the degree of abnormality a (R *) with a threshold value, determines that there is a possibility that an abnormal sound is generated when the degree of abnormality is equal to or greater than the threshold value, and sets “alarm” as the determination result 18. Output (step S10). If the degree of abnormality is less than the threshold, it is determined that there is a low possibility that abnormal sound has occurred, and “normal” is output as the determination result 18.

In the above embodiment, the time-frequency analysis unit 4 is configured to output the time-frequency distribution 5 by FFT calculation, but is not limited to FFT, and wavelet transform may be used.
Further, for the rectangular area of the prior knowledge 8a stored in the prior knowledge storage unit 8, a lower limit tmin may be provided for the difference t2-t1 between the upper limit time t2 and the lower limit time t1. That is, it is stored in the prior knowledge storage unit 8 only in the rectangular area where t2−t1 ≧ tmin.
Similarly, a lower limit fmin may be provided for the difference f2-f1 between the upper limit frequency f2 and the lower limit frequency f1. That is, it is stored in the table 8 only in the rectangular area where f2−f1 ≧ fmin.
Furthermore, the nonlinear function may be a function having nonlinear characteristics by broken line approximation in addition to an analytical function.

  As described above, according to the present invention, by providing means for extracting a time frequency region that is continuously formed with respect to the time frequency from the time frequency distribution, various bandwidths and durations appearing in the frequency analysis result can be reduced. It is not necessary to register a dedicated diagnostic procedure for the abnormal sound component, and the effect is that it can be diagnosed equally accurately.

  In addition, a region candidate generation unit that generates a region candidate, an evaluation unit that evaluates the goodness (condensation degree) of the generated region candidate, and a unit that selects a region having the highest goodness (condensation degree) are used. Thus, the optimum region having the best evaluation value is extracted from all the region candidates generated by the region candidate generating unit without using various threshold values specialized for the appearance pattern of the specific abnormal sound component. There is an effect. Thereby, there is an effect that abnormal sound components having various bandwidths and durations appearing in the frequency analysis result can be diagnosed equally accurately with no need to register a dedicated diagnostic procedure.

  In addition, as the goodness of the candidate area (condensation degree), by multiplying the amount of variation from the normal time frequency distribution by the number corresponding to the number of samples as a weight, the smaller the number of samples, the smaller the weight and the number of samples The larger the weight, the smaller the weight. Therefore, if the amount of mutation is the same, the region with the largest number of samples to be extracted is selected as much as possible (equivalently, the area of the region is large). Even if the amount is large, there is an effect that the goodness (condensation degree) of the region where the number of samples is small (equivalently the area of the region is small) becomes small. As a result, a region in which both the amount of mutation and the size of the sample are large in a balanced manner is extracted, which has the effect of improving the accuracy of diagnosis.

  In addition, when the sample average is used as the characteristic parameter of the distribution compared with normal, a meaningful result is obtained when the distribution of the sample follows a normal distribution, but the actual spectrum value has a non-negative asymmetric distribution. Therefore, non-linear transformation has the effect of bringing the distribution closer to the normal distribution, and makes it possible to make a meaningful comparison using the sample average. Thereby, the goodness of the area (condensation degree) can be appropriately evaluated, and as a result, the possibility of abnormal noise can be determined based on the appropriately extracted area, so that the accuracy of diagnosis is improved.

  In addition, by providing a non-linear characteristic (compression characteristic) with respect to the number of samples as a number according to the number of samples included in the region candidate as a parameter for determining the degree of condensation, the number of samples in the region (equivalent To prevent the area from becoming excessively large. Thereby, a balanced region having a large variation and a large number of samples can be extracted as the extraction region, and as a result, there is an effect that the accuracy of diagnosis of the determination result based on this can be improved.

  Further, by limiting the shape of the candidate area to be generated to a rectangle, generally, it works so as not to mistakenly extract an area having a shape other than the rectangle that is assumed to be impossible. As a result, an appropriate region can be extracted as the extraction region, and as a result, the accuracy of diagnosis of the determination result based on this can be improved.

  Similarly, by limiting the shape of the region using the prior knowledge regarding the temporal frequency distribution of the fluctuation component, the region that does not exist in the prior knowledge is prevented from being erroneously extracted. As a result, an appropriate region can be extracted as the extraction region, and as a result, the accuracy of diagnosis of the determination result based on this can be improved.

  Similarly, by limiting the shape of the region using prior knowledge about the operating state of the device, the region that does not exist in the prior knowledge is prevented from being erroneously extracted. As a result, an appropriate region can be extracted as the extraction region, and as a result, the accuracy of diagnosis of the determination result based on this can be improved.

  The abnormal sound diagnosis apparatus of the present invention may be used as a system apparatus that is a combination of a plurality of devices, for example, a detection apparatus that detects the location of an abnormal state in an elevator.

  DESCRIPTION OF SYMBOLS 1; Measurement signal, 2; Waveform acquisition part, 3; Waveform data, 4; Time frequency analysis part, 5; Time frequency distribution, 6: Time frequency distribution memory part at normal time, 7: Time frequency distribution memory part at test time, 8 Prior knowledge storage unit, 9; region candidate generation unit, 10; region candidate, 11; evaluation unit, 12; condensation degree, 13; region extraction unit, 15;

Claims (5)

Waveform data acquisition means for capturing waveform data of sound or vibration generated by the device to be inspected;
Time frequency analysis of the waveform data, time frequency analysis means for obtaining a time frequency distribution with one axis as a time axis and the other axis as a frequency axis;
Prior knowledge specifying the shape of the appearance region in the time frequency distribution of the abnormal sound component generated from the device obtained by analyzing the inspection target device in advance, which is defined by the time axis of the time frequency distribution and the coordinate value of the frequency axis. A region extracting means for generating a plurality of regions based on the above and extracting a region including a variation component different from the steady state of the time frequency distribution;
Determination means for determining and outputting an abnormality based on the time-frequency distribution included in the extraction region, and
The region extracting means includes
A region candidate generation unit that extracts a region including a variation component different from the steady state of the time frequency distribution as a region candidate;
An evaluation unit that obtains the degree of condensation from the relationship between the time frequency distribution included in the region candidate and the time frequency distribution at normal time;
An abnormal sound diagnosis apparatus comprising: a region candidate having a high degree of condensation as an extraction region . The evaluation unit nonlinearly converts the time frequency distribution included in the region candidate to the degree of condensation, and the characteristic parameter of the non-linearly converted time frequency distribution and the characteristic parameter of the normal time frequency distribution that is also non-linearly converted The abnormal sound diagnosis apparatus according to claim 1 , wherein the abnormal sound diagnosis apparatus is obtained by calculation with a number corresponding to the number of samples included in the region candidate. The abnormal sound diagnosis apparatus according to claim 2 , wherein the non-linear conversion for the evaluation unit to obtain the degree of condensation uses a conversion function having non-linear characteristics with respect to intensity. The number according to the number of samples included in the region candidate for calculating the degree of condensation by the evaluation unit is a number obtained by applying a function having nonlinear characteristics to the number of samples to the number of samples. The abnormal sound diagnosis apparatus according to claim 2 or claim 3 . A prior knowledge storage unit in which the prior knowledge is stored as a table;
Said region extraction means, abnormal sound diagnostic apparatus according to any one of claims 1 to 4, characterized in that to produce on the basis of the rectangular area candidates of the table stored in the pre-knowledge storage unit .

Images