Rolling bearing abnormal diagnosis method and computer program for abnormal diagnosis

Abstract

To propose an abnormality diagnosis method which can conduct measurement contactlessly and rapidly, and which can detect abnormality of a bearing even in cases when an S / N ratio of a bearing damage signal is low and in cases when bearing damage is at an initial stage.SOLUTION: A method for diagnosing abnormality of a bearing includes: an operation sound acquisition step ST1 of acquiring an operation sound signal of a bearing; a first resonance frequency estimation step ST2 of calculating kurtoses of respective envelope spectra of operation sound signals of respective bandwidths obtained by dividing the operation sound signal by a filter bank, and estimating a resonance frequency by using a kurtosis index including a maximum kurtosis from among the calculated kurtoses; a second resonance frequency estimation step ST3 of estimating the resonance frequency by using a spectrum correlation in which the operation sound signal is represented by a contamination frequency and a vibration frequency distribution, from the operation sound signal; and an abnormality determination step ST4 of comparing the estimated resonance frequencies with each other, and determining a case considered to match each other to be an abnormal case, meanwhile, determining a case not considered to match each other to be a normal case.SELECTED DRAWING: Figure 4

Description

Technical Field

[0001] The present invention relates to a method for diagnosing abnormalities in rolling bearings and a computer program for diagnosing abnormalities in rolling bearings.

Background Art

[0002] Rolling bearings (hereinafter simply referred to as "bearings") are very important mechanical elements incorporated in various rotating machines such as turbines, fans, and pumps. Since these machines rotate at high speeds, a bearing failure may lead to a major accident. For this reason, various abnormality diagnosis methods using temperature, wear powder, AE (Acoustic Emission), vibration, sound, etc. have been proposed.

[0003] Currently, the diagnosis method using vibration is common, but this diagnosis method requires contact with an acceleration sensor, so it is difficult to diagnose in situations where contact is difficult, and there are also problems such as the man-hours for attaching and detaching the acceleration sensor and the time required for measurement.

[0004] By the way, as characteristics of the operating sound and vibration waveform when the bearing is damaged, a periodic non-stationary waveform generated when the rolling element passes through the damaged part is included. This non-stationary waveform has the characteristic that it contains many resonance frequency components of the machine housing in which the bearing is incorporated.

[0005] As an analysis method for diagnosing bearing abnormalities by focusing on the periodicity of the vibration waveform, the envelope analysis proposed in Non-Patent Document 1 is widely used. Since the envelope analysis is a method using the envelope (envelope curve) of the signal waveform, there are problems such that periodicity cannot be detected when the non-stationary waveform is buried in broadband background noise or when the non-stationary waveform excited by a small degree of damage is weak. In this case, it is effective to extract the non-stationary waveform by applying a band-pass filter in the frequency band including the resonance frequency, but it is difficult to discriminate the resonance frequency and estimate the optimum center frequency and bandwidth.

[0006] On the other hand, extraction methods for non-stationary waveforms have also been widely studied. In Non-Patent Document 2, spectral kurtosis based on the kurtosis index of a waveform has been proposed, and its effectiveness has been shown.

[0007] In addition, in Non-Patent Document 3, a cartogram that generalizes spectral kurtosis and has a reduced computational cost, namely a high-speed cartogram, has been proposed. The high-speed cartogram is an analytical method for estimating the central frequency and frequency resolution suitable for extracting non-stationary waveforms. However, since the high-speed cartogram has a stronger sensitivity to aperiodic signals than to periodic signals, its effectiveness is limited in the extraction of periodic non-stationary waveforms.

[0008] In addition, in Non-Patent Document 4, as a method for extracting periodic non-stationary waveforms, "Protrugram" has been proposed, which uses the kurtosis of the envelope spectrum as an evaluation index instead of the kurtosis of the time-series signal. However, since Protrugram uses an empirically selected width for the bandwidth of the band-pass filter, there are problems in the design of the bandwidth.

[0009] In addition, in Non-Patent Document 5, as another method focusing on periodic non-stationary waveforms, spectral correlation has been proposed. Spectral correlation can analyze the spectral frequency components containing the frequency components of the envelope of a signal from the frequency representation of the autocorrelation function. Therefore, the dominant spectral frequency of the non-stationary waveform can be estimated from the frequency of the envelope of the non-stationary waveform. However, spectral correlation is an estimation method with low estimation accuracy or high computational cost, and there are problems in practical applications.

Prior Art Documents

Non-Patent Documents

[0010]

Non-Patent Document 1

[0011] As described above, in the currently common diagnostic method using vibration, it is difficult to diagnose in a situation where contact is difficult, and there are problems such as the time and labor required for attaching and detaching the acceleration sensor and the measurement taking a long time. Further, in the bearing abnormality diagnosis, when the S / N ratio of the bearing damage signal is low or when the bearing damage is in the initial stage, it is necessary to extract the non-stationary waveform, but a definitive estimation method has not yet been established for the method of extracting the non-stationary waveform. From the viewpoint of preventing accidents caused by bearings, it is urgent to establish a frequency analysis method suitable for extracting the non-stationary waveform effective for the bearing damage signal and a bearing abnormality diagnosis method. In addition, in the bearing abnormality diagnosis, it is more preferable if the degree of abnormality related to the bearing damage can be quantitatively estimated.

[0012] In view of such points, an object of the present invention is to propose an abnormality diagnosis method and a computer program for bearing abnormality diagnosis that can perform non-contact and rapid measurement and can detect bearing abnormalities even when the S / N ratio of the bearing damage signal is low or when the bearing damage is in the initial stage. Another object of the present invention is to propose an abnormality diagnosis method and a computer program for bearing abnormality diagnosis that can quantitatively estimate the degree of abnormality related to bearing damage.

Means for Solving the Problems

[0013] In order to solve the above problems, the bearing abnormality diagnosis method according to the present invention is a rolling bearing abnormality diagnosis method for diagnosing an abnormality due to damage of a rolling bearing, an operating sound acquisition step of acquiring an operating sound signal of the rolling bearing, a first resonance frequency estimation step of calculating the kurtosis of the envelope spectrum of the operating sound signal in each band obtained by dividing the operating sound signal acquired in the operating sound acquisition step by a filter bank arranged without gaps from low frequency to high frequency by a band-pass filter with a predetermined bandwidth, and estimating the resonance frequency using a kurtosis index having the largest kurtosis among the calculated kurtoses; A second resonance frequency estimation step of estimating a resonance frequency using spectral correlation obtained by expressing the operation sound signal obtained in the operation sound acquisition step in terms of a mixing frequency and a vibration frequency distribution; An abnormality determination step of comparing the respective resonance frequencies estimated in the first and second resonance frequency estimation steps, determining abnormality if they can be regarded as matching, and determining normality if they cannot be regarded as matching; It is characterized by including the above.

[0014] In this bearing abnormality diagnosis method, since the signal used for abnormality diagnosis is the operation sound signal acquired in the operation sound acquisition step, non-contact and rapid measurement can be performed. Also, in the abnormality determination step, since the logical product of the respective resonance frequencies estimated in the first and second resonance frequency estimation steps is taken to determine abnormality, the reliability of the diagnosis result is high. That is, in the first resonance frequency estimation step, the resonance frequency is estimated from the calculated value as a relative quantity using the kurtosis index, and in the second resonance frequency estimation step, the resonance frequency is estimated from the calculated value as an absolute quantity using spectral correlation. The resonance frequency is identified when the same result is obtained even with completely different methods, and the accuracy of identification is high. Also, in this bearing abnormality diagnosis method, in the first resonance frequency estimation step, the kurtosis of the envelope spectrum of the operation sound signal in each band divided by a filter bank arranged without gaps from the low frequency range to the high frequency range by a band-pass filter with a predetermined bandwidth is calculated. The envelope of the operation sound signal preprocessed by the band-pass filter is free from the influence of non-periodic noise components, and the periodic noise components are detected cleanly. As a result, according to this bearing abnormality diagnosis method, even when the S / N ratio of the bearing damage signal is low or when the bearing damage is in the initial stage, the non-stationary waveform of the bearing damage can be extracted using a frequency analysis method, and the abnormality of the bearing can be detected.

[0015] In the first resonance frequency estimation step, it is preferable to estimate the resonance frequency using a spectrogram that interprets the short-time Fourier transform for analyzing the time change of the frequency as a filter bank and calculates the kurtosis of the envelope spectrum of the operating sound signal for each time window length of the short-time Fourier transform. As this spectrogram, it is more preferable to use a high-speed spectrogram based on a dendritic multirate filter bank structure composed of filters with various passbands to speed up the calculation.

[0016] In the second resonance frequency estimation step, it is preferable to use the high-speed spectral correlation represented by the following Equation 1 as the spectral correlation. TIFF0007713696000001.tif141170

[0017] Further, in the bearing abnormality diagnosis method according to the present invention, When an abnormality is determined in the abnormality determination step, an optimal band-pass filter having a center frequency corresponding to the resonance frequency regarded as matching and an optimal bandwidth set according to the S / N ratio of the harmonic components centered on the center frequency is applied to the operating sound signal; an optimal band-pass filter application step, an optimal bandwidth envelope spectrum calculation step of calculating the envelope spectrum of the operating sound signal to which the band-pass filter having the optimal bandwidth is applied in the optimal band-pass filter application step; a foreign frequency amplitude value acquisition step of performing frequency analysis on the envelope spectrum calculated in the optimal bandwidth envelope spectrum calculation step to obtain the amplitude value of the foreign frequency; a quantitative diagnosis step of estimating the degree of abnormality of the rolling bearing from the magnitude of the amplitude value of the foreign frequency obtained in the foreign frequency amplitude value acquisition step and performing quantitative diagnosis; and further includes.

[0018] In this bearing abnormality diagnosis method, in the optimal band-pass filter application step, a band-pass filter having an optimal bandwidth set according to the signal-to-noise ratio of the harmonic components is applied to the operating sound signal. In the optimal bandwidth envelope spectrum calculation step, the envelope spectrum of the operating sound signal to which the band-pass filter having the optimal bandwidth is applied is calculated. In the mixing frequency amplitude value acquisition step, the envelope spectrum calculated in the optimal bandwidth envelope spectrum calculation step is frequency-analyzed to obtain the amplitude value of the mixing frequency. Then, in the quantitative diagnosis step, the degree of bearing abnormality is estimated from the magnitude of the amplitude value. According to this bearing abnormality diagnosis method, by using a band-pass filter having an optimal bandwidth set according to the signal-to-noise ratio, it is possible to estimate bearing abnormalities according to the purpose, such as when it is desired to identify the damage location or when it is desired to identify the type of damage. Further, according to this bearing abnormality diagnosis method, since the larger the degree of damage, the larger this amplitude value, the degree of bearing abnormality can be estimated from the magnitude of the amplitude value, so that the degree of bearing abnormality can be quantitatively diagnosed.

[0019] In the quantitative diagnosis step, it is preferable to focus only on the fundamental harmonic component of the mixing frequency and perform quantitative diagnosis. Regarding the fundamental harmonic component, it is a component that is always included regardless of the bandwidth determined as the optimal bandwidth, so it can be compared fairly.

[0020] The computer program for bearing abnormality diagnosis according to the present invention is a computer program for bearing abnormality diagnosis that is installed in a computer device to which a voice signal acquisition device is connected and causes the computer device to diagnose abnormalities due to damage to a rolling bearing, and is characterized in that the computer device executes the above-described bearing abnormality diagnosis method of the present invention.

[0021] According to this computer program for bearing abnormality diagnosis, since the above-described bearing abnormality diagnosis method can be executed on a computer device to which a voice signal acquisition device is connected, the same operations and effects as those of the above-described bearing abnormality diagnosis method can be obtained.

Effects of the Invention

[0022] According to the bearing abnormality diagnosis method of the present invention, non-contact and rapid measurement can be performed, and bearing abnormalities can be detected even when the S / N ratio of the bearing damage signal is low or when the bearing damage is in the initial stage. Also, according to the computer program for bearing abnormality diagnosis of the present invention, non-contact and rapid measurement can be performed by a computer device, and bearing abnormalities can be detected by the computer device even when the S / N ratio of the bearing damage signal is low or when the bearing damage is in the initial stage.

Brief Description of the Drawings

[0023]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Embodiment for Carrying Out the Invention

[0024] Hereinafter, with reference to the drawings, an abnormal diagnosis method for a bearing as an embodiment to which the present invention is applied will be described. First, the damage of the bearing that is the object of abnormal diagnosis, the envelope analysis which is basically used when analyzing the mixing frequency due to the damage in the present invention, and the pre-processing by a band-pass filter for detecting the envelope (also referred to as an envelope curve) cleanly in the envelope analysis will be described, and then the abnormal diagnosis method for the bearing of the embodiment will be described. Note that each figure does not necessarily reflect the actual state and all procedures strictly.

[0025] (Regarding the damage of the bearing) A rolling bearing is mainly composed of four parts: an inner ring, an outer ring, a cage, and rolling elements. When damage occurs to any of these four parts, an impact occurs each time the rolling element contacts the damaged area, exciting the vibration of the resonance frequency components of the machine housing. This impact occurs at regular time intervals when the rotational speed does not fluctuate. This time interval period or frequency is referred to as the contamination frequency in this specification. The contamination frequency is determined by four types of frequencies corresponding to the damaged part, specifically, the fundamental cage frequency (ETF), the ball spin frequency (BSF), the ball pass frequency of the outer ring raceway surface (BPFO), and the ball pass frequency of the inner ring raceway surface (BPFI). The contamination frequency can theoretically be calculated from the bearing specifications and the rotational speed. However, when actually acquiring the operating sound signal (time series signal) of a device incorporating the bearing, it is difficult to confirm the periodic transient signal of this contamination frequency from the operating sound signal. The reason for this is thought to be that the periodic transient signal due to bearing damage is buried in mechanical operating sounds such as motor operating sounds and background noise. In this case, it is effective to perform envelope analysis on the operating sound signal to confirm the periodic transient signal of the contamination frequency.

[0026] (Regarding preprocessing by envelope analysis and band - pass filter) FIG. 1 is a diagram showing the procedure of envelope analysis, FIG. 2 is a diagram showing the problem, and FIG. 3 is a diagram showing the effectiveness of preprocessing by a band-pass filter for solving the problem. In the abnormal diagnosis of a damaged bearing, there is an envelope analysis method that focuses on the periodicity of the bearing damage signal. In envelope analysis, as shown in FIG. 1, first, a time-series signal of the bearing is acquired (ST11). Next, in order to extract only the periodic information of the bearing damage signal, the envelope of the obtained vibration signal is calculated (ST12). Finally, by performing frequency analysis on the envelope, an envelope spectrum is obtained (ST13). Envelope analysis is an analysis method based on the premise that the bearing damage signal can be detected by the envelope. However, as shown in FIG. 2, in the case where the degree of bearing damage is small or the S / N ratio is small under conditions such as when the noise or background noise during operation is large, it becomes difficult to detect the mixed frequency. That is, for example, in the state where the degree of bearing damage is in the final stage, the S / N ratio becomes large, and it is highly possible to detect the mixed frequency in the envelope spectrum. However, in the state where the degree of bearing damage is in the initial stage, the S / N ratio becomes small, and the resonance frequency is often not detected in the envelope spectrum. In such a case, it is effective to perform envelope analysis after increasing the S / N ratio. Specifically, as shown in FIG. 3, as preprocessing before calculating the envelope (ST12), a band-pass filter with a determined center frequency and bandwidth is prepared (ST14), and the band-pass filter is applied to the time-series signal of the bearing (ST15). By this, the band-pass filter extracts an appropriate frequency band and increases the S / N ratio.

[0027] (Bearing Abnormal Diagnosis Method) FIG. 4 is a flowchart showing the procedure of the bearing abnormality diagnosis method according to the embodiment, FIG. 5 is a diagram for explaining the estimation of the resonance frequency by the kurtosis index, FIG. 6 is a diagram for explaining the envelope spectrum in the first resonance frequency estimation step ST2, FIG. 7 is a diagram showing the procedure for estimating the resonance frequency in the first resonance frequency estimation step, FIG. 8 is a diagram showing an example of result display by the method using the high-speed cartogram at that time, FIG. 9 is a diagram showing an example of result display by the evaluation method using the high-speed spectral correlation, FIG. 10 is a diagram for explaining the setting of the bandwidth focusing on the signal-to-noise ratio, FIG. 11 is a diagram showing the relationship between the signal-to-noise ratio and the bandwidth, FIG. 12 is a flowchart showing the procedure for setting the optimal bandwidth, and FIG. 13 is a diagram showing the relationship between the amplitude value of the mixing frequency and the degree of damage. With reference to these figures, the bearing abnormality diagnosis method according to the present invention will be described.

[0028] As shown in FIG. 4, the bearing abnormality diagnosis method according to the embodiment sequentially executes an operating sound acquisition step ST1, a first resonance frequency estimation step ST2, a second resonance frequency estimation step ST3, an abnormality determination step ST4, an optimal band-pass filter application step ST5, an optimal bandwidth envelope spectrum calculation step ST6, a mixing frequency amplitude value acquisition step ST7, and a quantitative diagnosis step ST8 to diagnose an abnormality due to bearing damage.

[0029] In the operating sound acquisition step ST1, an operating sound signal of the bearing is acquired using a microphone. In the bearing abnormality diagnosis method according to the embodiment, the operating sound signal acquired in this operating sound acquisition step ST1, that is, the time-series signal that can be acquired non-contact from the bearing, is analyzed in the following steps to diagnose the abnormality of the bearing.

[0030] In the first resonance frequency estimation step ST2, the resonance frequency is estimated based on the spectral kurtosis. Specifically, the operation sound signal acquired in the operation sound acquisition step ST1 is divided by a filter bank in which band-pass filters with a predetermined bandwidth are arranged without gaps from the low frequency range to the high frequency range, and the kurtosis of the envelope spectrum of each narrow-band operation sound signal is calculated. The resonance frequency is estimated using the kurtosis index having the largest kurtosis among the calculated kurtoses. At this time, the short-time Fourier transform (hereinafter referred to as "STFT") for analyzing the time change of the frequency is interpreted as a filter bank, and the resonance frequency is estimated using a cartogram that calculates the kurtosis of the envelope spectrum of the operation sound signal for each time window length of the short-time Fourier transform. As this cartogram, a high-speed cartogram with accelerated calculation is used based on a dendritic multirate filter bank structure composed of filters with various passbands. The method for estimating the resonance frequency using this kurtosis index will be described in detail below.

[0031] The kurtosis is represented by the following defining formula shown in Equation 3. TIFF0007713696000002.tif38170

[0032] In the method of estimating the resonance frequency using the kurtosis index, as shown in Fig. 5, the time-series signal is divided into narrow bands (divided by a time window) by a filter bank arranged without gaps from the low frequency range to the high frequency range by a band-pass filter with a predetermined bandwidth (Bw in Fig. 5). When comparing the time-series signals of each narrow band, it is based on the high possibility that the resonance frequency is included in the narrow band (the middle band in Fig. 5) where the amplitude that seems to have a large kurtosis is large. However, as shown in the lower band of Fig. 6, even in a band that does not include the resonance frequency, the amplitude may appear large due to the influence of non-periodic noise. Therefore, as shown in Fig. 6, by calculating the envelope spectrum of the time-series signal of each band, the periodic signal can be made to react strongly, and the influence of non-periodic noise can be removed. For this reason, in the first resonance frequency estimation step ST2, the kurtosis of the envelope spectrum of the operating sound signal is calculated. The kurtosis of the envelope spectrum of a time-series signal such as an operating sound signal is calculated by STFT that analyzes the time change of the frequency by performing spectrum analysis while sliding the time window.

[0033] In the method of estimating the resonance frequency using the kurtosis index, as shown in Fig. 7, it is preferable to develop it into a method using a spectrogram when estimating the resonance frequency. A spectrogram is a representation of spectral kurtosis in the (frequency / frequency resolution) plane and is a process of interpreting STFT as a filter bank. A spectrogram is a method of calculating the spectral kurtosis for each time window length of STFT using the interpretation of STFT as a filter bank. The method of using a spectrogram when estimating the resonance frequency, as shown in Fig. 7, is to acquire a time-series signal (ST21), divide it into narrow-band time-series signals using a multirate filter bank (ST22), perform envelope analysis on each narrow-band time-series signal (ST23), calculate the kurtosis of all envelope spectra (ST24), represent a spectrogram from the calculated kurtosis (ST25), and select the band with a large kurtosis by comparing the kurtosis (ST26).

[0034] Here, in the method using a cartogram, since the frequency resolution is investigated in detail, the calculation cost tends to be high. Therefore, in the first resonance frequency estimation step ST2, a high-speed cartogram that speeds up the calculation of the cartogram is used based on a dendritic multirate filter bank structure composed of filters with various passbands. In the first resonance frequency estimation step ST2, in the method of estimating the resonance frequency using the high-speed cartogram, the sharpness of two bands with the same bandwidth is compared, and the band with the larger sharpness is selected. At this time, the bandwidth is gradually narrowed from a wide bandwidth to a narrow bandwidth, and this selection is repeated until a very narrow bandwidth is finally reached. As shown in FIG. 8 (from top to bottom in FIG. 8), the bandwidth to be compared gradually narrows, and finally, the resonance frequency can be estimated as the remaining frequency.

[0035] In the second resonance frequency estimation step ST3, the resonance frequency is estimated using the spectral correlation that expresses the operation sound signal in terms of the mixing frequency and the vibration frequency distribution from the operation sound signal acquired in the operation sound acquisition step ST1. At this time, it is preferable to use a high-speed spectral correlation that enables high-speed calculation of the spectral correlation. The method of estimating the resonance frequency using this spectral correlation will be described in detail below.

[0036] The spectral correlation is the frequency representation of the autocorrelation function and can be expressed as in the following Equation 4. TIFF0007713696000003.tif60170

[0037] The estimation of the spectral correlation in signal processing is performed based on the above Equation 4, and a method based on the STFT that expresses the signal in terms of the mixing frequency - vibration frequency distribution is used. In the spectral correlation based on the STFT, first, time-frequency analysis of the signal is performed, and then correlation processing in the frequency domain is performed according to the above Equation 4.

[0038] Here, as an estimation method based on STFT, there is Averaged Cyclic Periodogram (hereinafter referred to as "ACP"). ACP is expressed by the following Equation 2, and this is called scanning vector correlation. TIFF0007713696000004.tif90170

[0039] Also, as an estimation method based on STFT, although different from spectral correlation, there is Cyclic Modulation Spectrum (hereinafter referred to as "CMS") which has the same purpose as spectral correlation in terms of analyzing a signal in the frequency-spectrum frequency representation of the envelope. CMS performs the frequency-spectrum frequency representation of the envelope by performing frequency analysis in the time direction with respect to the power of the signal obtained by time-frequency analysis, and effective calculation is performed by performing discrete Fourier transform (DFT) on the spectrogram in the time direction. ACP has high estimation accuracy but tends to have a high calculation cost. On the contrary, CMS has a low calculation cost but tends to have a low estimation accuracy. Therefore, in the second resonance frequency estimation step ST3, high-speed spectral correlation that takes advantage of ACP and CMS is used.

[0040] High-speed spectral correlation is a method that expands the concept of CMS and approaches ideal spectral correlation while maintaining a low calculation cost, and can be expressed as in the following Equation 1. High-speed spectral correlation can reduce the calculation cost while having statistical performance comparable to ACP. TIFF0007713696000005.tif141170

[0041] In the second resonance frequency estimation step ST3, when estimating the resonance frequency using high-speed spectral correlation, the evaluation is performed by spectral coherence in order to impart sensitivity due to periodicity. Spectral coherence is a function representing the strength of correlation in the frequency domain and is given by the following Equation 5. By spectral coherence, the spectral correlation is normalized from 0 to 1. Also, it has the characteristic of emphasizing signals that are periodic but have a low level, and can be made more sensitive to periodicity. TIFF0007713696000006.tif31170

[0042] In the second resonance frequency estimation step ST3, as shown in FIG. 9, on the frequency-spectral frequency plane of the envelope, the calculated spectral coherence is displayed with color-coding according to the strength of the correlation (FIG. 9 is a monochrome diagram and is not color-coded). In the second resonance frequency estimation step ST3, on this display, the frequency of the envelope at the location where the correlation is strong, that is, where the spectral coherence is large, is estimated as the resonance frequency.

[0043] In the abnormality determination step ST4, the respective resonance frequencies estimated in the first resonance frequency estimation step ST2 and the second resonance frequency estimation step ST3 are compared. If they can be regarded as matching, it is determined as abnormal; if they cannot be regarded as matching, it is determined as normal. Here, various variation factors such as variations in the measuring instrument, variations in measurement accuracy, and variations in the measurement environment are included in the estimated values of the resonance frequencies in the first resonance frequency estimation step ST2 and the second resonance frequency estimation step ST3. Therefore, in the determination in the abnormality determination step ST4 where it is determined that they can be regarded as matching, it is not required that the estimated values exactly match. Conditions for regarding them as matching considering these variations are appropriately set. For example, when the resonance frequency can be estimated by both, it may be set as a condition for regarding them as matching within a certain percentage of either estimated value, within a certain percentage of the average of the two estimated values, or within a range of the difference, etc.

[0044] In the optimal band - pass filter application step ST5, when it is determined as abnormal in the abnormality determination step ST4, a band - pass filter having a center frequency corresponding to the resonance frequency regarded as matching and an optimal bandwidth set according to the S / N ratio of the harmonic components centered on the center frequency is applied to the operating sound signal.

[0045] Here, the relationship between the bandwidth and the S / N ratio will be explained. As shown in FIGS. 10 and 11, when the set bandwidth of the band - pass filter is narrow, although the S / N ratio is high and it becomes easy to detect the mixing frequency, only the information of the mixing period is shown in the calculated envelope spectrum. On the other hand, when the set bandwidth of the band - pass filter is wide, although the S / N ratio is low and it becomes difficult to detect the mixing frequency, not only the information of the mixing period but also the information of the waveform shape is shown in the calculated envelope spectrum. That is, the harmonic components are considered to have information regarding the waveform shape of the bearing damage signal and the fluctuation of the collision time. When it is desired to specify the damage location of the bearing, it is advisable to set a high threshold value for the S / N ratio, and when it is desired to specify the type of bearing damage, it is advisable to set a low threshold value for the S / N ratio.

[0046] When diagnosing the bearing, it is preferable to select a bandwidth at which the harmonic components can be detected in order to obtain as much information as possible contained in the bearing damage signal. For this reason, it is preferable to extend the optimal bandwidth to the width at which a peak is seen in the high - frequency components.

[0047] As shown in FIG. 12, the specific procedure for setting the optimal bandwidth for diagnosing the bearing is as follows. First, a band - pass filter with a bandwidth shown in the following formula 6 is applied to the operating sound signal to calculate the envelope spectrum (ST31). TIFF0007713696000007.tif23170Next, according to formula 7, the S / N ratio in the calculated envelope spectrum is calculated (ST32). TIFF0007713696000008.tif25170 In addition, when calculating the S / N ratio (ST32), as shown on the waveform of the envelope spectrum in FIG. 10, while maintaining the analysis window length, it is slid according to the bandwidth. Thereby, it is possible to select a bandwidth in which a peak can be confirmed in the amplitude value of the nth harmonic, and a bandwidth that does not lose information on the harmonic components can be determined. Next, when the S / N ratio exceeds an arbitrarily determined threshold value, n is updated and the process returns to the first step (ST31) (ST33). On the other hand, when the S / N ratio does not exceed the arbitrarily determined threshold value, the optimal bandwidth is determined to be the bandwidth shown in the following Equation 8 (ST34). TIFF0007713696000009.tif23170

[0048] In the optimal bandwidth envelope spectrum calculation step ST6, as shown in FIG. 11, the envelope spectrum of the operating sound signal to which the band-pass filter having the optimal bandwidth is applied in the optimal band-pass filter application step ST5 is calculated.

[0049] In the interference frequency amplitude value acquisition step ST7, the envelope spectrum calculated in the optimal bandwidth envelope spectrum calculation step ST6 is frequency-analyzed to acquire the amplitude value of the interference frequency.

[0050] In the quantitative diagnosis step ST8, the degree of abnormality of the bearing is estimated from the magnitude of the amplitude value of the contaminating frequency obtained in the contaminating frequency amplitude value acquisition step ST7, and quantitative diagnosis is performed. More specifically, as shown in FIG. 13, the relationship between the amplitude value of the contaminating frequency and the degree of damage is such that the larger the degree of damage, the larger the amplitude value of the contaminating frequency. That is, in FIG. 13, the degree of damage is large from (a) to (f), but the amplitude value of the contaminating frequency appears large from (a) to (f). The same applies when only focusing on the first harmonic component of the contaminating frequency. In the quantitative diagnosis step ST8, it is preferable to estimate the degree of abnormality of the bearing by focusing only on the magnitude of the amplitude value of the first harmonic component. This is because the first harmonic component is a component that is always included regardless of the determined bandwidth, enabling a fair comparison.

[0051] As described above, in the bearing abnormality diagnosis method as an embodiment to which the present invention is applied, in the operating sound acquisition step ST1, an abnormality is diagnosed from the operating sound signal of the bearing acquired non - contact. Also, in the first resonance frequency estimation step ST2 and the second resonance frequency estimation step ST3, the resonance frequency is estimated using different analysis methods, and when the estimation results of both can be regarded as the same, it is determined that an abnormality has occurred. And when an abnormality has occurred, in the optimal band - pass filter application step ST5 and the optimal bandwidth envelope spectrum calculation step ST6, the state of the bearing abnormality is estimated. Further, in the contaminating frequency amplitude value acquisition step ST7 and the quantitative diagnosis step ST8, the degree of damage of the bearing is quantitatively estimated.

[0052] (System for diagnosing bearing abnormalities) The above - described bearing abnormality diagnosis method can be reflected in a bearing abnormality diagnosis system for diagnosing bearing abnormalities. An abnormality diagnosis system as an example is configured as a computer device to which an audio signal acquisition device is connected. By installing a bearing abnormality diagnosis computer program for executing various processes for performing the above - described bearing abnormality diagnosis method to which the present invention is applied in this computer device, a bearing abnormality diagnosis system can be constructed.

[0053] That is, the computer program for abnormality diagnosis first causes the computer device to execute an operating sound acquisition process for acquiring the operating sound signal of the rolling bearing. Next, the envelope spectrum sharpness of the operating sound signals in each band obtained by dividing the operating sound signal acquired in the operating sound acquisition process by a filter bank arranged without gaps from the low frequency range to the high frequency range by a band-pass filter with a predetermined bandwidth is calculated, and a first resonance frequency estimation process for estimating the resonance frequency using a sharpness index having the largest sharpness among the calculated sharpness values, and a second resonance frequency estimation process for estimating the resonance frequency using the spectral correlation representing the operating sound signal in terms of the mixing frequency and the vibration frequency distribution from the operating sound signal acquired in the operating sound acquisition process are executed. Next, the respective resonance frequencies estimated in the first and second resonance frequency estimation processes are compared, and an abnormality determination process is executed to determine abnormality if they can be regarded as matching, and to determine normality if they cannot be regarded as matching. Next, when it is determined as abnormal in the abnormality determination process, an optimum band-pass filter application process is executed to apply a band-pass filter having a center frequency corresponding to the resonance frequency regarded as matching and an optimum bandwidth set according to the S / N ratio of the harmonic components centered on the center frequency to the operating sound signal. Next, an optimum bandwidth envelope spectrum calculation process is executed to calculate the envelope spectrum of the operating sound signal to which the band-pass filter having the optimum bandwidth has been applied in the optimum band-pass filter application process. Next, a mixing frequency amplitude value acquisition process is executed to perform frequency analysis on the envelope spectrum calculated in the optimum bandwidth envelope spectrum calculation process to acquire the amplitude value of the mixing frequency. Next, a quantitative diagnosis process is executed to estimate the degree of abnormality of the rolling bearing from the magnitude of the amplitude value of the mixing frequency acquired in the mixing frequency amplitude value acquisition process and to perform quantitative diagnosis.

[0054] In the first resonance frequency estimation process, it is preferable to cause the computer device to estimate the resonance frequency using a kurtogram that calculates the kurtosis of the envelope spectrum of the operating sound signal for each time window length of the short-time Fourier transform by interpreting the short-time Fourier transform that analyzes the time change of the frequency as a filter bank. At this time, as the kurtogram, it is preferable to use a high-speed kurtogram with accelerated calculation based on a dendritic multirate filter bank structure composed of filters with various passbands.

[0055] In the second resonance frequency estimation process, it is preferable to use high-speed spectral correlation as the spectral correlation.

[0056] In the quantitative diagnosis process, it is preferable to cause the computer device to focus only on the first harmonic component of the mixing frequency and perform quantitative diagnosis.

[0057] (Function and Effect) In the bearing abnormality diagnosis method of the embodiment to which the present invention is applied, since the signal used for abnormality diagnosis is the operating sound signal acquired in the operating sound acquisition step ST1, non-contact and rapid measurement can be performed. Further, in the abnormality determination step ST4, since the logical product of the respective resonance frequencies estimated in the first resonance frequency estimation step ST2 and the second resonance frequency estimation step ST3 is taken to determine abnormality, the reliability of the diagnosis result is high. That is, in the first resonance frequency estimation step ST2, the resonance frequency is estimated from the calculated value as a relative amount using the kurtosis index, and in the second resonance frequency estimation step ST3, the resonance frequency is estimated from the calculated value as an absolute amount using spectral correlation. The resonance frequency is identified when the same results are obtained even with methods having completely different properties, and the accuracy of identification is high. Further, in this bearing abnormality diagnosis method, in the first resonance frequency estimation step ST2, the kurtosis of the envelope spectrum of the operating sound signal in each band divided by a filter bank arranged without gaps from the low frequency range to the high frequency range by a band-pass filter having a predetermined bandwidth is calculated. The envelope of the operating sound signal preprocessed by the band-pass filter is free from the influence of the aperiodic noise component, and the periodic noise component is detected cleanly. As a result, according to this bearing abnormality diagnosis method, even when the S / N ratio of the bearing damage signal is low or when the bearing damage is in the initial stage, the non-stationary waveform of the bearing damage can be extracted using a frequency analysis method, and the reliability is also improved to detect the bearing abnormality.

[0058] Also, in this bearing abnormality diagnosis method, in the optimal band-pass filter application step ST5, a band-pass filter having an optimal bandwidth set according to the S / N ratio of the harmonic components is applied to the operating sound signal. In the optimal bandwidth envelope spectrum calculation step ST6, the envelope spectrum of the operating sound signal to which the band-pass filter having the optimal bandwidth is applied is calculated. In the mixing frequency amplitude value acquisition step ST7, the envelope spectrum calculated in the optimal bandwidth envelope spectrum calculation step ST6 is frequency-analyzed to obtain the amplitude value of the mixing frequency. Moreover, in the quantitative diagnosis step ST8, the degree of bearing abnormality is estimated from the magnitude of the amplitude value. According to this bearing abnormality diagnosis method, by using a band-pass filter having an optimal bandwidth set according to the S / N ratio, when it is desired to identify the damage location, when it is desired to identify the type of damage, etc., the bearing abnormality can be estimated according to the purpose. Also, according to this bearing abnormality diagnosis method, since the amplitude value increases as the degree of damage increases, the degree of bearing abnormality can be estimated from the magnitude of the amplitude value, so that the degree of bearing abnormality can be quantitatively diagnosed.

[0059] According to the bearing abnormality diagnosis computer program of the embodiment to which the present invention is applied, since the bearing abnormality diagnosis method of the embodiment can be executed on a computer device to which an audio signal acquisition device is connected, the same operations and effects as those of the bearing abnormality diagnosis method of the embodiment can be obtained.

[0060] [Other Forms] As described above, the present invention has been described based on the above embodiments, but the present invention is not limited to the above embodiments. It can be implemented in various forms without departing from the gist thereof. For example, the following modifications are possible.

[0061] (1) The specific methods, mathematical formula forms, waveform and analysis result illustrations, etc. described in the above embodiments are examples and can be changed without impairing the effects of the present invention.

[0062] (2) In the above-described embodiments, the fast cartogram is used in the first resonance frequency estimation step ST2, and the fast spectral correlation is used in the second resonance frequency estimation step ST3. However, the present invention is not limited thereto. In the first resonance frequency estimation step ST2, the resonance frequency may be estimated from the kurtosis index of the envelope spectrum of the operating sound signal without using the fast cartogram. Further, in the second resonance frequency estimation step ST3, the resonance frequency may be estimated by spectral correlation without using the fast spectral correlation.

Example

[0063] As an example, two samples each of an actually normally operating bearing and an actually abnormal bearing were prepared, and for these, the operating sound acquisition step ST1 to the abnormality determination step ST4 of the bearing abnormality diagnosis method of the embodiment were executed and determined, and the results in Table 1 below were obtained. In the example, in the abnormality determination step ST4, the condition that the estimated value by the first resonance frequency estimation step ST2 and the estimated value by the second resonance frequency estimation step ST3 can be regarded as being the same is set as "when the resonance frequency can be estimated by both, if the difference is within 10% of the higher estimated resonance frequency, it is regarded as being the same."

Table 1

[0064] In the example, as shown in Table 1, for the two samples of normal products, although the resonance frequency (6400 Hz for one sample and 1700 Hz for the other sample) was estimated in the first resonance frequency estimation step ST2, in the second resonance frequency estimation step ST3, no peak that could be estimated as the resonance frequency was confirmed at the mixing frequency of 130 Hz in the cartogram, and the resonance frequency could not be estimated. Therefore, for the two samples of normal products, the estimated value by the first resonance frequency estimation step ST2 and the estimated value by the second resonance frequency estimation step ST3 could not be regarded as being the same, and they were determined to be normal in the abnormality determination step ST4.

[0065] On the other hand, for two abnormal product samples, in the first resonance frequency estimation step ST2, the resonance frequencies (both samples are 2600 Hz) were estimated. In the second resonance frequency estimation step ST3, peaks were confirmed at the mixing frequency of 130 Hz in the cartogram, and the resonance frequencies (both samples are 2500 Hz) were estimated. The difference between the two estimated values is 100 Hz, which is within 10% (3.8%) of the higher resonance frequency of 2600 Hz. Therefore, they are considered to be in agreement, and an abnormality was determined in the abnormality determination step ST4.

[0066] In the embodiment, it was confirmed that the abnormality of the bearing could be correctly determined.

Explanation of Signs

[0067] ST1... Operating sound acquisition step, ST2... First resonance frequency estimation step, ST3... Second resonance frequency estimation step, ST4... Abnormality determination step, ST5... Optimal band-pass filter application step, ST6... Optimal bandwidth envelope spectrum calculation step, ST7... Mixing frequency amplitude value acquisition step, ST8... Quantitative diagnosis step

Claims

1. A rolling bearing abnormality diagnosis method for diagnosing an abnormality due to damage to a rolling bearing, comprising: an operating sound acquisition step of acquiring an operating sound signal of the rolling bearing; a first resonance frequency estimation step of calculating the kurtosis of the envelope spectrum of the operating sound signal in each band obtained by dividing the operating sound signal acquired in the operating sound acquisition step by a filter bank arranged without gaps from low frequency to high frequency by a band-pass filter having a predetermined bandwidth, and estimating the resonance frequency using a kurtosis index having the largest kurtosis among the calculated kurtoses; a second resonance frequency estimation step of estimating the resonance frequency using the spectral correlation representing the operating sound signal in terms of the mixing frequency and the vibration frequency distribution from the operating sound signal acquired in the operating sound acquisition step; an abnormality determination step of comparing the respective resonance frequencies estimated in the first and second resonance frequency estimation steps, determining that there is an abnormality if they can be regarded as matching, and determining that it is normal if they cannot be regarded as matching; A rolling bearing abnormality diagnosis method characterized by including the above steps.

2. In the rolling bearing abnormality diagnosis method according to Claim 1, in the first resonance frequency estimation step, a short-time Fourier transform for analyzing the time change of the frequency is interpreted as the filter bank, and the resonance frequency is estimated using a spectrogram that calculates the kurtosis of the envelope spectrum of the operating sound signal for each time window length of the short-time Fourier transform.

3. In the rolling bearing abnormality diagnosis method according to Claim 2, as the spectrogram, a high-speed spectrogram based on a dendritic multirate filter bank structure composed of filters with various passbands and with calculation speed increased is used.

4. In the rolling bearing abnormality diagnosis method according to Claim 1, in the second resonance frequency estimation step, as the spectral correlation, a high-speed spectral correlation represented by the following Equation 1 is used.

5. In the rolling bearing abnormality diagnosis method according to any one of Claims 1 to 4, When it is determined as an abnormality in the abnormality determination step, an optimal band-pass filter having a center frequency corresponding to the resonance frequency regarded as being in agreement and an optimal bandwidth set according to the S / N ratio of harmonic components centered on the center frequency is applied to the operating sound signal; an optimal band-pass filter application step; an optimal bandwidth envelope spectrum calculation step of calculating an envelope spectrum of the operating sound signal to which the band-pass filter having the optimal bandwidth is applied in the optimal band-pass filter application step; a mixing frequency amplitude value acquisition step of performing frequency analysis on the envelope spectrum calculated in the optimal bandwidth envelope spectrum calculation step to obtain an amplitude value of the mixing frequency; a quantitative diagnosis step of estimating the degree of abnormality of the rolling bearing from the magnitude of the amplitude value of the mixing frequency obtained in the mixing frequency amplitude value acquisition step and performing quantitative diagnosis; A method for diagnosing an abnormality of a rolling bearing, further comprising:

6. In the method for diagnosing an abnormality of a rolling bearing according to claim 5, in the quantitative diagnosis step, paying attention only to the first harmonic component of the mixing frequency, a method for diagnosing an abnormality of a rolling bearing for performing the quantitative diagnosis.

7. A computer program for diagnosing an abnormality of a rolling bearing, which is installed in a computer device to which a voice signal acquisition device is connected, and causes the computer device to diagnose an abnormality due to damage to the rolling bearing, in the computer device, an operating sound acquisition process for acquiring an operating sound signal of the rolling bearing; calculating the kurtosis of the envelope spectrum of the operating sound signal in each band obtained by dividing the operating sound signal acquired in the operating sound acquisition process by a filter bank in which band-pass filters having a predetermined bandwidth are arranged without gaps from low frequency to high frequency, and estimating a resonance frequency using a kurtosis index having the largest kurtosis among the calculated kurtosis values; a first resonance frequency estimation process; a second resonance frequency estimation process of estimating a resonance frequency using the spectral correlation in which the operating sound signal acquired in the operating sound acquisition process is expressed in terms of a mixing frequency and a vibration frequency distribution; an abnormality determination process of comparing the respective resonance frequencies estimated in the first and second resonance frequency estimation processes, determining as an abnormality if they can be regarded as being in agreement, and determining as normal if they cannot be regarded as being in agreement; A computer program for diagnosing abnormalities in rolling bearings, characterized by causing the following to be executed. **Claim 8** In the computer program for diagnosing abnormalities in rolling bearings according to Claim 7, in the first resonance frequency estimation process, the computer device interprets the short-time Fourier transform for analyzing the time change of the frequency as the filter bank, and uses a kurtogram for calculating the kurtosis of the envelope spectrum of the operating sound signal for each time window length of the short-time Fourier transform to estimate the resonance frequency. A computer program for diagnosing abnormalities in rolling bearings. **Claim 9** In the computer program for diagnosing abnormalities in rolling bearings according to Claim 8, as the kurtogram, a computer program for diagnosing abnormalities in rolling bearings that uses a high-speed kurtogram with calculations speeded up based on a dendritic multirate filter bank structure composed of filters with various passbands. **Claim 10** In the computer program for diagnosing abnormalities in rolling bearings according to Claim 7, in the second resonance frequency estimation process, as the spectral correlation, a computer program for diagnosing abnormalities in rolling bearings that uses the high-speed spectral correlation represented by the following Equation 1. **Claim 11** In the computer program for diagnosing abnormalities in rolling bearings according to any one of Claims 7 to 10, further, the computer device when it is determined as abnormal in the abnormality determination process, an optimal band-pass filter application process for applying a band-pass filter having a center frequency corresponding to the resonance frequency regarded as matching and an optimal bandwidth set according to the S / N ratio of the harmonic components centered on the center frequency to the operating sound signal, an optimal bandwidth envelope spectrum calculation process for calculating the envelope spectrum of the operating sound signal to which the band-pass filter having the optimal bandwidth has been applied in the optimal band-pass filter application process, a mixing frequency amplitude value acquisition process for performing frequency analysis on the envelope spectrum calculated in the optimal bandwidth envelope spectrum calculation process to obtain the amplitude value of the mixing frequency, a quantitative diagnosis process for estimating the degree of abnormality of the rolling bearing from the magnitude of the amplitude value of the mixing frequency obtained in the mixing frequency amplitude value acquisition process and performing quantitative diagnosis. A computer program for diagnosing abnormalities in rolling bearings, characterized by causing the following to be executed. **Claim 12** In the computer program for diagnosing abnormalities of a rolling bearing according to claim 11, In the quantitative diagnosis process, the computer device is made to focus only on the primary harmonic component of the mixing frequency, and a computer program for diagnosing abnormalities of a rolling bearing that performs the quantitative diagnosis.

Images