JP5143863B2 - Bearing condition monitoring method and bearing condition monitoring apparatus - Google Patents

Description

  The present invention relates to a bearing state monitoring method and a bearing state monitoring device.

  When diagnosing abnormalities by monitoring the state of bearings in rotating machinery equipment, etc., for example, whether the amplitude of vibration or acoustic (Acoustic Emission: AE) measured by a sensor attached to the bearing exceeds a preset threshold In general, it is determined whether there is an abnormality. In the case of measuring AE, it is also known that the number of occurrences within a certain period of events in which the amplitude of AE exceeds a preset threshold is used as a monitoring index.

  When measuring vibration, for low-speed rotating equipment with a rotational speed generally lower than 100 rpm, the vibration intensity generated is weak even if the bearing is flawed, and it is difficult to reliably monitor the condition. (For example, see Non-Patent Document 1). Even in such a low-speed rotating facility, sufficient damage sensitivity can be obtained when AE is measured, but it is difficult to quantitatively relate the amplitude and the number of events to damage. It is necessary to accumulate sufficient data on-site every time (the number of continuous rotations and the period of intermittent rotations), and so-called spot measurement is difficult to diagnose.

  Noriaki Inoue, “Equipment Diagnosis by Practical Vibration Method to Answer Field Questions”, Japan Plant Maintenance Association, September 20, 1998, p.91-92

  It is an object of the present invention to make it possible to determine the state of a bearing by processing measurement data without setting a reference value based on data accumulation for each measurement site or operation condition.

In a first aspect of the present invention, a detection waveform is calculated by performing a detection process on a time waveform of a signal generated by rotation of the rotation shaft in a bearing that holds the rotation shaft, and the amplitude of the detection waveform is logarithmized. The amplitude distribution is calculated from the detected waveform obtained by logarithmizing the amplitude, a logarithmic amplitude distribution obtained by logarithmizing the frequency of the amplitude distribution is obtained, a mode value of the amplitude distribution is obtained, and the mode value of the amplitude distribution is obtained. An estimation composed of a low-amplitude side distribution obtained by approximating data having a lower amplitude than the normal amplitude by a normal distribution and a high-amplitude side distribution obtained by applying this low-amplitude side distribution to a higher amplitude side than the amplitude of the mode value Obtaining a normal distribution, obtaining a logarithmic estimated normal distribution obtained by logarithmizing the frequency of the estimated normal distribution, obtaining a reference distribution by adding a margin to the logarithmized estimated normal distribution, obtaining the logarithmized amplitude distribution and the reference distribution Jo of the bearing by comparison with Determining, providing a bearing condition monitoring methods.

  Since the bearing state is determined by comparing the amplitude distribution with the reference distribution obtained from the amplitude distribution, even when the rotational speed of the rotating shaft is relatively low (for example, about 0.1 to 200 rpm), it is determined for each measurement site and operating condition. It is possible to determine the state of the bearing by processing the detected waveform without setting a reference value by data accumulation.

It is possible to improve the reliability of the determination by adding margin to the number of estimated normal distribution pairs.

  More specifically, the approximation coefficient is obtained by approximating the data included in the logarithmic amplitude distribution whose logarithmization frequency exceeds the frequency of the same amplitude in the reference distribution with a polynomial. The approximate coefficient is used as an index indicating the difference between the logarithmic amplitude distribution and the reference distribution in determining the state of the bearing. For example, the polynomial is a linear expression, and the gradient of this linear expression or the absolute value of the inverse thereof is used as the index.

  As an alternative, the index area calculated by the following equation for the logarithmic amplitude distribution data whose logarithmization frequency exceeds the same amplitude frequency of the reference distribution, the logarithmic amplitude distribution and the reference distribution of It is used to determine the state of the bearing as an index indicating the difference.

  In the detection processing, the number of sampling points per one rotation or intermittent operation of the rotation shaft in the detection process is set to a constant value regardless of the rotation speed of the rotation shaft or the speed of the intermittent operation, and the detection waveform in which the amplitude is logarithmically used is used. The frequency of the amplitude distribution calculated in the above is normalized by dividing by the constant value. By standardizing, the influence due to the difference in measurement site and operating conditions can be further reduced, and the state of the bearing can be determined with higher accuracy.

  If the vibration distribution is obtained while including the fluctuation component as disturbance noise, the width of the vibration distribution is determined by the fluctuation component instead of the signal to be originally measured (for example, AE), and the accuracy of determining the state of the bearing may be lowered. In such a case, the calculation of the amplitude distribution is performed by calculating a fluctuation component of the detection waveform, removing the fluctuation component from the detection waveform, and using the detection waveform after removing the fluctuation component. It is preferable to calculate the distribution. For the calculation of the fluctuation component, a moving average or a low-pass filter can be used.

  The signal generated by the rotation of the rotating shaft is either AE, vibration, or ultrasonic wave.

According to a second aspect of the present invention, a sensor unit that detects a signal generated by rotation of the rotary shaft in a bearing that holds the rotary shaft, and a detection waveform is calculated by performing detection processing on the measurement waveform detected by the sensor unit. A detection processing unit that performs logarithm of the amplitude of the detection waveform, calculates an amplitude distribution from the detection waveform obtained by logarithmizing the amplitude, obtains a mode value of the amplitude distribution, A low-amplitude distribution obtained by approximating data having a lower amplitude than the amplitude of the mode with a normal distribution, and a high-amplitude side distribution obtained by applying the low-amplitude side distribution to a higher amplitude than the amplitude of the mode. A reference waveform generation unit that obtains a configured estimated normal distribution, obtains a logarithmic estimated normal distribution obtained by logarithmizing the frequency of the estimated normal distribution, adds a margin to the logarithmized estimated normal distribution, and obtains a reference distribution ; The frequency of the amplitude distribution is By a phased the logarithmic amplitude distribution comparison of the reference distribution provides a bearing condition monitoring apparatus and a determination unit for determining status of said bearing.

  According to the present invention, since the state of the bearing is determined by comparing the amplitude distribution with the reference distribution obtained from the amplitude distribution, the bearing can be processed by the detection waveform without setting a reference value based on data accumulation for each measurement region and operating condition. The state can be determined.

The schematic diagram which shows the bearing state diagnostic apparatus concerning embodiment of this invention. The graph which shows the measurement waveform and detection waveform of AE. A graph showing an amplitude distribution and an estimated normal distribution. The graph which shows an amplitude distribution, a reference distribution, and an approximate curve. The graph which shows the amplitude distribution at the time of a normal bearing, a reference distribution, and an approximate line. The graph which shows the amplitude distribution at the time of bearing abnormality occurrence, a reference distribution, and an approximate line. The bar graph which shows the reciprocal number of the gradient of the approximate line in various measurement conditions. A graph which shows amplitude distribution, standard distribution, and index area. The bar graph which shows the parameter | index area in various measurement conditions. The graph which shows a detection waveform and a fluctuation component. The graph which shows the detection waveform which removed the fluctuation component.

Next, embodiments of the present invention will be described with reference to the accompanying drawings. The amplitude distribution of RF waveforms such as acoustic (AE), vibration, and ultrasonic waves when the bearing is normal is known to follow a normal distribution, but the present inventor has found the following knowledge. When the bearing is normal and the background noise is low, the amplitude distribution of the detected waveform can be approximated by a normal distribution. When the bearing is damaged, the amplitude distribution deviates from the normal distribution due to the appearance of high-amplitude data, but the amplitude distribution on the lower amplitude side than the mode amplitude can be approximated by the normal distribution. Furthermore, the amplitude distribution on the lower amplitude side than the amplitude of the mode value approximated by the normal distribution when the bearing is damaged is substantially the same as the normal distribution when the bearing is normal. In the present invention, these findings are applied to monitoring the state of a bearing.

  FIG. 1 shows a bearing state monitoring device (hereinafter, monitoring device) 1 according to an embodiment of the present invention. The bearing 3 supports a rotating shaft 2 of rotating machine equipment (a belt conveyor equipment in this embodiment, but the kind of equipment or machine is not particularly limited). The monitoring device 1 monitors the occurrence of abnormality due to wear and damage in the bearing 3.

  The monitoring device 1 includes an AE sensor 10 fixed to the bearing 3 via a coplant. Moreover, the monitoring apparatus 1 includes a filter 12, an amplifier 13, and a signal processing unit 21 that performs various arithmetic processes. The monitoring device 1 also includes a determination unit 22 that determines whether or not an abnormality is determined in the bearing 3 based on the processing result in the signal processing unit 21, and a monitor for displaying the determination result of the determination unit 22, for example. The display part 23 which is an apparatus is provided. Furthermore, the monitoring apparatus 1 includes a storage unit 24 that stores various data, calculation results, and the like in cooperation with the signal processing unit 21 and the determination unit 22. The signal processing unit 21 includes a detection processing unit 30, a sampling circuit 31, an amplitude distribution calculation unit 32, and a reference wave generation unit 33.

  Hereinafter, a bearing state monitoring method executed by the monitoring device 1 will be described.

  The AE sensor 10 detects an AE signal generated by the rotation of the rotary shaft 2 in the bearing 3. Instead of detecting the AE signal by the AE sensor 10, vibration generated when the rotating shaft 2 rotates may be detected by a vibration sensor. Moreover, you may detect the ultrasonic wave which generate | occur | produces at the time of rotation of the rotating shaft 2 with an ultrasonic sensor. The following processing can be similarly applied when detecting vibrations and ultrasonic waves.

  A measurement waveform (AE time waveform) from the AE sensor 10 is input to the signal processing unit 21 via a preamplifier, a filter 12, and an amplifier 13 (not shown). A weak output signal from the AE sensor 10 is first amplified by a preamplifier. The preamplifier may be provided in the AE sensor 10 or may be provided between the AE sensor 10 and the filter 12. The filter 12 removes noise from the preamplifier signal and passes only an appropriate frequency band. The signal that has passed through the filter 12 is amplified to an intensity suitable for processing by the signal processing unit 21 by the amplifier 13.

  The detection processing unit 30 performs detection processing on the measurement waveform (AE time waveform input from the amplifier 13) to calculate a detection waveform (see FIG. 2). The time length of this detection waveform has at least one rotation of the rotating shaft 2 (one cycle of intermittent operation when the rotating shaft 2 operates intermittently). For example, a measurement waveform of about 10 rotations of the rotating shaft 2 is obtained. The time length for one rotation of the rotating shaft 2 may be determined by the set number of rotations of the rotating shaft 2 or may be actually measured.

  The sampling circuit 31 performs sampling on the detection waveform from the detection processing unit 30.

  The amplitude distribution calculation unit 32 calculates the amplitude distribution by performing the following processing on the detected waveform after sampling. First, the amplitude of the detected waveform after sampling is logarithmized (natural logarithmization). Using the detection waveform obtained by logarithmizing the amplitude, the amplitude distribution (the distribution of the frequency at which the amplitude appears in the detection waveform) is calculated (see FIGS. 3 and 4). Since AE has a wide range of amplitude changes, it is possible to detect low amplitude changes with high sensitivity by logarithmically increasing the weight of information on the low amplitude side. Further, the logarithm makes a ratio difference (that is, a linear difference of 10 times becomes a logarithm becomes +1), so that the handling of the amplitude becomes easy.

  The frequency of the amplitude distribution may be normalized. In this case, the sampling frequency at which the sampling circuit 31 samples the measurement waveform is changed according to the number of rotations of the rotating shaft 2 (the number of operations per unit time in the case of intermittent operation), whereby one rotation of the rotating shaft 2 or The number N of sampling points per cycle of the intermittent operation is set to a constant value regardless of the rotation speed and the speed of the intermittent operation. Then, the frequency is normalized by dividing the frequency of the amplitude distribution by the number of sampling points N (a constant value). The vertical axis of the graphs of FIGS. 3 to 5B and 7 is the normalized frequency.

The reference distribution generation unit 33 obtains a reference distribution used for determining the state of the bearing. The reference distribution is obtained based on an estimated normal distribution that is an estimation of the amplitude distribution when the bearing 3 is normal. When the bearing 3 is damaged as described above, the amplitude distribution deviates from the normal distribution due to the appearance of high amplitude data, but the amplitude distribution on the lower amplitude side than the mode amplitude can be approximated by the normal distribution. The amplitude distribution on the low amplitude side is almost the same as the normal distribution when the bearing 3 is normal. Also, by approximating the amplitude distribution on the lower amplitude side than the amplitude of the mode value of the reference distribution with a normal distribution, the amplitude of the mode value is folded back at the boundary so that the amplitude of the mode value when the bearing is normal The amplitude distribution on the higher amplitude side can also be estimated. Based on this principle, the reference distribution generation unit 33 estimates the amplitude distribution when the bearing 3 is normal.

Hereinafter, a specific procedure in which the reference distribution generation unit 33 obtains the reference distribution will be described with reference to FIG. First, the mode value F _max of the amplitude distribution (the amplitude distribution after the frequency is normalized when the above-described normalization is performed) is obtained. Next, a low-amplitude distribution is obtained by approximating the data included in the amplitude distribution with a normal distribution having a lower amplitude than the amplitude of the mode _Fmax . Moreover, by folding the lower amplitude side distribution with an amplitude of the mode F _max, requiring a high amplitude side distribution is also obtained by estimating a distribution of high amplitude side than the amplitude of the mode F _max. A combination of the low amplitude side distribution and the high amplitude side distribution is the above-described estimated normal distribution. Next, a logarithmic estimated normal distribution obtained by logarithmizing the estimated normal distribution (natural logarithmization) is obtained (see also FIGS. 5A and 5B). A margin α for improving determination reliability by preventing erroneous determination is added to the frequency of the logarithmic estimated normal distribution. The distribution obtained by adding the margin α to the logarithmic estimated normal distribution is the reference distribution. For example, the margin α is set to a value (α = ln (2)) corresponding to twice the frequency of data included in the estimated normal distribution before logarithmization.

  On the other hand, the amplitude distribution calculation unit 32 obtains the amplitude distribution obtained by using the detection waveform obtained by logarithmizing the amplitude (natural logarithmization) as described above (after performing frequency normalization in the case of performing the normalization described above). A logarithmic amplitude distribution is calculated by further performing logarithmization (natural logarithmization) of the frequency of the amplitude distribution on the amplitude distribution) (see FIGS. 3 and 4). Since the frequency of abnormal AEs is much lower than that of background noise AEs, the difference between normal and abnormal is small when the frequency is viewed linearly. By logarithmizing the amplitude distribution, it is possible to relatively increase the weight of an AE caused by an abnormality that is infrequent, and to improve the sensitivity to the abnormality.

The determination unit 22 determines whether the bearing 3 is normal by comparing the logarithmized amplitude distribution calculated by the amplitude distribution calculation unit 32 and the reference distribution generated by the reference distribution generation unit 33 (an abnormality caused by wear or damage occurs). Is not determined). The determination of the state of the bearing 3 by the determination unit 22 is that the logarithmization frequency exceeds the frequency of the same amplitude of the reference waveform among the data included in the logarithmic amplitude distribution ( higher amplitude side than the amplitude of the mode _Fmax ). That is, it is performed by evaluating a region in the logarithmic amplitude distribution where the frequency exceeds the reference waveform. Specifically, there are the following two types of determination methods.

  The first determination method is as follows. An approximation coefficient is obtained by approximating the data included in the logarithmic amplitude distribution whose logarithmization frequency exceeds the frequency of the same amplitude in the reference distribution by a polynomial. Then, the state of the bearing 3 is determined using this approximation coefficient as an index indicating the difference between the logarithmic amplitude distribution and the reference distribution. For example, as shown in FIG. 4 to FIG. 5B, an approximate straight line obtained by approximating a data whose logarithmization frequency exceeds the frequency of the same amplitude of the reference distribution among data included in the logarithmic amplitude distribution by a linear function is obtained. The state of the bearing 3 is determined based on the absolute value of the reciprocal of the gradient. As is apparent from a comparison between FIG. 5A (during normal operation) and FIG. 5B (during abnormality), when an abnormality occurs in the bearing 3, the slope of the approximate straight line becomes gentler than that during normal operation. That is, when an abnormality occurs in the bearing 3, the absolute value of the reciprocal of the gradient is larger than that in the normal state. Accordingly, an appropriate threshold for the reciprocal of the gradient is set, and if the absolute value of the reciprocal of the gradient is equal to or less than the threshold, it is determined that the bearing 3 is normal, and if the absolute value exceeds the threshold, it is determined that an abnormality has occurred in the bearing 3. can do.

  FIG. 6 shows the absolute value of the reciprocal of the slope of the approximate curve obtained experimentally when the number of rotations of the bearing 3 is different (10 rpm, 20 rpm, 80 rpm, 100 rpm). In FIG. 6, Nos. 1 to 12 are cases where the bearing 3 is normal, and Nos. 13 to 16 are cases where an abnormality occurs in the bearing 3. The absolute value of the reciprocal of the slope of the approximate curve is approximately 1 when normal (No. 1 to 12), but is 2 or more when abnormal (No. 13 to 16), and there is a clear difference. . In the case of the example in FIG. 8, for example, it is conceivable to set the threshold value for determining whether or not the bearing 3 needs attention for occurrence of abnormality to 1.5. In this embodiment, since the frequency is standardized, a common threshold value can be used, and the threshold value is set by measuring the absolute value of the reciprocal of the slope of the approximate curve at normal and abnormal times at each measurement location. There is no need to do.

  When approximation by a linear function is adopted in the first determination method, the gradient itself or the absolute value of the gradient may be used as an index in determining the state of the bearing 3. Further, approximation by a quadratic or higher-order function may be performed instead of approximation by a linear function, and the coefficient of the obtained polynomial may be used as an index in state determination.

The second determination method is as follows. Referring to FIG. 7, the area (index area) of the region surrounded by the reference frequency F _ref and the logarithmization frequency among the data included in the logarithmic amplitude distribution that exceeds the frequency of the same amplitude in the reference distribution is _expressed as follows. Calculate with the formula.

The reference frequency F _ref used for calculating the index area AR is preferably smaller than the frequency corresponding to the situation in which a certain amplitude appears once in the detection waveform, and is not too small. For example, when the above-described normalization by the sampling point number N is not performed, the frequency when a waveform appears only once in the detected waveform is 1, so the reference frequency F _ref is less than ln (1) = 0. Is set to a value that is not too small (for example, -1). In addition, when normalization is performed with the frequency sampling point N, the frequency when the waveform appears once in the detected waveform is 1 / N, so the reference frequency F _ref is less than ln (1 / N). It is set to a value that is not too small (for example, the largest integer less than ln (1 / N)). When the number of sampling points N used for normalization is 10,000, since ln (1 / N) = − 9.2, the reference frequency F _ref is set to −10, for example.

  FIG. 8 shows the index area AR obtained experimentally when the number of rotations of the bearing 3 is different (10 rpm, 20 rpm, 80 rpm, 100 rpm). In FIG. 8, Nos. 1 to 12 are cases where the bearing 3 is normal, and Nos. 13 to 16 are cases where an abnormality occurs in the bearing 3. The value of the index area AR is less than 20 at the normal time (No. 1 to 12), but is 40 or more at the abnormal time (No. 13 to 16), and there is a clear difference. In the case of the example in FIG. 8, for example, it is conceivable to set the threshold value for determining whether or not the bearing 3 needs attention to occurrence of abnormality to 20. When comparing the frequency with the number of sampling points N, a common threshold can be used if the measurement systems are the same, and the threshold value is determined by measuring the normal and abnormal index areas AR at each measurement location. There is no need to set.

  The determination unit 22 may execute either one of the first determination method and the second determination method, or may execute both of these determination methods. For example, the determination unit 22 determines that an abnormality has occurred in the bearing 3 when the determination of occurrence of abnormality is established in both the first and second determination methods. However, the determination unit 22 uses the first and second determination methods. If the determination of occurrence of abnormality is established with only one of them, the bearing 3 may be determined to be normal.

  In the present embodiment, the state of the bearing 3 is determined by comparing the logarithmic amplitude distribution with a reference distribution calculated from the amplitude distribution itself before logarithmization. Therefore, even when the rotational speed of the rotating shaft is relatively low (for example, about 0.1 to 200 rpm), the detection waveform processing is performed without setting a reference value based on data accumulation for each measurement site and operating condition. The state can be determined.

  Since the logarithmic estimated normal distribution is obtained by approximation, the actual data varies around the logarithmic estimated normal distribution. For this reason, if the logarithmic estimated normal distribution is used as the reference distribution as it is, there is a possibility that the reference value may be exceeded even in a normal AE portion that is not an AE caused by damage on the high amplitude side that is originally desired due to variations in the actual amplitude distribution. Yes, leading to misdiagnosis. By using the reference distribution obtained by adding the margin α to the logarithmic estimated normal distribution, it is possible to prevent misdiagnosis due to variations in the distribution of actual data.

When the vibration distribution is obtained while including the fluctuation component as disturbance noise, the width of the vibration distribution is determined by the fluctuation component, not the signal to be originally measured (for example, AE), and the amplitude change due to the vibration component due to the damage is included therein. There is a risk of being buried. For this reason, the accuracy of determining the state of the bearing 3 may decrease due to the fluctuation component. In such a case, the amplitude distribution may be calculated after removing the fluctuation component. Specifically, first, the fluctuation component included in the detection waveform after logarithmizing the amplitude is calculated (see FIG. 9). Next, the fluctuation component is removed from the detected waveform (see FIG. 10). For example, a moving average or a low-pass filter can be used to calculate the fluctuation component.

DESCRIPTION OF SYMBOLS 1 Bearing state monitoring apparatus 2 Rotating shaft 3 Bearing 10 Acoustic sensor 11 Preamplifier 12 Filter 13 Amplifier 14 Sampling circuit 21 Signal processing part 22 Determination part 23 Display part 24 Storage part 31 Detection process part 32 Amplitude distribution calculation part 33 Reference distribution generation part

Claims (10)

In the bearing holding the rotary shaft, the detection waveform is calculated by performing detection processing on the time waveform of the signal generated by the rotation of the rotary shaft,
Logarithmically the amplitude of the detected waveform,
Calculating the amplitude distribution from the detected waveform obtained by logarithmizing the amplitude;
Obtaining a logarithmic amplitude distribution obtained by logarithmizing the frequency of the amplitude distribution;
A mode value of the amplitude distribution is obtained, a low-amplitude side distribution obtained by approximating data having a lower amplitude than the amplitude of the mode value of the amplitude distribution by a normal distribution, and the low-amplitude side distribution is represented by the amplitude of the mode value. An estimated normal distribution composed of a higher amplitude side distribution applied to a higher amplitude side than that, a logarithmic estimated normal distribution obtained by logarithmizing the frequency of the estimated normal distribution is obtained, and a margin is provided in the logarithmized estimated normal distribution. To obtain the reference distribution,
A bearing state monitoring method for determining a state of the bearing by comparing the logarithmic amplitude distribution and the reference distribution . Approximating the data included in the logarithmic amplitude distribution with a logarithmization frequency exceeding the frequency of the same amplitude of the reference distribution by a polynomial to obtain an approximation coefficient,
The bearing state monitoring method according to claim 1 , wherein the approximation coefficient is used for determining the state of the bearing as an index indicating a difference between the logarithmic amplitude distribution and the reference distribution. The bearing state monitoring method according to claim 2 , wherein the polynomial is a linear expression, and an absolute value of a gradient of the linear expression or its reciprocal is used as the index. The index area calculated by the following formula for the logarithmic amplitude distribution data whose logarithmization frequency exceeds the same amplitude frequency of the reference distribution is an index indicating the difference between the logarithmic amplitude distribution and the reference distribution. The bearing state monitoring method according to claim 1 , wherein the bearing state monitoring method is used to determine the state of the bearing.

The number of sampling points per one rotation or intermittent operation of the rotating shaft in the detection process is set to a constant value regardless of the number of rotations of the rotating shaft or the speed of the intermittent operation,
Normalized by dividing the frequency of the amplitude distribution which is calculated by using the detection waveform amplitude and logarithmic in the constant value, bearing condition monitoring according to any one of claims 1 to 4 Method. Wherein removing the fluctuation component prior to calculating the amplitude distribution from the logarithmic to said detection waveform amplitude, bearing condition monitoring method according to any one of claims 1 to 5. The bearing state monitoring method according to claim 6 , wherein a moving average or a low-pass filter is used for calculating the fluctuation component. The signal generated by the rotation of the rotary shaft, AE, vibration, or any of the ultrasonic, bearing condition monitoring method according to any one of claims 1 to 7. A sensor unit for detecting a signal generated by rotation of the rotary shaft in a bearing holding the rotary shaft;
A detection processing unit that calculates a detection waveform by performing detection processing on the measurement waveform detected by the sensor unit;
An amplitude distribution calculating unit that logs the amplitude of the detection waveform and calculates an amplitude distribution from the detection waveform obtained by logarithmizing the amplitude;
A mode value of the amplitude distribution is obtained, a low-amplitude side distribution obtained by approximating data having a lower amplitude than the amplitude of the mode value of the amplitude distribution by a normal distribution, and the low-amplitude side distribution is represented by the amplitude of the mode value. An estimated normal distribution composed of a higher amplitude side distribution applied to a higher amplitude side than that, a logarithmic estimated normal distribution obtained by logarithmizing the frequency of the estimated normal distribution is obtained, and a margin is provided in the logarithmized estimated normal distribution. A reference waveform generation unit for obtaining a reference distribution by adding
A bearing state monitoring device comprising: a determination unit that determines the state of the bearing by comparing the logarithmic amplitude distribution obtained by logarithmizing the frequency of the amplitude distribution and the reference distribution . The bearing state monitoring apparatus according to claim 9 , wherein the sensor unit detects any one of AE, vibration, and ultrasonic waves generated in the bearing by rotation of the rotating shaft.

Images