JP5780870B2 - Rotating equipment soundness diagnosis apparatus, method and program - Google Patents

Description

  The present invention relates to a technique for performing a soundness diagnosis of a rotating device based on vibration measurement data.

  Soundness diagnosis of rotating equipment uses the vibration measurement data as the original waveform data to calculate various statistics and analyze trends, or to determine the appropriate amount (displacement, speed, (Acceleration), evaluation is performed, and abnormal signs are captured in the frequency domain by FFT or envelope processing.

Such analysis method, evaluation method and processing method of vibration measurement data largely depend on the know-how of skilled engineers.
For this reason, when the number of rotating devices to be monitored is enormous, such as a nuclear power plant, the work load, cost, and work time of engineers are increased.
In order to overcome such a situation, in recent years, an abnormality diagnosis of a rotating device is performed by using a pattern recognition technique, thereby reducing the work load on an engineer (for example, Patent Documents 1 and 2).

JP 2008-262375 A JP 2010-191556 A

By the way, when performing the soundness diagnosis of the rotating device using the pattern recognition technology, it is necessary to register the pattern information of the normal state and the abnormal state in advance. However, depending on the type of rotating device, there is a situation that the pattern information of the normal state and the abnormal state is different. Furthermore, there are a wide variety of types of rotating equipment to be diagnosed in the plant.
For this reason, when performing the soundness diagnosis of the rotating device by applying the pattern recognition technology, it is necessary to prepare different pattern information for all the rotating devices to be monitored.
However, the occurrence of abnormal events in an actual plant is extremely rare, and it is extremely difficult to satisfy the accumulation of abnormal pattern information in all types of rotating equipment.

  The present invention has been made in consideration of such circumstances, and an object of the present invention is to provide a soundness diagnosis technique for a rotating device having excellent adaptability without depending on the type of the rotating device.

In the rotating device soundness diagnosis apparatus, the first storage unit that stores vibration measurement data of the rotating device, the analysis unit that performs spectrum analysis based on the vibration measurement data, and the attribute of the abnormal event of the rotating device A second storage unit for storing a calculation formula for a characteristic frequency of a simple vibration, and a standardizing characteristic frequency calculated based on the type data of the rotating device and the calculation formula, and correlating a corresponding frequency component of the spectrum analysis A normalization unit, a second normalization unit that normalizes the peak value of the spectrum peak obtained as a result of the spectrum analysis by dividing by an effective value in the frequency domain, and the pattern data of the spectrum peak of the abnormal event a storage unit for classifying the attributes and normal or damaged and storage, it is determined whether the associated was said frequency component is a spectral peak further before Characterized in that it and a determination unit attributes and damage level of the abnormal event based on the pattern data.

  According to the present invention, there is provided a soundness diagnosis technique for a rotating device having excellent adaptability without depending on the type of the rotating device.

The block diagram which shows embodiment of the soundness diagnostic apparatus of the rotary equipment which concerns on this invention. The graph which shows the vibration measurement data of rotating equipment. (A) A table showing a calculation formula of a characteristic frequency of vibration specific to a bearing component which is an example of a rotating device, and (B) a table showing bearing type data. Graph of waveform data obtained by spectral analysis of vibration measurement data. Graph of waveform data obtained by envelope processing of vibration measurement data. (A) A graph of waveform data obtained by spectral analysis of envelope processing data, and (B) an arithmetic expression for normalizing spectral peaks. The table which shows the position of the spectrum peak corresponding to the attribute of the abnormal event of a rotation apparatus. The graph which shows the probability density distribution of the normalized spectrum peak. The display screen which shows the file list of vibration measurement data, and the judgment result of each file. A display screen showing a feature vector in which a horizontal axis is a character string in which a characteristic frequency is normalized and a vertical axis is a crest factor in which a spectrum peak is normalized.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.
As shown in FIG. 1, the health diagnostic apparatus 10 for a rotating device performs a spectrum analysis based on the first storage unit 11 that stores vibration measurement data (FIG. 2) of the rotating device and the vibration measurement data (FIG. 2). Analysis unit 15, a second storage unit 12 that stores a calculation formula (FIG. 3A) of the characteristic frequency of vibration unique to the attribute of the abnormal event of the rotating device, and type data of the rotating device (FIG. 3B )) And the first standard that relates the frequency components of the corresponding spectrum analysis (FIG. 6A) by normalizing the characteristic frequency (FIG. 3A) calculated based on the calculation formula (FIG. 3A) And a determination unit 23 that determines an attribute of the abnormal event (FIG. 7) based on whether or not the associated frequency component is a spectrum peak.

FIG. 2 is a graph showing a waveform of vibration measurement data of a rotating device.
The first storage unit 11 stores a plurality of such vibration measurement data as independent files. The vibration measurement data stored in the first storage unit 11 is measured by attaching an acceleration sensor to a plurality of rotating devices (rolling bearings, sliding bearings, gears, pumps, motors, etc.) installed in the plant. Therefore, it is data for which a health diagnosis has not been performed.

FIG. 3A shows a calculation formula for the characteristic frequency of vibration inherent to the components of the bearing, which is an example of the rotating device, and the data value of the type of the rotating device illustrated in FIG. 3B is substituted. The characteristic frequency calculated as follows is shown.
The bearing is composed of an inner ring, an outer ring, a rolling element, and a cage. When any one of these is damaged, vibration unique to each of the bearings is superimposed on the vibration measurement data (FIG. 2). And the characteristic frequency of such a natural vibration can be derived from the type data (FIG. 3B) and the calculation formula (FIG. 3A) of the rotating device.

The second storage unit 12 stores the calculation formulas (not shown) of the characteristic frequencies of other rotating devices, including the calculation formulas of the characteristic frequencies of the bearings illustrated in FIG.
The 3rd preservation | save part 13 is storing the specification data (classification data) regarding the classification of the rotary equipment used for the plant so that it may be illustrated by FIG. 3 (B).
Then, the characteristic frequency calculator 19 calculates the characteristic frequency according to the stored calculation formula and the type data of the rotating device.

The graph of FIG. 5 shows waveform data obtained by performing envelope processing on vibration measurement data.
The envelope processing unit 14 acquires vibration measurement data stored in the first storage unit 11 and executes envelope processing (envelope processing).
There are a plurality of specific methods of envelope processing. An outline of a method for obtaining an envelope waveform by performing a Hilbert transform on vibration measurement data (FIG. 2) to obtain a complex analysis signal will be described.

(1) Fast Fourier transform (FFT) is performed on the vibration measurement data in the time domain, and the spectrum x (i); [i = 0, 1,..., N / 2-1, n / 2, n / 2 +1,..., N-1 (n; even number)] is obtained (FIG. 4).
(2) The following operation is performed on the spectrum x (i).
y (i) = x (i) × h (i)
Here, when i = 0, h (i) = 1.
When i = 1,..., n / 2-1, h (i) = 2.
When i = n / 2, h (i) = 1.
When i = n / 2 + 1,..., n−1, h (i) = 0.

(3) Inverse fast Fourier transform (IFFT) is performed on the spectrum y (i).
When the real part of the obtained complex signal x _a (n) is r (n) and the imaginary part is h (n), this imaginary part h (n) is Hilbert transform.
The instantaneous amplitude A (n) at an arbitrary time is expressed by the following equation.
A (n) = √ (r (n) ² + h (n) ² )
This utilizes the property that the signal subjected to the Hilbert transform is delayed in phase by π / 2 in all frequency bands with respect to the original signal.
An envelope waveform (FIG. 5) of the original vibration measurement data is obtained by connecting the instantaneous amplitudes A (n).

The graph of FIG. 4 shows waveform data obtained by spectral analysis of vibration measurement data (FIG. 2), and the graph of FIG. 6A shows waveform data obtained by spectral analysis of envelope processing data (FIG. 5). .
The spectrum analysis unit 15 executes spectrum analysis of vibration measurement data stored in the first storage unit 11 in addition to executing spectrum analysis of envelope waveform data created in the envelope processing unit 14. The data subjected to the spectrum analysis in this way is stored in the analysis data storage unit 16.

The table of FIG. 7 shows the position of the spectrum peak corresponding to the attribute of the abnormal event of the rotating device. Character strings (elements) such as 1 / n × Fr, n × Fr, n × Fin, n × Fout, and 2n × Fball are standardized characteristic frequencies. Here, “1” in the table indicates that a spectrum peak is observed at the characteristic frequency corresponding to the corresponding character string.
Thereby, the presence of the spectrum peak corresponding to the vibration unique to the abnormal event is indicated by the character string normalized from the characteristic frequency without depending on the type of the rotating device to be diagnosed.

The first normalization unit 20 is a character string (1 / n × Fr, n × Fr, n × Fin, n × Fout) obtained by normalizing the characteristic frequency (FIG. 3) of the rotating device calculated by the characteristic frequency calculation unit 19. , 2n × Fball) and the frequency components of the corresponding spectrum analysis (FIG. 4, FIG. 6A).
The peak search unit 21 searches for a character string indicating a spectrum peak among the frequency components related in this way.
The determination unit 23 compares the character string pattern searched by the peak search unit 21 with the table shown in FIG. 7 and determines the attribute of the abnormal event of the rotating device to be diagnosed.

FIG. 6B shows an arithmetic expression for normalizing the spectrum peak.
The 2nd normalization part 22 performs the calculation which normalizes a spectrum peak according to these arithmetic expressions. Here, a method for normalization by a crest factor (crest factor) in the frequency domain is illustrated.

The overall value X _OA in the frequency domain is shown as the equation (1) in FIG.
Here, V _k ; k-th component of the envelope curve spectrum H _f ; Hanning window correction value (H _f = 2/3)
K: Number of sample points (for example, 4096 points)

Then, the effective value X _rms in the frequency domain is shown as shown in FIG. 6B (2) and the crest factor CF in the frequency domain is shown as shown in FIG. 6 (B) and (3). Here, X _peak is the value of the spectrum peak of the waveform data obtained by spectrum analysis.

  In this way, the crest factor CF corresponding to each spectrum peak is handled as normalized data. As a result, the spectrum peak and effective value change depending on the type of rotating equipment, but the crest factor CF has no such dependence, so the analysis results obtained from rotating equipment of different types are compared. it can.

On the other hand, since the crest factor CF reflects the damage level of the abnormal event of the rotating device, the damage level can be accurately determined even by comparing rotating devices of different types.
Thereby, in the determination part 23, not only the attribute of an abnormal event but a damage level can be determined based on the normalized spectrum peak.

In this case, the determination unit 23 can be realized by using a support vector machine (SVM). In this case, it is necessary to previously accumulate vibration measurement data of a rotating device whose attribute of an abnormal event is known as pattern data in the accumulation unit 18.
Based on the pattern data stored in the storage unit 18, the support vector machine determines which abnormal event attribute or damage level the rotating device to be diagnosed corresponds to.

  In the initial state, such pattern data is classified by an engineer having judgment ability. In other words, the results of observation of vibration measurement data (Fig. 2) or its envelope processing data (Fig. 5) or their spectrum analysis data (Figs. 4 and 6 (A)) are compared with the results of direct examination of rotating equipment. By doing so, each spectrum analysis data is classified into pattern data of corresponding abnormal event attributes.

The pattern learning unit 17 takes in the spectrum analysis data stored in the analysis data storage unit 16 as pattern data and updates the determination parameter of the determination unit 23. That is, since the spectrum analysis data of the vibration measurement data subjected to the diagnosis is known data, it can be adopted as the next diagnosis pattern data.
Thereby, the determination precision in the determination part 23 improves as the diagnostic score of a rotating device increases.

The graph of FIG. 8 shows the probability density distribution of the standard value of the spectrum peak at the feature frequency of interest. It is clear that the normalized spectral peak intensity statistically shows a normal distribution as the number of diagnostic points increases.
Therefore, the damage level determination of the abnormal event in the determination unit 23 can be executed based on the probability density distribution of the standard value of the spectrum peak.

The display screens shown in FIGS. 9 and 10 are examples of determination results displayed on the determination result display unit 24.
In the file list shown in FIG. 9, “file name” corresponds to each vibration measurement data, and “judgment result” corresponding to “unexecuted” is stored in the first storage unit 11. Those corresponding to “normal” or “damage” are stored in the pattern data storage unit 18.

When any one “file name” is selected from the file list, detailed information of the file is displayed at the top of the screen.
When the “unexecuted” file is selected and the “diagnosis” button is pressed, the soundness diagnosis is executed, and the display of the “judgment result” is switched to “normal” or “damage”.

FIG. 10 shows a feature vector in which a character string in which the characteristic frequency is normalized is represented on the horizontal axis and a crest factor in which the spectrum peak is normalized is represented on the vertical axis.
When a file other than “Not implemented” is selected on the screen of FIG. 9 and the “Confirm” button is pressed, the screen of FIG. 10 is displayed. The feature vector of the file selected by the “Confirm” button is highlighted (indicated by a one-dot chain line in FIG. 10).
Furthermore, by returning to the file list of FIG. 9 and selecting another file and pressing the “Add” button, the feature vector of the selected file can be overwritten (displayed with a solid line in FIG. 10).

  Further, by selecting the other pull tabs shown in FIG. 10, vibration measurement data (FIG. 2), original waveform spectrum (FIG. 4), envelope waveform (FIG. 5) corresponding to the file selected by the “confirm” button. ), Envelope spectrum (FIG. 6), probability density distribution (FIG. 8).

  According to the above-described embodiment, by normalizing the characteristic frequency or normalizing the spectrum peak, it is possible to provide a soundness diagnosis technology for a rotating device that is excellent in adaptability without depending on the type of the rotating device. . Thereby, while reducing the work burden, cost, and time of the engineer who performs the soundness evaluation of the rotating equipment, the quality of the evaluation work can be homogenized.

Although the embodiments of the present invention have been described, these embodiments are presented as examples, and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are also included in the invention described in the claims and the equivalents thereof.
The rotating device soundness diagnosis apparatus can also realize each means as a function program by a computer, and can be operated by a rotating device soundness diagnosis program that combines the function programs.

  DESCRIPTION OF SYMBOLS 10 ... Soundness diagnostic apparatus of rotating equipment, 11 ... Vibration measurement data storage part (1st storage part), 12 ... Calculation formula storage part (2nd storage part), 13 ... Type data storage part (3rd storage part), DESCRIPTION OF SYMBOLS 14 ... Envelope processing part (processing part), 15 ... Spectrum analysis part (analysis part), 16 ... Analysis data storage part (storage part), 17 ... Pattern learning part, 18 ... Pattern data storage part (storage part), 19 ... Feature frequency calculation unit, 20 ... first normalization unit, 21 ... peak search unit, 22 ... second normalization unit, 23 ... abnormal event attribute determination unit (determination unit), 24 ... determination result display unit.

Claims (6)

A first storage unit for storing vibration measurement data of the rotating device;
An analysis unit for performing spectrum analysis based on the vibration measurement data;
A second storage unit that stores a calculation formula of a characteristic frequency of vibration unique to the attribute of the abnormal event of the rotating device;
A first normalization unit that normalizes a characteristic frequency calculated based on the type data of the rotating device and the calculation formula and relates a corresponding frequency component of the spectrum analysis;
A second normalization unit that normalizes the peak value of the spectrum peak obtained as a result of the spectrum analysis by dividing by the effective value in the frequency domain;
An accumulator for classifying and accumulating the spectral peak pattern data into the abnormal event attribute and normal or damage;
A rotating device comprising: a determination unit that determines whether or not the related frequency component is a spectrum peak, and further determines an attribute and a damage level of the abnormal event based on the pattern data Health diagnostic equipment. The soundness diagnostic apparatus for a rotating device according to claim 1,
The apparatus for diagnosing soundness of a rotating device, wherein the spectrum analysis is performed after envelope processing is performed on the vibration measurement data. In the health diagnostic apparatus for a rotating device according to claim 1 or 2 ,
The spectral peak normalization is performed by calculating a crest factor of a spectrum of an envelope curve. The soundness diagnostic apparatus for a rotating device according to any one of claims 1 to 3 ,
The determination of the damage level is executed based on a probability density distribution of a standard value of the spectrum peak. Storing the vibration measurement data of the rotating device;
Performing spectral analysis based on the vibration measurement data;
Storing a formula for calculating a characteristic frequency of vibration unique to the attribute of the abnormal event of the rotating device;
Normalizing the characteristic frequency calculated based on the type data of the rotating device and the calculation formula and relating the corresponding frequency components of the spectrum analysis;
Dividing the peak value of the spectral peak obtained as a result of the spectral analysis by the effective value in the frequency domain, and normalizing;
Classifying and accumulating the spectral peak pattern data into the abnormal event attributes and normal or damaged;
Determining whether or not the related frequency component is a spectrum peak, and further determining an attribute and damage level of the abnormal event based on the pattern data . Health diagnostic method. On the computer,
Function to save vibration measurement data of rotating equipment,
A function of performing spectrum analysis based on the vibration measurement data;
A function for storing a calculation formula of a characteristic frequency of vibration unique to an attribute of an abnormal event of the rotating device;
A function of normalizing the characteristic frequency calculated based on the type data of the rotating device and the calculation formula and relating the corresponding frequency component of the spectrum analysis;
A function of normalizing by dividing the peak value of the spectrum peak obtained as a result of the spectrum analysis by the effective value in the frequency domain,
A function of classifying and storing the spectral peak pattern data into the abnormal event attribute and normal or damage;
A function of determining whether or not the related frequency component is a spectrum peak, and further executing a function of determining an attribute and damage level of the abnormal event based on the pattern data . Health diagnostic program.

Images