JP2010066244A - Method and system for diagnosis of abnormal conditions in facilities - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To efficiently analyze frequencies of sound generated from these facilities in daily patrol, and precisely detect signs of any abnormal condition in the facilities to prevent the occurrence of serious accidents. <P>SOLUTION: A method includes a step of collecting steady-state sound data per facility and store the above collected data as constant domain data, a step of collecting sound data of the facility and store them as measurement data, a macro diagnostic step of calculating correlation coefficient of frequency spectrum on the section including constant domain and anomaly estimation domain and then determine the presence of abnormal conditions based on the above correlation coefficient, and a micro diagnostic step of calculating standard deviation multiple which indicates how many times of standard deviation of the whole differential data each value of differential spectrum between anomaly estimation domain spectrum and constant domain spectrum per category is, then use this standard deviation multiple to specify abnormal frequency band, if determined as abnormal through the macro diagnostic step. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a facility abnormality diagnosis method and system for determining the presence or absence of an abnormality by frequency analysis of sound data generated from the facility.

  With the progress of electric power liberalization, the maintenance of facilities is also required to be rationalized, and CBM (Condition Based Maintenance) is being promoted.

  Conventionally, as one of the abnormality diagnosis methods for power equipment, there is a method in which sound data generated from the equipment is subjected to frequency analysis by Fourier analysis, and the presence / absence of abnormality is determined based on frequency components contained therein. (For example, see Patent Documents 1 and 2).

  In general, frequency analysis of equipment data includes DFT (Discrete Fourier Transform) and FFT (Fast Fourier Transform), but FFT is often used because of its high calculation speed. However, the analysis by FFT has a restriction that the number of data must be 2 to the nth power, and when the number of data is insufficient, it is usually necessary to add zero to create data for analysis. Also, the data is multiplied by a window function so that the first and last values of the data are equal to make periodic changes smooth.

  For this reason, the FFT results in analyzing the processed data, not the true measured data, and the data processing method (window function selection) as well as the quality of the measured data as the cause of error in the analysis results. Therefore, there is a problem in that the validity of the analysis result becomes complicated.

In addition, as a technique for collecting sound data of equipment, for example, in Patent Document 3, airborne sound is collected by a parabolic sound collector, but when the equipment is installed adjacently, It is difficult to determine clearly from which equipment the sound is. In particular, if it is attempted to detect a faint abnormal sound before failure of the equipment, such an abnormal sound often does not propagate as a sound in the air and may not be accurately detected.
JP 2008-140222 A JP 2008-33532 A JP 2004-20484 A

  The present invention has been made in view of the above-described circumstances, and is intended to grasp the progress of anomalies over time. When the calculation processing time is not limited, measurement data is obtained by frequency analysis using DFT. The data can be used as it is without limitation, and the frequency of sound generated from the equipment during daily patrols is efficiently analyzed to accurately detect signs of equipment abnormalities and prevent serious accidents from occurring. An object of the present invention is to provide a facility abnormality diagnosis method and system that can be prevented in advance.

  The present invention includes (1) global diagnosis of sounds generated from equipment (an anomaly estimation function based on human hearing sounds and an overall waveform display function and an input function of estimation results), and (2) macro diagnosis (comparison coefficient comparison of frequency spectra). Provides a method and system for diagnosing equipment abnormality by three-stage diagnosis of overall normal / abnormality determination) and (3) micro-diagnosis (extraction of abnormal frequency band for an object determined to be abnormal).

The concept of the abnormality estimation function according to the present invention is as follows.
(A) Measurement of sound propagation in solids:
(1) Small abnormal sounds cannot be transmitted in the air, and if the equipment to be measured is adjacent, the sound source cannot be clearly identified. Use a “solid-state propagation sound collection microphone” that can collect sound.
(2) If there is an abnormality in the rotating machine, the abnormality is considered to occur somewhere within the range in which the device makes one rotation, so if you measure data of more than one rotation of the device, you can find an abnormality in this It is considered possible.
(3) Since the rotation speed of the three-phase AC device is usually in the range of 100 to 6000 rotations / minute (2 to 100 rotations / second), abnormalities can be found by measuring data for 1 second or more. However, taking into account abnormalities in the attached valves, it is preferable to measure the data for about 30 seconds with an expected safety factor so as to find abnormalities including the attached devices.

(B) Propagation sound in solids: <Estimate abnormal part by listening sound>
An efficient, effective and best method for estimating an abnormal part from measurement data is to listen to the measurement sound directly with the human ear. The stationary part and the abnormal part have distinctly different sounds, and the human ear can easily recognize the difference. For this reason, it is effective to provide a function for inputting a judgment result by a person.

(C) Overall waveform confirmation: <Identify the abnormal part>
(1) Frequency analysis cannot be performed unless the abnormal part is specified and extracted. (2) An efficient, effective and best method for estimating an abnormal portion from measurement data is to graph the measurement data and view it directly as a waveform with the human eye. (3) Although the entire waveform draws a certain envelope curve, the abnormal part often shows a waveform that partially protrudes from the envelope curve, such as a whisker-like pulse. In addition, (4) by comparing with the listening result, the estimation of the abnormal part becomes more reliable.

The concept of the diagnostic function according to the present invention is as follows.
(D) Macro diagnosis: <Evaluation of whether analysis spectrum includes abnormal part>
(1) The average of the correlation coefficients of “steady part spectrum vs. abnormal estimation part spectrum” at a plurality of locations is obtained, and a. Above the specified value is considered normal, and b.
(2) b. If the value is less than the specified value, further micro-evaluation is performed (the specified value is determined by the measurement and analysis results for each target facility).
(3) Even if, for example, the excitation vibration sound of a regular transformer is different, if the extraction location in the data is different, the analysis spectrum value is slightly different, so the spectrum value is different for each of p (for example, p = 10) data. Collectively in the frequency band (hereinafter referred to as Σp spectrum) for macroscopic diagnosis.

(E) Micro diagnosis: <Specify frequency band of abnormal part>
Anomalous frequency is obtained by obtaining the standard deviation multiple of how many times the standard deviation of the entire Σp difference spectrum is the difference spectrum between the Σp spectrum of the stationary part and the Σp spectrum of the anomaly estimation part (hereinafter referred to as Σp difference spectrum). Specify the bandwidth. (For example, 3σ (= 99.73%).)

  In addition, when microdiagnosis is performed from the beginning for each Σp difference spectrum of “steady part data vs. steady part data” and “steady part data vs. abnormal assumption part data”, there are places where the standard deviation multiple is always exceeded in any case Therefore, it is necessary to first determine whether there is an abnormality by macro diagnosis.

  Specifically, the equipment abnormality diagnosis method according to the present invention is a equipment abnormality diagnosis method that collects sound data generated from equipment and diagnoses the presence or absence of abnormality by frequency analysis by a computer, and is in a steady state for each equipment. Collecting sound data, storing the collected data as stationary part data, collecting sound data of the equipment and storing it as measurement data, and frequency of a section including the steady part data and an abnormality estimation part of the measurement data A macro diagnosis step of obtaining a spectrum, calculating a correlation coefficient between values obtained by dividing and summing the two spectrum values for each predetermined number of frequency orders, and determining the presence or absence of an abnormality based on the correlation coefficient If the macro diagnosis process determines that there is an abnormality, the difference spectrum of the aggregated values for each of the two spectral value categories is the standard for the entire difference data. Calculating a standard deviation multiples indicating which is many times the deviation, characterized in that it comprises a, and micro diagnosis step of identifying an abnormal frequency band by the standard deviation multiples.

  In the present invention, sound data (stationary part data) in a steady state with no abnormality is collected in advance for the equipment to be measured, and the correlation coefficient between the Σp spectrum of the measurement data and the Σp spectrum of the steady part data is below a certain value. In this case, it is determined that there is an abnormality, and the abnormal frequency band is specified by the Σp difference spectrum of both. Since there are various abnormal modes, it is more effective and effective than patterning the spectrum distribution at the time of abnormality and determining whether or not the pattern corresponds to the pattern.

  The equipment abnormality diagnosis method according to the present invention further uses a Fourier transform as a frequency analysis to calculate a Fourier coefficient of sound data. When the number of sound data is an even number, the maximum of the calculated general discrete Fourier coefficients is calculated. Only the order coefficient is corrected to a half of the calculated value, and the frequency spectrum is calculated.

  A high restoration rate can be realized as compared with the use of a Fourier coefficient by a general discrete Fourier transform.

  Preferably, in the macro diagnosis step, the presence / absence of an abnormality may be determined by comparing a frequency spectrum value of a section including the abnormality estimation unit and a total value of frequency spectrum values of a plurality of sections adjacent to the section. Thereby, an abnormality can be detected easily.

  In addition, it is possible to evaluate the validity of the analysis data by including a step of displaying the ratio (restoration rate) between the value of the measurement data and the restoration value when the original data is restored from the frequency spectrum of the measurement data. Become.

  Furthermore, by including the step of accumulating the frequency spectrum data of the measurement data and displaying the trend of secular change, it becomes easy to determine an abnormality due to the change tendency.

  More preferably, it is preferable to include a step of collecting the propagation sound in the solid of the facility as sound data and designating an abnormal portion estimated by the listening sound or the displayed data waveform. By diagnosing the frequency division around the specified abnormal part, the processing time can be shortened. Instead of specifying the abnormal part, the result of the abnormality determination by a person may be input.

  Moreover, the equipment abnormality diagnosis method according to the present invention includes a step of displaying a failure mode corresponding to an abnormal frequency band for each equipment type. By displaying the failure modes expected from those having a high correlation coefficient, it is possible to link to preventive maintenance.

  Similarly, the standard deviation multiple and the average of the period until the occurrence of a failure may be calculated for each equipment type, and thereby the failure time may be predicted.

An equipment abnormality diagnosis system according to the present invention is an equipment abnormality diagnosis system that collects sound data generated from equipment and diagnoses the presence or absence of abnormality by frequency analysis, and uses steady state sound data for each equipment as steady-state data. A stationary part spectrum database to be stored; measurement data input means for collecting sound data of equipment and storing it as measurement data; a p-value table for storing the number (p) of collected frequency orders of sound data for each equipment; and p-value Referring to the table, extract the number of pieces corresponding to the equipment to be measured, calculate the correlation coefficient between the Σp spectrum of the stationary part data of the equipment and the Σp spectrum of the data including the abnormality estimation part of the measured data, Macro diagnosis means for determining the presence or absence of an abnormality based on the correlation coefficient, and when the macro diagnosis means determines that there is an abnormality, Each time, a standard deviation multiple indicating how many times the standard deviation of the entire Σp difference spectrum is calculated for each value of the Σp difference spectrum between the Σp spectrum of the abnormal part estimation data and the Σp spectrum of the stationary part data is calculated. And a micro diagnostic means for specifying an abnormal frequency band by a deviation multiple.
Here, Σp means that p spectrum values are added and totaled.

  According to the present invention, the macro diagnosis is first executed based on the measurement data to determine the presence or absence of an abnormality, and when it is determined that there is an abnormality, the abnormal frequency band is further specified by the micro diagnosis, leading to equipment failure. The sign of the previous abnormality can be detected with high accuracy.

  Moreover, since the frequency spectrum of the measurement data is divided and calculated for each predetermined number of pieces in each diagnosis, it is possible to detect abnormalities with high accuracy by simple calculation even for abnormal sounds in bands where the frequencies are slightly shifted. .

  Furthermore, by using a solid-borne propagation sound collection microphone, it is possible to determine from which equipment the sound comes from, reducing the influence of external noise, and high accuracy even for equipment with abnormal noise Diagnosis is possible.

  Embodiments of the present invention will be described below. FIG. 1 is a functional block diagram of the equipment abnormality diagnosis system according to the present embodiment. Here, the equipment abnormality diagnosis system 1 has, as input functions, a solid-propagating sound collection microphone (hereinafter referred to as a sound collector) 60 and a recorder 70 that records sound data collected by the sound collector 60. is doing. The recorder 70 includes a storage unit 71 that stores sound data and an output unit 72 that outputs the sound data to the outside as a sound or a sound signal.

  Here, as shown in FIG. 2, the sound collector 60 is filled in the solid-propagating sound collecting microphone container 62 whose one surface is the sound collecting section 64 and the solid-propagating sound collecting microphone container 62. A solid-borne propagation sound transmitting material 63 such as rubber clay, and a microphone 61 provided in the solid-borne propagation sound collecting microphone container 62 so as to be covered with the solid-borne propagation sound transmitting material 63. That is, the sound collector 60 covers the entire microphone 61 such as a condenser microphone with the in-solid propagation sound transmitting material 63 in order to block the propagation sound in the air and efficiently collect the abnormal sound of the equipment 80. While being accommodated in the propagation sound collection microphone container 62, the sound collection portion 64 of the solid-state propagation sound collection microphone container 62 can be brought into close contact with the surface of the facility 80. Thereby, in the sound collector 60, only the abnormal sound from the facility 80 can be taken in from the sound collecting portion 64 of the propagation sound collecting microphone container 62 in the solid and detected by the microphone 61.

  FIG. 3 (a) shows a graph of measurement data of the propagation sound in the solid of the compressor as an example of equipment, and FIG. 3 (b) shows the measurement data of the propagation sound in the air. As for the propagation sound in the solid, a whisker-like pulse is seen in the waveform, and an abnormal noise can be heard in synchronization with the generation timing of this pulse. On the other hand, abnormal noise cannot be heard in the air-borne sound, and no whisker-like pulse can be confirmed in the waveform.

  The data processing function of the equipment abnormality diagnosis system 1 includes an input unit 10 for inputting data from the outside, an output unit 20 configured by a display or the like and having functions such as display output, an arithmetic unit 30 for executing arithmetic processing, and storing data. Storage unit 50.

  Further, the calculation unit 30 receives the measurement data digitized by the A / D conversion unit 11 and stores the measurement data in the storage unit 50, and the judgment of the listening result by a patrolman or the like via the input unit 10. Listening result input means 22 for inputting, Fourier coefficient calculating means 23 for calculating the Fourier coefficient of the input measurement data, waveform generating means 24 for generating a waveform using the Fourier coefficient and displaying it on the output unit 20, macro diagnosis Means 25, micro diagnostic means 26, and determination result output means 27 for outputting a determination result are provided. Each means 21-27 is a function realizable by a program. Note that the A / D conversion unit 11 may be provided in the recording device 70 instead of being provided in the input unit 10, and the data after A / D conversion may be stored in the storage unit 71.

(Operation of Fourier coefficient calculation means)
The Fourier coefficient calculation means 23 calculates the Fourier coefficient of the measurement data. Hereinafter, a general Fourier coefficient and a formula for calculating the Fourier coefficient used in the present embodiment will be described.

In general, the Fourier series and Fourier coefficient of discrete data (number of data m) are derived by the equations (1) to (4).
Where x = 1,2,3, ------ m and n = 1,2,3, ------ N (N ≦ 1 / 2m from the sampling theorem).

On the other hand, in the Fourier coefficient calculation means 23 according to the present embodiment, Fourier coefficients a _k and b _k (k <1 / 2m) and a _N and b _N (N = 1 / 2m) are respectively calculated from (5) to Calculate with equation (8).
Comparing Eqs. (3) and (7) here, it can be seen that only the maximum order coefficient is 1/2 times the other order coefficient (a _N = 1 / 2a _k ). 7) The feature of this embodiment is that the calculation is performed using the equation.

That is, where m is the number of discrete data and n is the maximum frequency contained therein, the Fourier coefficient a _{N of the} data is expressed by equation (7) when n = N = (1/2) m. By using the Fourier coefficient obtained by this equation, the original data can be completely restored.

According to the sampling theorem, the order of the maximum frequency included in the data is n, and when the number m of data extracted at a certain sampling time is an even number of 2 or more, the equation (9). 10) The formula holds.
n ≦ m / 2 ----------------------------------------- (9)
n ≦ (m−1) / 2 ---------------------------- (10)

Next, the flow of the equipment abnormality diagnosis method according to the present embodiment will be described with reference to FIG.
(Sound data collection stage)
First, sound is collected by the solid-borne propagation sound collection microphone 60 and recorded in the storage means 71 of the recorder 70 (S101). Then, the recorder 70 is played back, and the presence or absence of abnormal noise is confirmed by listening to sound from a patrolman or the like (S102). The abnormal sound is a high / loud sound generated in a pulse manner different from the continuous stationary sound. Next, the playback sound of the recorder 70 is converted into digital data by the A / D conversion means 11 (S103). This digital data is stored in the storage unit 50 as a measurement data file 51 by the measurement data input means 21.

  Further, the Fourier coefficient calculation means 23 calculates Fourier coefficients using the discrete data of the measurement data file 51 as described above, and stores them in the measurement target spectrum data file 52.

(Overall diagnosis stage)
Next, the measurement data 51 is converted into a waveform graph by the waveform generation means 24 and displayed on the output unit 20. FIG. 5 is an enlarged view of the beard-like pulse portion of the compressor displayed on the output unit 20. The patrolman etc. confirms the presence or absence of abnormalities such as sudden changes in the waveform and whiskers using this graph, and also correlates with the abnormal noise generation unit in step S102, that is, abnormal noise and abnormal waveform generation on the same time axis. Is checked (S104).

(Macro diagnosis stage)
As a result, if it is determined that there is a suspicion of abnormality (“Yes” in S105), then the macro diagnosis means 25 is activated, and a p-value table storing the number of frequency orders collected for each facility. 54, and the number of frequency orders collected in association with the type of equipment to be diagnosed is extracted (S106), and the spectrum data file 52 to be measured is accessed, and an arbitrary plurality of sections of the continuous steady portion (Corresponding to B and C in FIG. 5) is extracted and subjected to frequency analysis (S107). In this case, it is confirmed that the restoration rate is 100%. FIG. 6B shows the frequency analysis result of the stationary part displayed on the output unit 20. The frequency order is taken on the horizontal axis, and the spectrum value is taken on the vertical axis.

  When frequency analysis is performed, the quality judgment of the result becomes a problem. However, if the restoration rate is perfect (100%), even if there is a missing frequency component in the analysis result, it is not an analysis error but an original data. It is possible to clearly determine that the frequency component is not included in “.

Here, the restoration rate is obtained by substituting the frequency spectrum obtained as a result of the frequency analysis as a coefficient of the Fourier series, calculating a value (restoration value) for each sampling time, and restoring the original data. Based on the above, the ratio to the measured data is obtained and averaged over all data.
Restoration rate = restoration value / measured data value * 100 (11)

In equation (11), frequency analysis is performed by adding twice the maximum value of the absolute value of the data to each data in consideration of the fact that if the measured data value contains zero, it cannot be calculated. Is preferred. In this case, the entire data is translated upward by twice the maximum value, but the frequency analysis results in only the a ₀ term (constant term) of the Fourier coefficient changing, and the frequency spectrum does not change.

  In the following description, it is assumed that the number of collectives p = 10. The steady part spectrum is accumulated in the steady part spectrum DB 53 each time, and is compared with the abnormality estimation part. Further, frequency analysis is also performed for the abnormality estimation unit (corresponding to part A in FIG. 5) (S108). A graph of the frequency analysis result of the abnormality estimation unit is shown in FIG.

  Next, the correlation coefficient of the spectrum value Σ10 (where 10 is the total number) is calculated using the frequency band order as a common horizontal axis for both graphs of the above-mentioned total number 10 and steps S107 and S108 (S109), It is determined whether this correlation coefficient is equal to or greater than a predetermined value (S110). FIG. 7 is a table of correlation coefficients of the compressor sound data spectrum. The correlation coefficient based on all data and the correlation coefficient of Σ10 data are shown by month. In the case of a compressor, the criterion for macro diagnosis of correlation coefficient is appropriate to be 40% for all data and 70% for Σ10 data, but this specified value depends on the target equipment, sampling time, number of analysis data, etc. It is necessary to select an appropriate value.

  If it is equal to or less than the specified value in step S110 (“No” in S110), after accumulating data in the steady spectrum DB 53, the process ends as having no abnormality.

(Micro diagnosis stage)
On the other hand, if the correlation coefficient is greater than or equal to the specified value in step S110 (“Yes” in S110), the micro diagnostic means 26 is activated, and the frequency band order is set as a common horizontal axis for both graphs in steps S107 and S108, and Σ10 A difference spectrum (= | abnormality estimation part spectrum value−steady part spectrum value |) is calculated (S112).

  Next, the standard deviation value (σ) of this Σ10 difference spectrum and the standard deviation multiple of each difference spectrum (= Σ10 difference spectrum / σ) are calculated (S113).

  When this standard deviation multiple is a frequency band of a specified value, for example, 3σ (exceeding 99.73%) or more (“Yes” in S114), the determination result output means 27 is used as a caution or abnormality source frequency band. Display output to the output unit 20. FIG. 8A is a standard deviation multiple graph of the difference spectrum of all data, and FIG. 8B is a standard deviation multiple graph of Σ10 data. In FIG. 8 (a), abnormalities are observed in two places, A and B, whereas in FIG. 8 (b), abnormalities are observed only in B. The result of the Σ10 difference spectrum well represents the tendency of all data spectrum comparison, and is suitable for grasping the abnormal frequency band as a whole. A patrolman or the like performs follow-up observation with this output (S115), and when the situation advances, investigates the cause and implements countermeasures (S116, S117).

In the follow-up observation, it is effective to predict the failure mode and the failure time using the prediction tables shown in FIGS. 12 and 13 and take measures in advance. Details will be described below.
FIG. 12 is a failure mode prediction table in which the abnormal frequency band and the failure mode of each facility are recorded for each type of facility. The abnormal frequency band specified by the micro diagnostic means 26 and the history of the failure mode when an actual failure occurs thereafter are stored, and this history data is used for each type of equipment such as a compressor and a transformer. It is stored as a failure mode prediction table in which the frequency band and the failure mode are compared. And when abnormality is discovered about the measurement object equipment, the failure mode is displayed from the abnormal frequency band with reference to this table. As a result, the patrolman can take measures such as repairing that part or preparing maintenance items in advance. In particular, when the facility is composed of a plurality of devices (or parts), the abnormal frequency band may be different for each device, and maintenance for each device becomes possible, and maintenance costs can be reduced.

  FIG. 13 is a failure time prediction table in which the standard deviation multiple and the average period until failure occurrence are recorded for each equipment type. With reference to this table, the failure period is extracted from the standard deviation multiple of the equipment to be measured, and the failure time is predicted. Note that the failure period can be calculated by using the shortest period, standard deviation, or the like, instead of obtaining the average period until the actual failure occurs.

(Example)
Hereinafter, as an example, the validity of comparison between a stationary part and an abnormality estimation part based on Σ10 data will be described.

FIG. 9 is a graph of measurement data in a specific range including the compressor abnormality estimation unit. In FIG. 9, the P part is a beard pulse-like abnormality generating part.
(1a) to (5a) are abnormality estimation parts including the P part, and (1b) to (5b) are stationary parts. Here, the frequency analysis spectra of (1a) to (5a) and (1b) to (5b) are SP different (1a) to SP different (5a) and SP constant (1b) to SP constant (5b), respectively. , SP difference (1) (= | SP difference (1a) −SP constant (1b) |), and similarly, SP difference (2),..., SP difference (5) are obtained.

  Next, standard deviation values ST (1),..., ST (5) of each SP difference are obtained, and a value obtained by dividing each SP difference value by each ST value is a standard deviation multiple, nst (1),. Calculate nst (5).

  When frequency analysis is performed on the measurement data, the analysis spectrum value also varies slightly due to the difference in the analysis data extraction interval. The sound data of the equipment is not always a regular periodic waveform such as a sine wave, but complex waveform data including noise etc. Even if A / D conversion is performed with the same sampling and frequency analysis is performed with a constant number of data, data measurement is possible. Since it is difficult to perform analysis data extraction under exactly the same conditions as in the previous analysis, a shift in the analysis data extraction interval is unavoidable.

  Therefore, it is necessary to confirm how much the analysis data extraction interval shifts as the analysis spectrum shift.

  In this embodiment, the frequency analysis is performed on 1000 data obtained by A / D converting the propagation sound in the solid during the operation of the compressor for generating compressed air with a sampling time of 20 μS.

  Here, the abnormality was estimated by comparing the spectrum between the abnormal part and the steady part (correlation coefficient), and by the standard deviation multiple of the spectrum difference between the abnormal part and the steady part, but the above-mentioned nst (1),. Nst (5) may be considered to indicate a variation state of the standard deviation multiple of the abnormal / steady part spectrum difference due to the deviation of the data extraction interval.

FIG. 10 is a graph of these nst (1),..., Nst (5) with the frequency order as the horizontal axis.
Here, in the sections (a) to (g) where nst ≧ 3.0, the frequency order is 10 or less. This indicates that the shift in the range where the spectral difference is large (3σ or more) due to the shift in the analysis data extraction interval is within 10 frequency orders.

  From this, it can be seen that the comparative study based on the Σ10 data alleviates the shift of the analysis spectrum value and the frequency order caused by the shift of the analysis data extraction interval, and makes the determination of the spectrum difference comparison more appropriate.

  Incidentally, the nst graph of Σ10 data is shown in FIG. 11, and the Σ10 frequency order bands a, b, and c where nst ≧ 3.0 are one frequency order band. In this embodiment, the frequency order is 10 units. The Σ10 analysis summarized in is considered appropriate.

  In addition, it is necessary to select an appropriate value for the total number p (Σp: p = 10 in this case) of frequency orders depending on the target equipment, sampling time, number of analysis data, and the like.

  As described above, according to the present embodiment, first, macro diagnosis is performed to calculate the correlation coefficient between the steady-state Σp spectrum and the Σp spectrum of the abnormality estimation unit, and when the correlation coefficient is lower than a predetermined value, it is determined that there is an abnormality. Thus, since the micro diagnosis is executed to identify the abnormal frequency band, it is possible to accurately detect signs of abnormality before the equipment failure.

  Moreover, since the measurement data is divided and calculated for each predetermined number of collected data in each diagnosis, it is possible to detect an abnormality with high accuracy with simple calculation even for an abnormal sound in a band whose frequency is slightly shifted.

It is a functional block diagram of an equipment abnormality diagnosis system according to an embodiment of the present invention. It is a block diagram of the sound collector of FIG. FIG. 3A is a graph of the sound data of the compressor, FIG. 3A is a graph of the measurement data of the propagation sound in the solid, and FIG. 3B is a graph of the measurement data of the propagation sound in the air. It is a flowchart which shows the flow of the equipment abnormality diagnosis method by embodiment of this invention. 3 is an enlarged view of a beard-like pulse portion of a compressor displayed on an output unit 20. FIG. FIG. 6A is a graph of the frequency analysis result of the compressor, FIG. 6A is a graph of the frequency analysis result of the abnormality estimation unit, and FIG. 6B is a graph of the frequency analysis result of the stationary part. It is a table | surface of the correlation coefficient of the compressor sound data spectrum calculated by the macro diagnostic means of FIG. FIG. 8A is a graph of the standard deviation multiple calculated by the micro diagnostic means of FIG. 1, FIG. 8A is a standard deviation multiple graph of the difference spectrum of all data, and FIG. 8B is a standard deviation multiple graph of the Σ10 difference spectrum. It is. It is a graph of the specific range and stationary part data including the abnormality estimation part of the compressor by the Example of this invention. FIG. 10 is a graph of standard deviation multiples of the difference spectra of all data when the extraction sections of the abnormality estimation part and the steady part in FIG. 9 are changed to 1a to 5a and 1b to 5b, respectively. It is a graph of the standard deviation multiple of (SIGMA) 10 data by a present Example. It is a data block diagram of the failure mode prediction table by embodiment of this invention. It is a data block diagram of the failure time prediction table by embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Equipment abnormality diagnosis system 10 Input part 11 A / D conversion means 20 Output part 21 Measurement data input means 22 Listening sound result input means 23 Fourier coefficient calculation means 24 Waveform generation means 25 Macro diagnosis means 26 Micro diagnosis means 27 Determination result output means 30 Calculation unit 50 Storage unit 51 Measurement data storage means 52 Measurement target spectrum data storage means 53 Steady-state spectrum database 54 p-value table 60 Sound collector 61 Microphone 62 Solid-state propagation sound collection microphone container 63 Solid-state propagation sound transmission substance 64 Collection Sound part 70 Recorder 71 Storage means 72 Output means 80 Equipment

Claims (7)

An equipment abnormality diagnosis method that collects sound data generated from equipment and diagnoses the presence or absence of abnormality by frequency analysis using a computer,
Collecting steady-state sound data for each facility, and storing the collected data as steady-state data;
Collecting sound data of equipment and storing it as measurement data;
Finding the frequency spectrum of the section including the stationary part data and the abnormal estimation part of the measurement data, calculating a correlation coefficient between values obtained by dividing and summing the two spectrum values for a predetermined number of frequency orders, A macro diagnosis step for determining presence / absence of abnormality based on the correlation coefficient, and each value of the difference spectrum of the aggregated value for each of the two spectrum values when the macro diagnosis step determines that there is an abnormality Calculating a standard deviation multiple indicating how many times the standard deviation of the entire difference data is, and identifying the abnormal frequency band by the standard deviation multiple,
An equipment abnormality diagnosis method comprising: The frequency analysis is an analysis by Fourier transform, and the Fourier coefficient of the sound data is calculated by the following equation, where m is the number of discrete data and N is the maximum frequency contained therein. The equipment abnormality diagnosis method according to claim 1, wherein
  2. The macro diagnosis step of determining presence / absence of abnormality by comparing a frequency spectrum value of the division including an abnormality estimation unit and a total value of frequency spectrum values of a plurality of divisions adjacent to the division. Or the facility abnormality diagnosis method according to 2;   The equipment according to any one of claims 1 to 3, further comprising a step of displaying a ratio between the value of the measurement data and a restoration value when the original data is restored from the frequency spectrum of the measurement data. Abnormal diagnosis method.   The equipment abnormality diagnosis method according to any one of claims 1 to 4, further comprising a step of accumulating frequency spectrum data of the measurement data and displaying a trend of secular change.   6. The method according to claim 1, further comprising: collecting the propagation sound in the solid of the facility as sound data, and specifying an abnormal portion estimated by listening sound or the displayed data waveform. Equipment abnormality diagnosis method. An equipment abnormality diagnosis system that collects sound data generated from equipment and diagnoses the presence or absence of abnormality by frequency analysis,
Steady state spectrum database storing steady state sound data for each facility as steady part data;
Measurement data input means for collecting equipment sound data and storing it as measurement data;
A p-value table for storing the number (p) of frequency order of sound data for each facility;
With reference to the p-value table, the number of collected items corresponding to the equipment to be measured is extracted, the frequency spectrum of the section including the stationary part data of the equipment and the abnormal estimation part of the measured data is obtained, and the two spectrum values A macro diagnostic means for calculating a correlation coefficient between values obtained by dividing and summing up each frequency order, and determining whether there is an abnormality based on the correlation coefficient;
If the macro diagnostic means determines that there is an abnormality, the standard deviation indicating how many times the standard deviation of the difference data of each value of the difference spectrum of the aggregate value for each of the two spectral values is A micro diagnostic means for calculating a multiple and identifying an abnormal frequency band by the standard deviation multiple;
An equipment abnormality diagnosis system characterized by comprising: