JP2007051982A - Method and apparatus for evaluating object of diagnosis - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an evaluation method which enables accurate evaluation of the state of an object of evaluation to automatically determine the presence of abnormalities or the like, and to provide an evaluation apparatus. <P>SOLUTION: The change the degrees of reference state signal, acquired preliminarily from the object of evaluation and the measuring state signal acquired at measurement are calculated by statistical inspection or the like, by using the statistical values of frequency spectral components and the state of the evaluation target is determined, on the basis of the inspection result. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

The present invention relates to a technique for evaluating the state of an evaluation object, and in particular, the state of the evaluation object is highly reliable using state signals such as vibrations, sounds, and AE signals detected from the evaluation object. The present invention relates to a novel evaluation method and evaluation apparatus for a diagnostic object that can be efficiently evaluated.

  In various devices, systems, natural objects, etc., it may be necessary to determine the normal or abnormal state by determining the state thereof, or to determine the necessity of maintenance. Specifically, for example, in machine tools such as NC processing machines, production facilities such as semiconductor manufacturing plants, turbines in power plants, home appliances such as air conditioners and refrigerators, and Yamano where there is a risk of collapse, In order to continuously realize a stable operation and to avoid occurrence of danger, it is effective to determine whether or not the state is normal.

  However, it is often difficult to easily determine the state of such various evaluation objects simply by observing from outside.

  Therefore, conventionally, as a state reflection signal reflecting the state of the evaluation object, for example, sound, vibration, AE signal or the like is detected, and this state reflection signal is analyzed, so that the occurrence of an abnormality in the evaluation object can be quickly detected. It has been proposed to detect. The state reflection signal is analyzed by observing the maximum value of the state reflection signal and detecting a peak value exceeding the threshold value to determine that there is an abnormality, or a specific frequency obtained from the state reflection signal by Fourier calculation. Frequency analysis and the like for determining an abnormality by detecting a component are known. (See Patent Document 1)

  By the way, the state reflection signal detected from the evaluation object by an acceleration sensor, a voice coil, or the like includes many noise signals that are not related to the state of the evaluation object. Therefore, in order to determine the state of the evaluation object with high accuracy and increase the reliability of the determination result, it is required to remove such a noise signal. For this reason, conventionally, by using a low-pass filter, a high-pass filter, or a band-pass filter, a noise signal is removed from the state-reflecting signal and only a signal in a specific frequency range that is considered to reflect the state of the evaluation object is extracted. The extracted signal in the specific frequency range is analyzed.

  However, when only a signal in a specific frequency region is extracted from the state reflection signal using such a filter, it is not clear whether only a signal that really reflects the state of the evaluation object is extracted. The determination of the cut-off frequency by the filter is performed only empirically or by trial and error.

  Therefore, the conventional state evaluation method and state evaluation device require skill when used, and the objective reliability and stability of the determination result are not necessarily high. Moreover, even if a skilled person operates the signal, it is impossible to deny the possibility that even a signal that is significant in determining the state of the evaluation object has been removed. It was not possible to wipe it off.

JP 2002-372400 A

Here, the present invention has been made in the background as described above, and the problem to be solved is that the state of the evaluation object can be easily and accurately determined and the occurrence of abnormality is high. An object of the present invention is to provide a novel evaluation method and evaluation apparatus for an evaluation object that can be detected with reliability.

  Hereinafter, the aspect of this invention made | formed in order to solve such a subject is described. In addition, the component employ | adopted in each aspect as described below is employable by arbitrary combinations as much as possible. Further, aspects or technical features of the present invention are not limited to those described below, but are described in the entire specification and drawings, or an invention that can be understood by those skilled in the art from those descriptions. It should be understood that it is recognized based on thought.

(Aspect 1 of the present invention relating to an evaluation method of an evaluation object)
In the aspect 1 of the present invention relating to the evaluation method of the evaluation object, when the state of the evaluation object is evaluated based on the state reflection signal detected from the evaluation object, the state reflection signal is used under the condition as a determination criterion. A plurality of reference state signals over a unit time length are obtained by detecting from the evaluation object, and frequency spectrum information is obtained for each of the plurality of reference state signals, while the state reflection signal is evaluated under the state to be measured. At least one measurement state signal over a unit time length is obtained by detecting from an object, frequency spectrum information is obtained with respect to the measurement state signal, frequency spectrum information obtained with respect to the reference state signal, and the measurement state For each frequency spectrum information obtained for the signal, a statistical value is obtained and based on the reference state signal. Based on the degree of difference between the statistical value based on a total value as the measurement state signal, in the evaluation method of diagnosis target object so as to evaluate the measured state of the evaluation object.

  According to such a method of the present invention, it is possible to evaluate a state to be measured using a reference state such as a normal state of an evaluation object as a determination criterion. In short, in the measurement state, the state of the evaluation object can be evaluated only for the portion that has changed from the reference state. Therefore, it is not necessary to remove all the signals in the frequency range uniquely set in advance using the filter means as in the conventional method.

  Therefore, according to the method of the present invention, it is possible to prevent the information signal from being inadvertently deleted by the intention of the operator, and it is possible to easily and efficiently extract a significant signal for evaluation. It becomes. The state of the evaluation object can be determined with high accuracy based on the extracted state signal.

  In addition, in the method of the present invention, a plurality of reference state signals of the evaluation object are obtained, and the state is evaluated using their statistical values (average value, variance, etc.). Reliability and stability can be improved.

  If a more detailed aspect is illustrated, the evaluation method according to the present invention may take the following aspects. First, in order to determine the state of an object, a spectrum of a signal (reference state signal) measured in a reference state (determination reference state) is obtained by FFT, and a statistical value (average value or variance) of spectrum components at each frequency. Etc.) in advance. Next, in order to perform a state diagnosis, a spectrum of a signal (measurement state signal) newly measured under a state to be measured is obtained by FFT, and a statistical value of a spectrum component of a reference state obtained in advance at each frequency ( The degree of change of the spectrum component of each frequency with respect to the spectrum in the reference state is obtained by a statistical test or the like (for example, a spectral component difference calculation value described later). Further, a spectral component having a large change is extracted, and the presence / absence of a state change and the state type are identified using a frequency domain feature parameter (for example, a feature parameter described later).

  In the aspect 1 of the present invention relating to the evaluation method of the evaluation object, it is generally employed that (e) the unit time length in the reference state signal is the same as the unit time length in the measurement state signal. .

  Thus, by aligning the unit time lengths of the reference state signal and the measurement state signal, the detection conditions for both state signals can be made closer. Therefore, it is possible to further improve the determination accuracy based on the comparison of both statistical values obtained from the reference state signal and the measurement state signal. However, in the present invention, it is possible to perform the corresponding processing even if the time lengths of both signals are not equal to each other.

(Aspect 2 of the present invention relating to an evaluation method of an evaluation object)
Aspect 2 of the present invention relating to an evaluation method of an evaluation object is the evaluation method according to aspect 1 described above, wherein a signal having a predetermined time length is acquired as the state reflection signal under a state as the determination criterion. A plurality of the reference state signals are obtained by dividing the state reflection signal into a plurality of parts with a constant division time length, and a signal having a predetermined time length is obtained as the state reflection signal under the state to be measured. Then, a plurality of the measurement state signals are obtained by dividing the state reflection signal into a plurality of parts with a constant division time length.

  According to this aspect, it is possible to easily and efficiently acquire a plurality of reference state signals and measurement state signals. Further, by obtaining a plurality of measurement state signals in addition to the reference state signal in this way, it is possible to further improve the determination accuracy based on the difference between the respective statistical values.

(Aspect 3 of the present invention relating to an evaluation method of an evaluation object)
Aspect 3 of the present invention relating to the evaluation method of the evaluation object is the evaluation method according to aspect 2 described above, and is divided when dividing the state reflection signal acquired under the condition used as the criterion. The division time length is set so that the average value and the standard deviation of the absolute values of the respective data in the plurality of obtained reference state signals are approximately equal, and obtained under the state to be measured. When dividing the state reflection signal into a plurality, the division time length is set so that the average value and the standard deviation of the absolute values of the respective data in the plurality of measurement state signals obtained by the division are approximately equal. It is characterized by setting.
In addition, that the average value and the standard deviation of the absolute values of each data are approximately equal means that the following [Formula 1] can be satisfied.
(Where i ≠ j, j = 1 to N, M _i and M _j are the numbers of the i-th and j-th divided data, respectively, and t (α, ∞) is the probability density function of the t distribution is the upper probability α (Also, α = 0.00001 to 0.9, and more preferably 0.005 to 0.3.)

  According to this aspect, each of the plurality of reference state signals and the plurality of measurement state signals is approximately a steady signal, and both the reference state signal and the measurement state signal are obtained with higher accuracy. Is possible. In particular, even when the measurement state signal is a non-stationary signal over the entire time length, the division time length is set to be sufficiently short, and the above conditions (the average value of the absolute value of each data and By dividing so as to satisfy (equal standard deviation), it is possible to obtain approximately a steady signal.

(Aspect 4 of the present invention relating to an evaluation method of an evaluation object)
Aspect 4 of the present invention relating to an evaluation method for an evaluation object is the evaluation method according to any one of the aforementioned aspects 1 to 3, wherein the statistical value of the frequency spectrum information related to the reference state signal and the measurement An average value and a standard deviation are employed as the statistical values of the frequency spectrum information related to the state signal, respectively.

  According to this aspect, the frequency spectrum information of the reference state signal and the measurement state signal can be grasped efficiently and easily by adopting the average value and the standard deviation as the statistical values. Therefore, by adopting such a statistical value, for example, as illustrated in Aspect 7 to be described later, the difference (degree of change) between the frequency spectrum information of the reference state signal and the frequency spectrum information of the measurement state signal is advantageously reduced. It becomes possible to evaluate.

(Aspect 5 of the present invention relating to an evaluation method of an evaluation object)
Aspect 5 of the present invention relating to an evaluation method for an evaluation object is the evaluation method according to any one of the aforementioned aspects 1 to 4, wherein the statistical value based on the reference state signal and the statistical value based on the measured state signal In order to evaluate the degree of difference from the values, a spectral component difference calculation value representing the magnitude of the difference between the frequency spectral components in both statistical values is used.

  In this mode, for example, even if the vibration level component is changing even if the vibration level as a whole is hardly changed between the reference state and the measurement state, the change is grasped and abnormality is detected. Can also be detected. In particular, this aspect is more suitably realized as the following aspects 6 and 7.

(Aspect 6 of the present invention relating to an evaluation method of an evaluation object)
Aspect 6 of the present invention relating to an evaluation method for an evaluation object is an identification index obtained by the following formula: [Equation 2] for each frequency spectrum component as the spectrum component difference calculation value in the evaluation method according to aspect 5 described above: It is characterized in that at least one of the average value difference: DA obtained by the following formula: [Equation 3] is adopted for each DI and frequency spectrum component.

(Aspect 7 of the present invention relating to an evaluation method of an evaluation object)
Aspect 7 of the present invention relating to the evaluation method of the evaluation object is the evaluation method according to aspect 5 or 6 described above, wherein a threshold is set for the calculated spectral component difference, and the calculated spectral component difference is larger than the threshold. Determining whether or not, extracting a characteristic spectrum component reflecting the characteristic of the state change based on the determination result, and evaluating the measurement target state in the evaluation object using this characteristic spectrum component, Features.

  In this aspect, it is possible to accurately and more easily determine an abnormal state or the like by efficiently targeting only the frequency spectrum component that has changed compared to the reference state.

(Aspect 8 of the present invention relating to the evaluation method of the evaluation object)
Aspect 8 of the present invention relating to an evaluation method for an evaluation object is the evaluation method according to aspect 7 described above, wherein at least one of the characteristic parameters described in (1) to (8) below is used. It is characterized by evaluating the measurement object state in the evaluation object by grasping the characteristic.
(1) “Residual spectrum power” calculated by the following formula
(2) "Spectrum residual power" calculated by the following formula
(3) "Average characteristic frequency" calculated by the following formula
(4) “Frequency of closing time average per unit time” calculated by the following formula
(5) “Waveform stability index” calculated by the following formula
(6) “Variation rate” calculated by the following formula
(7) “Strain” found by the following formula
(8) “kurtosis” calculated by the following formula

  In this aspect, it is possible to accurately grasp the characteristics of the characteristic spectrum component using mathematical formulas, and it is possible to more efficiently determine the state of the evaluation object.

(Aspect 9 of the present invention relating to an evaluation method of an evaluation object)
Aspect 9 of the present invention relating to the evaluation method of the evaluation object is the evaluation method according to any one of the aforementioned aspects 1 to 8, wherein M (p _{1 to} p _M ) arbitrary dimensionless feature parameters are selected. From the M feature parameters, the following formula:
The integrated feature parameter represented by: Z is adopted,
As the reference state in the evaluation object, two different states, state n and state a, are selected, and the N reference state signals obtained in the state n and the N pieces obtained in the state a are obtained. The integrated feature parameters: Z _ni and Z _ai (i = 1 to N) are obtained based on the reference state signal, respectively, and an average value of the N integrated feature parameters: Z _{ni in} the state n: μ _Zn and standard deviation: S _Zn and, N pieces of the integrating feature parameters in the state a: average value of Z _ai: mu _Za and standard deviation: with S _Za, the following equation:
While obtaining the absolute criterion: DI _Z
Based on the measurement state signal obtained in the measurement target state of the evaluation object, the integrated feature parameter: Z is obtained,
According to the MA minimax absolute determination method shown in the following determination condition I and determination condition II, the measurement object state in the evaluation object is evaluated.
Determination condition I: If μ _Zn > μ _Za , state n when Z> DI _Z , state a when Z <DI _Z.
Judgment condition II: If μ _Zn <μ _Za , state n when Z <DI _Z , state a when Z> DI _Z.

  In this aspect, generally, for the state n and the state a, a category or the like to be obtained as a determination result is selected. More specifically, a normal state is selected as the state n, and an abnormal state is selected as the state a. In addition, different abnormal states may be selected as the state n and the state a. In short, in this aspect, as the determination result, it is determined which of the two states selected in advance as the reference state: the state n and the state a is the measurement target state in the evaluation object. Therefore, two different states to be obtained as determination results are selected as these two reference states: state n and state a.

On the other hand, in the measurement target state, an integrated feature parameter: Z is obtained from the obtained measurement state signal. Then, taking into account the relationship between this Z value and the above-described two reference states: the information of absolute state criterion: DI _Z obtained from the information of state n and state a, the measurement target state is two reference states. : It is determined whether the state is the state n or the state a.

In this embodiment, the dimensionless feature parameters: p _{1 to} p _M that are considered when obtaining the integrated feature parameter: Z are degrees of difference in statistical values between the reference state signal and the measurement state signal. Appropriate reflected parameters can be adopted, more suitably appropriate parameters reflecting the characteristic spectrum components can be adopted, and more preferably, the above-described aspects (1) to (9) in the aspect 8 can be adopted. An appropriate number is appropriately selected from the characteristic parameters. Either two may be selected, or all nine may be selected. The selection is made in accordance with the evaluation state of the measurement target state or the type of the measurement target in consideration of, for example, past determination experience data and experimental results. In addition, the value of a in the integrated feature parameter Z is a constant that is appropriately set with reference to, for example, past determination experience data and experimental results, according to the weighting ratio to be considered in each corresponding feature parameter. .

(Aspect 10 of the present invention relating to an evaluation method of an evaluation object)
Embodiment 10 of the present invention relates to method for evaluating evaluation target is the evaluation method according to the embodiment 9 described above, the absolute criterion: lower instead of DI _Z formula:
Relative criteria represented by: DI _Zn , DI _Za are adopted,
Instead of the determination condition I and the determination condition II, the measurement object state in the evaluation object is evaluated according to the MA minimax relative determination method shown in the following determination condition III.
Determination Condition III: State n if DI _Zn <DI _Za , State a if DI _Zn > DI _Za .

  According to the above-described aspects 9 and 10 of the present invention, any effective and simple determination based on the integrated feature parameter can be realized. The effectiveness of these determinations is also apparent from Example 3 to be described later, in which the normal state is adopted as the state n as the two reference states and the abnormal state is adopted as the state a.

(Aspect 11 of the present invention relating to an evaluation object evaluation apparatus)
Aspect 11 of the present invention relating to the evaluation method of the evaluation object is the evaluation method according to any one of the aforementioned aspects 1 to 10, wherein the reference state signal and the measurement state signal are obtained from the state reflection signal. It is characterized by performing at least one signal processing of signal amplification, noise removal, and target frequency band filtering.

  In this aspect, information that should not be considered in the state evaluation can be removed in advance from the state reflection signal detected from the evaluation object. Thereby, it becomes possible to further improve the accuracy of state evaluation.

(Aspect 1 of the present invention relating to an evaluation apparatus for an evaluation object)
A feature of the aspect 1 of the present invention relating to the evaluation object evaluation apparatus is that (a) a state reflection signal detecting means for detecting a state reflection signal including information reflecting the state from the evaluation object; And (c) a reference state signal acquisition unit for acquiring a plurality of reference state signals over a unit time length from the state reflection signal detected by the state reflection signal detection unit under a state as a determination reference for the evaluation object; ) Reference spectrum information calculation means for obtaining frequency spectrum information for the reference state signal acquired by the reference state signal acquisition means; and (d) a statistical value of the frequency spectrum information of the reference state signal obtained by the reference spectrum information calculation means. And (e) the state reflected signal detecting means detected by the state reflected signal detecting means in a state to be measured by the evaluation object. Measurement state signal acquisition means for acquiring a measurement state signal over a unit time from the state reflection signal, and (f) measurement spectrum information calculation means for obtaining frequency spectrum information for the measurement state signal acquired by the measurement state signal acquisition means, (G) a measurement state statistical value calculation unit that calculates a statistical value of frequency spectrum information of the measurement state signal obtained by the measurement spectrum information calculation unit; and (h) a statistical value obtained by the reference state statistical value calculation unit; A comparison calculation means for comparing and calculating the statistical value obtained by the measurement state statistical value calculation means; (i) a determination means for determining the state of the evaluation object from the comparison calculation result obtained by the comparison calculation means; (J) The diagnostic object evaluation apparatus includes display means for displaying the determination result by the determination means to the outside.

  In the evaluation apparatus having the structure according to this aspect, the method of the present invention as described above can be advantageously performed, and thereby the same effect as that obtained by the method of the present invention can be obtained. It becomes.

  In particular, in this aspect, (a) as the state reflection signal detection means, for example, a sensor such as a force sensor such as an acceleration sensor or a strain sensor, or a sound sensor such as a voice coil is attached to the evaluation object, and an electrical vibration signal is detected. In addition, the control system provided in the evaluation object itself is used as the state reflection signal detection means when the evaluation object is provided with a control computer such as an internal combustion engine for automobiles. It is also possible to use an appropriate signal in such a control system as a state reflection signal.

  Further, in this aspect, (b) the reference state signal acquisition unit and (e) the measurement state signal acquisition unit are signals that divide the state reflection signal obtained by the state reflection signal detection unit at a predetermined time. It can be constituted by a processing device. In addition, (c) the reference spectrum information calculation means and (f) the measured spectrum information calculation means use the calculation means based on the fast Fourier calculation (FFT calculation) for the data processed with the A / D converter as necessary. Depending on the processing arrangement, it can be advantageously realized. In addition, (d) reference recitation singing value calculation means, (g) measurement state statistic value calculation means, (h) comparison calculation means, and (i) determination means are all calculated according to a program stored in advance in a ROM or the like. It is advantageously configured by an arithmetic processing unit configured using a CPU or the like that performs processing. Further, (j) the display means only needs to display the determination result so that it can be recognized by a person. In addition to an image display device such as a CRT or a liquid crystal monitor, various display devices such as a sound display device and a light display device can be used. Can be employed, and may be employed in appropriate combination.

  As is clear from this, the evaluation apparatus having the structure according to the present invention can be configured by a computer or the like in addition to the state reflected signal detection means configured by a sensor or the like, thereby Evaluation processing can be performed automatically and at high speed without requiring manual adjustments and processing operations.

  Further, in this aspect, preferably, (d) a storage unit that stores the statistical value of the frequency spectrum information obtained by the reference state statistical value calculating unit is provided, and thereafter (g) the measurement state statistical value calculating unit appropriately And (h) the statistical value stored in the storage means when the calculation process is performed by the comparison calculation means.

(Aspect 2 of the present invention relating to an evaluation apparatus for an evaluation object)
Aspect 2 of the present invention relating to an evaluation apparatus for an evaluation object is the evaluation apparatus according to aspect 1, further comprising: (k) whether the state reflected signal detected by the state reflected signal detecting means is a steady signal or an unsteady signal. And (l) when the state reflection signal is determined to be an unsteady signal by the signal discrimination means, so that the state reflection signal can be processed approximately as a steady signal. Signal dividing means for obtaining the reference state signal and the measurement state signal by dividing the state reflection signal into a short time length so that the average value and standard deviation of the absolute values of the divided data are approximately equal. And a stationary signal converting means.
Note that being approximately equal means that the following [Equation 19] can be satisfied.
(Where i ≠ j, j = 1 to N, M _i and M _j are the numbers of the i-th and j-th divided data, respectively, and t (α, ∞) is the probability density function of the t distribution is the upper probability α (Also, α = 0.00001 to 0.9, and more preferably 0.005 to 0.3.)

In this aspect, for example, even if the state reflection signal is a non-stationary signal, it can be treated as a stationary signal approximately by dividing it into a sufficiently short time length, and evaluation based on statistical processing according to the present invention can be performed. It can be applied.

  As is clear from the description of each aspect of the present invention described above and the description of the embodiment of the present invention described later, in the present invention, the determination of the presence or absence of a change in the state of the evaluation object and the identification of the state type are performed by complicated human resources. This operation can be performed easily and with high accuracy without the need for this operation.

Therefore, for example, the present invention can provide a highly accurate condition diagnosis algorithm in the field of actual condition diagnosis and pattern recognition, and is useful for automation of condition diagnosis by a computer and realization of a diagnosis apparatus.

  Hereinafter, in order to clarify the present invention more specifically, embodiments of the present invention will be described. First, FIG. 1 shows a flow of data processing of various signals as state signals in the state diagnosis method as one embodiment of the present invention. As the status signal, a vibration signal, an acoustic signal, an AE (Acoustic Emission) signal, or the like measured by a signal measuring unit such as various known sensors is preferably used.

  As shown in FIG. 1, the flow of data processing is roughly divided into “preparation for diagnosis” and “execution of diagnosis”. Here, we will explain the detailed algorithm along this flow.

1. Preparation for diagnosis (1) Determine the reference state and frequency band.
The reference state is a state used as a reference when determining whether or not there is a change in the state of the diagnostic object (evaluation object). For example, in the case of equipment diagnosis, the reference state is often “normal state”.
Further, the frequency band referred to here is a frequency range from the minimum frequency to the maximum frequency considered in diagnosis, and the characteristic spectrum of each state to be diagnosed is obtained in this frequency range.

(2) Measure the signal in the reference state.
A signal (reference signal) reflecting the state of the diagnostic object is measured with a sufficient length of time by an acceleration sensor or the like attached to the diagnostic object so that subsequent statistical processing can be performed. By dividing this reference signal by an appropriate time length as will be described later, a reference state signal over a unit time length is acquired. Further, the sampling frequency is determined based on the result of the examination in advance of the vibration frequency to be considered for the diagnosis according to the characteristics of the diagnostic object.

(3) When the reference signal is a stationary signal, the spectrum of the reference state is obtained at each frequency. Here, the stationary signal is a signal whose average value and variance of the signal are substantially constant with time.
That is, the N reference state signals are obtained by equally dividing the reference signal obtained by measurement with a sufficient length of time into N times. Then, a frequency spectrum is obtained by FFT for each of the N reference state signals. According to statistical theory, it is desirable that N> 5.

(4) In the case of an unsteady signal, the measured signal is divided into short time-length signals (referred to as divided data) so that the measured signal can be processed approximately as a steady signal. The time length of the divided data (called the divided time length) is determined as follows.
First, an absolute value (referred to as absolute value data) of signal levels of N divided data is obtained. When N pieces respectively mu ₁ the average value and the standard deviation of the absolute value data in each data segment of ～Myu _N and S ₁ to S _N, giving significance level alpha, the hypothesis μ ₁ = μ ₂ = ··· = Μ _i = ... = μ _N is tested. That is, the division time length is shortened until this hypothesis holds.
For example, when the significance level α is given, the above hypothesis is established if the following mathematical formula is established.
Here, i ≠ j, i, j = 1 to N, M _i and M _j are the numbers of the i-th and j-th divided data, respectively, and t (α, ∞) is the probability density function of the t distribution is the upper probability. Percentage point for α. In general, α = 0.00001 to 0.9, and more preferably 0.005 to 0.3.

(5) For each of the divided N reference state signals, the frequency spectrum is obtained in the same manner as (3) above.

(6) At each frequency, the statistical value of the frequency spectrum component in the reference state is obtained.
At each frequency, typical statistics of the spectral components in the reference state are as follows: The spectrum component of the j-th reference state at the frequency f _i is F _j (f _i ), and i = 1 to I. I is the number of divisions of the frequency spectrum.
1) Average value An average value is calculated | required by the numerical formula shown next.
2) Standard deviation The standard deviation is obtained by the following mathematical formula.

(7) The statistical values of the spectral components in the reference state are stored in the database.
The statistical values of the spectral components of the reference state obtained in the above (6) are stored in a database constituted by a RAM as storage means, various storage devices, etc., and used for state diagnosis described later. .

2. Execution of diagnosis (1) A signal for state diagnosis is measured.
Signals (measurement signals) are measured in the same measurement conditions as the reference state (various conditions such as the detection means (sensor) to be used and the mounting position of the detection means on the diagnostic object) and the same frequency band. The measurement signal length of the measurement signal is the shortest and 1 / N of the measurement time length of the reference signal (N is the number of reference signal obtained by dividing the reference signal). More preferably, based on statistical theory, the measurement time length of the measurement signal is set to 5 / N or more of the measurement time length of the signal in the reference state.

(2) In the case of a stationary signal, the degree of change with respect to the spectrum component in the reference state is determined by statistical test or the like at each frequency f _i . Here, the stationary signal is a signal whose average value and variance of the signal are substantially constant with time.
That is, similarly to the above-described processing of the reference signal, an appropriate number of measurement state signals are acquired by equally dividing a measurement signal obtained by measurement with a predetermined time length into an appropriate number. Then, the frequency spectrum is obtained by FFT for each of the appropriate number of measurement state signals.

(3) In the case of an unsteady signal, an appropriate number of measurement state signals are obtained by dividing the measured signal into a sufficiently short time length so that the measured signal can be processed approximately as a steady signal. The division theory is the same as the processing of the reference signal described above.

(4) The spectrum F _u (f _i ) of each divided signal is obtained in the same manner as the above-described reference state signal.

(5) At each frequency, the degree of change with respect to the spectrum component in the reference state is evaluated by the statistical values (average value and standard deviation) shown in the above [Equation 21] and [Equation 22]. In order to evaluate the degree of change, the following method is preferably used.

1) Identification index method The identification index DI is defined by the following equation.
Here, μ ₁ and μ ₂ are average values of frequency spectrum components under two different states (state 1 and state 2), and σ ₁ and σ ₂ are standard deviations thereof.
At this time, the average value and the standard deviation of the spectrum component F _u (f _i ) at the frequency f _i of the signal measured for diagnosis are expressed as follows.
When i = 1 to I, the identification index DI (f _i ) is obtained by the following equation.
In short, the greater the identification index DI (f _i ), the greater the difference from the spectral component in the reference state at the frequency f _i .
When the time length of the measurement signal measured for diagnosis is 1 / N of the time length of the reference signal, the average value and standard deviation of the spectrum component F _u (f _i ) at the frequency f _i can be obtained. Absent. Therefore, using the spectrum component F _u (f _i ) at the frequency f _i , the identification index DI (f _i ) is obtained by the following equation.

2) Mean value difference test method The maximum estimated value of the mean value difference DA (f _i ) is obtained by the following equation.
Here, the values shown below are percentage points of the probability density function of the standard normal distribution with respect to the lower probability α / 2, and generally α is set to 0.02 to 0.2.
N _u is the number of data used when obtaining the average value and standard deviation of the spectral component F _u (f _i ) at the frequency f _i . The average value and standard deviation are expressed as follows.
The average value difference DA (f _i ) represents the magnitude of the spectral component difference between the two, similarly to the identification index DI (f _i ), but it is desirable that the number of data N and N _u is greater than 5.

(6) At each frequency, a threshold value for extracting a spectral component F _u (f _i ) having a large degree of change in the measurement state with respect to the spectral component in the reference state is determined.
The threshold value of the identification index DI (f _i ) is generally 1 to 12, more preferably 3 to 5, and the threshold value of the average value difference DA (f _i ) is defined as the following equation.
That is, if the average value difference DA (f _i ) and the identification index DI (f _i ) are larger than the above threshold, it is considered that the difference from the spectrum component in the reference state is large at the frequency f _i .

(7) At each frequency, a spectral component F _u (f _i ) having a large degree of change with respect to the spectral component in the reference state is extracted.
When the threshold value of the average value difference DA (f _i ) or the identification index DI (f _i ) set at the frequency f _i is not exceeded, the current spectral component F _u (f _i ) is set to zero and the other spectral component F _{u is set.} (F _i ) is extracted as a spectrum reflecting the characteristics of the state change.

(8) A feature parameter in the frequency domain for expressing the feature of the extracted spectral component is calculated.
As characteristic parameters in the frequency domain, the following p _f1 to p _f9 are preferably obtained.

1) Residual spectral power The residual spectral power p _f1 is obtained by the following equation.
Here, F _u (f _i ) is a spectral component that does not exceed the threshold value of the average value difference DA (f _i ) or the identification index DI (f _i ) set at the frequency f _i .

2) Spectral residual power The spectral residual power _pf2 is obtained by the following equation.
Here, F _u (f _i ) is a spectral component that does not exceed the threshold value of the average value difference DA (f _i ) or the identification index DI (f _i ) set at the frequency f _i , or is a total spectral component. . Hereinafter, the same applies to the p _f3 ~p _f9 shown in Equation 14 to Equation 22.

3) Average characteristic frequency The average characteristic frequency _pf3 is obtained by the following equation.

4) Frequency of closing time average per unit time Frequency _{f f4} of closing time average per unit time is obtained by the following formula.

5) Waveform Stability Index The waveform stability index _pf5 is obtained from the following equation.

6) Fluctuation rate Fluctuation rate _pf6 is obtained by the following equation.
Here, each parameter in _pf6 is obtained by the following formulas.

7) Skewness The skewness _pf7 is obtained by the following equation.

8) Kurtosis The kurtosis p _f8 and p _f9 are obtained by the following equations.

(9) The presence / absence of the state change and the state type are identified using the value of each feature parameter.
When the value of the characteristic parameter in the time domain or the frequency domain is calculated, the presence / absence of state change is diagnosed or the state type is identified. Here, a feature parameter integration method is preferably used. Examples include “self-reorganization of frequency domain feature parameters by genetic programming” disclosed in the following Non-Patent Document 1, “principal component analysis method” and “discriminant analysis” disclosed in Non-Patent Document 2. Law ".

Chen, Toyoda: Self-reorganization of frequency domain feature parameters by genetic programming, Transactions of the Japan Society of Mechanical Engineers (C), 65 (633), pp. 1946-1953, 1999. Wakimoto Kazumasa et al .: PC Statistical Analysis Handbook II Multivariate Analysis, Kyoritsu Publishing Co., Ltd., 1984.

  Here, an example of state diagnosis using the state diagnosis method according to an embodiment of the present invention when the state signal is a steady signal (AE signal) will be described.

FIG. 2A shows an example of a time-series signal in a reference state (here, “normal state”) measured with the frequency band set to 0 to 500 kHz.
FIG. 2B is an example of the spectrum in the reference state. FIGS. 2C and 2D are the average value and standard deviation of the spectrum in the reference state. In addition, in order to obtain the average value and standard deviation of the reference state, time series signals of 15 reference states were used.
FIGS. 2E and 2F are examples of a time-series signal measured for diagnosis and its spectrum, respectively.
FIG. 2 (g) is an identification index DI (f _i ) obtained by Equation 7.
In this example, there are two states to be diagnosed: a “normal state” and an “abnormal state”. The number of cases to be diagnosed is 30.
FIG. 2 (i) shows the residual spectrum power p _f1 represented by Equation 12 obtained using the spectrum of the time series data measured at the time of each diagnosis with the threshold value of the identification index DI (f _i ) being 3.

  In this example, the data numbers 1, 2, 5, 6, 7, 8, 9, 15, 17, 19, 21, 25, 26, 28, 29 are “normal”, and 3, 4, 10, 11, 12, 13, 14, 16, 18, 20, 22, 23, 24, 27, 30 are "abnormal conditions".

FIG. 2 (i) shows that the remaining spectrum power p _f1 in the “normal state” is almost 0, whereas the remaining spectrum power p _{f1 in} the “abnormal state” is large. In this case, since the value of the minimum residual spectral power p _{f1 in} the “abnormal state” is about 0.002, if the threshold value of the residual spectral power p _f1 is set to 0.0015, “normal state” and “abnormal state”. Can be identified.

  Here, an example of a state diagnosis using the state diagnosis method as one embodiment of the present invention when the state signal is an unsteady signal (AE signal) will be described.

FIG. 3 shows an example of unsteady signals of “normal state” and “abnormal state”. These signals are particularly unsteady in the latter half. Therefore, the latter half of the waveform data was divided into three for processing.
FIG. 4 shows an example of spectra of “normal state” and “abnormal state” corresponding to the time-series waveform division unit of FIG.
FIG. 5 shows the average value of the spectra of “normal state” and “abnormal state” corresponding to each division unit.
FIG. 6 shows the spectrum remaining power _pf2 obtained by Expression 12 obtained using the spectrum of the time series data of each division unit with the threshold value of the identification index DI (f _i ) set to 3 as in the previous example.

Also in this example, data numbers 1, 2, 5, 6, 7, 8, 9, 15, 17, 19, 21, 25, 26, 28, 29 are “normal states”, and 3, 4, 10, 11 , 12, 13, 14, 16, 18, 20, 22, 23, 24, 27, 30 are "abnormal conditions". FIG. 6 shows that the spectrum remaining power p _f2 in the “normal state” is almost 0, whereas the spectrum remaining power p _{f2 in} the “abnormal state” is large.

In this case, about the value of the minimum spectral residual power p _f2 of the "abnormal state" in the first, second, and third dividing each 0. 08,0. 02,0. 03 is because the threshold of the spectral residual power p _f2 If they are set to 0.07, 0.01, and 0.02, respectively, it is possible to distinguish between “normal state” and “abnormal state”.

FIG. 7 shows an average value of the spectrum remaining power _pf2 in the “normal state” and the “abnormal state” in the first, second, and third divisions. In this case, since the minimum mean value of the spectrum remaining power p _f2 of the "abnormal state" is about 0.05, by setting the threshold value of the spectral residual power p _f2 to 0.04, the "normal state", "abnormal state Can be identified.

  Here, an example of state diagnosis by integration of frequency domain feature parameters is shown.

FIGS. 8A and 8B are examples of time-series signals (vibration acceleration) in a normal state and an abnormal state, respectively, measured by setting the frequency band to 0 to 50 kHz.
FIGS. 8C and 8D are examples of spectra in a normal state and an abnormal state, respectively.
In order to distinguish between a normal state and an abnormal state, frequency domain characteristic parameters p _{f3 to} p _f9 are used. First, the normal state and abnormal state spectra shown in FIGS. 8C and 8D are obtained N times, respectively. Using these spectra, N are obtained for p _f3 to p _f9 in the normal state and the abnormal state, respectively.
By integrating canonical discriminant analysis, p _f3 to p _f9 are integrated into the following equation to obtain an integrated feature parameter Z.

In this example, a ₁ , a ₂ , a ₃ , a ₄ , a ₅ , a ₆ , and a ₇ are -1.24, -1.62, -33.58, 8.30, and -0.07, respectively. 1.91, -1.24. Here, the integrated feature parameters obtained from the normal state and abnormal state spectra are Z _ni and Z _ai (i = 1 to N), respectively. _Assuming that the average value and standard deviation of Z _ni and Z _ai are μ _Zn , μ _Za and S _Zn , and S _Za , respectively, the absolute judgment criterion DI _Z is given by the following equation.

That is, the method for determining the state using the absolute determination reference DI _Z at the time of the state diagnosis is as follows. This determination method is called MA minimax determination method.
If mu _Zn> mu _Za, when Z> DI _Z, and determined to be normal, when Z <DI _Z, determined as abnormal.
If μ _Zn <μ _Za, it is determined to be normal when Z <DI _Z , and it is determined to be abnormal when Z> DI _Z.

Alternatively, the state is determined based on the relative state determination criteria DI _Zn and DI _Za obtained by the following equations.
If DI _Zn <DI _Za, it is determined as normal, and if DI _Zn > DI _Za, it is determined as abnormal.

FIG. 9 shows an example of DI _Zn and DI _Za values obtained from normal state data. In this case, since DI _Zn <DI _Za, it can be correctly determined as normal.
FIG. 10 shows an example of the values of DI _Zn and DI _Za obtained from the abnormal state data. In this case, since DI _Zn > DI _Za, it can be correctly determined as abnormal.

  FIG. 11 is a block diagram schematically showing the configuration of a diagnostic apparatus using the state diagnostic method according to an embodiment of the present invention.

This apparatus includes a signal acquisition unit 10 for measuring vibration signals, acoustic signals, AE signals, etc., a signal processing unit 12 for amplification of measured signals, noise removal, filtering, etc., and the state diagnosis processing shown in FIG. A diagnosis processing unit 14 for diagnosing the state according to the flow and a display unit 16 for displaying the diagnosis result and state information are included.

It is a figure which shows the flow of the state diagnostic process in one Embodiment of the state diagnostic method according to this invention. It is a graph which shows the example of a stationary signal (AE signal) and a state diagnosis. It is a graph which shows the example of the unsteady signal (AE signal) of a normal state and an abnormal state. It is a graph which shows the example of the spectrum of the unsteady signal (AE signal) of a normal state and an abnormal state. It is a graph which shows the average value of the spectrum of a normal state corresponding to each division part of an unsteady signal, and an abnormal state. It is a graph which shows the spectrum residual power of the normal state and abnormal state corresponding to the spectrum of the time series data of each division | segmentation part of an unsteady signal. It is a graph which shows the average value of the spectrum residual power of the normal state and abnormal state in each division part of an unsteady signal. It is a graph which shows the example of the time series signal (vibration acceleration) of a normal state and an abnormal state, and a spectrum. It is a graph which shows the value of DI _Zn and DI _Za when the data of a normal state are input. It is a graph which shows the value of DI _Zn and DI _Za when the data of an abnormal condition are input. It is a figure which shows the structure of the diagnostic apparatus which shows one Embodiment of the diagnostic apparatus using the diagnostic method according to this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Signal acquisition part 12 Signal processing part 14 Diagnosis processing part 16 Display part

Claims (13)

When evaluating the state of the evaluation object based on the state reflection signal detected from the evaluation object,
While obtaining a plurality of reference state signals over a unit time length by detecting the state reflection signal from the evaluation object under the condition as a determination reference, while obtaining frequency spectrum information for each of the plurality of reference state signals,
By obtaining at least one measurement state signal over a unit time length by detecting the state reflection signal from the evaluation object under the state to be measured, obtaining frequency spectrum information regarding the measurement state signal,
A statistical value is obtained for each of the frequency spectrum information obtained for the reference state signal and the frequency spectrum information obtained for the measurement state signal, and a statistical value based on the reference state signal and a statistical value based on the measurement state signal are calculated. A diagnostic object evaluation method, comprising: evaluating a measurement target state of the evaluation object based on a degree of difference. While obtaining a signal having a predetermined time length as the state reflection signal under the condition used as the determination reference, the state reflection signal is divided into a plurality of pieces with a predetermined division time length to obtain a plurality of the reference state signals. ,
A plurality of measurement state signals are obtained by acquiring a signal having a predetermined time length as the state reflection signal under the state to be measured and dividing the state reflection signal into a plurality of divisions with a predetermined division time length. Item 2. The diagnostic object evaluation method according to Item 1. When dividing the state reflection signal acquired under the condition to be used as the determination reference, the absolute value of each data in the plurality of the reference state signals obtained by the division: μ and the standard deviation: S However, the division time length is set so as to satisfy the following formula: [Equation 1],
When dividing the state reflection signal acquired under the condition to be measured into a plurality, the average value of the absolute value of each data in the plurality of measurement state signals obtained by division: μ and the standard deviation: S The diagnostic object evaluation method according to claim 2, wherein the division time length is set so as to satisfy the following formula: [Equation 1].
(Where i ≠ j, j = 1 to N, M _i and M _j are the numbers of the i-th and j-th divided data, respectively, and t (α, ∞) is the probability density function of the t distribution is the upper probability α (Percentage point for .alpha. = 0.00001 to 0.9.)   4. The average value and the standard deviation are employed as the statistical value of the frequency spectrum information related to the reference state signal and the statistical value of the frequency spectrum information related to the measurement state signal, respectively. Method for evaluating diagnostic objects.   In order to evaluate the degree of difference between the statistical value based on the reference state signal and the statistical value based on the measurement state signal, a spectral component difference calculation value representing the magnitude of the difference between the frequency spectral components in both statistical values is used. The diagnostic object evaluation method according to any one of claims 1 to 4. As the spectral component difference calculation value, for each frequency spectrum component, an identification index obtained by the following formula: [Equation 2]: DI and for each frequency spectrum component, an average value difference obtained by the following formula: [Equation 3]: at least of DA The diagnostic object evaluation method according to claim 5, wherein one of them is adopted.
  A threshold is set for the spectrum component difference calculation value, and it is determined whether or not the spectrum component difference calculation value is larger than the threshold, and a feature spectrum component reflecting the state change feature is extracted based on the determination result The diagnostic object evaluation method according to claim 5, wherein the measurement target state of the evaluation object is evaluated using the characteristic spectrum component. The measurement target state in the evaluation object is evaluated by grasping the feature of the feature spectrum component using at least one of the feature parameters described in the following (1) to (8). Evaluation method for diagnostic objects.
(1) “Residual spectrum power” calculated by the following formula
(2) “Spectral residual power” calculated by the following formula
(3) "Average characteristic frequency" calculated by the following formula
(4) “Frequency of closing time average per unit time” calculated by the following formula
(5) “Waveform stability index” calculated by the following formula
(6) “Variation rate” calculated by the following formula
(7) “Strain” found by the following formula
(8) “kurtosis” calculated by the following formula
Arbitrary dimensionless feature parameters are selected from M (p _{1 to} p _M ), and the following formula is obtained from these M feature parameters:
The integrated feature parameter represented by: Z is adopted,
As the reference state in the evaluation object, two different states, state n and state a, are selected, and the N reference state signals obtained in the state n and the N pieces obtained in the state a are obtained. The integrated feature parameters: Z _ni and Z _ai (i = 1 to N) are obtained based on the reference state signal, respectively, and an average value of the N integrated feature parameters: Z _{ni in} the state n: μ _Zn and standard deviation: S _Zn and, N pieces of the integrating feature parameters in the state a: average value of Z _ai: mu _Za and standard deviation: with S _Za, the following equation:
While obtaining the absolute criterion: DI _Z
Based on the measurement state signal obtained in the measurement target state of the evaluation object, the integrated feature parameter: Z is obtained,
The diagnostic object evaluation method according to any one of claims 1 to 8, wherein a measurement target state of the evaluation object is evaluated according to an MA minimax absolute determination method indicated by the following determination condition I and determination condition II. .
Determination condition I: If μ _Zn > μ _Za , state n when Z> DI _Z , state a when Z <DI _Z.
Judgment condition II: If μ _Zn <μ _Za , state n when Z <DI _Z , state a when Z> DI _Z. Absolute criteria: In place of DI _Z :
Relative criteria represented by: DI _Zn , DI _Za are adopted,
The evaluation of the diagnostic object according to claim 9, wherein the measurement object state in the evaluation object is evaluated according to the MA minimax relative determination method shown in the following determination condition III instead of the determination condition I and the determination condition II. Method.
Determination Condition III: State n if DI _Zn <DI _Za , State a if DI _Zn > DI _Za .   11. When obtaining the reference state signal and the measurement state signal from the state reflection signal, at least one signal processing of signal amplification, noise removal, and target frequency band filtering is performed. Method for evaluating diagnostic objects. State reflected signal detection means for detecting a state reflected signal including information reflecting the state from the evaluation object;
Reference state signal acquisition means for acquiring a plurality of reference state signals over a unit time length from the state reflection signal detected by the state reflection signal detection means under a state as a determination reference for the evaluation object;
Reference spectrum information calculation means for obtaining frequency spectrum information for the reference state signal acquired by the reference state signal acquisition means;
Reference state statistic value calculating means for calculating a statistic value of frequency spectrum information of the reference state signal obtained by the reference spectrum information calculating means;
Measurement state signal acquisition means for acquiring a measurement state signal over a unit time from the state reflection signal detected by the state reflection signal detection means under a state to be measured of the evaluation object;
Measurement spectrum information calculation means for obtaining frequency spectrum information for the measurement state signal acquired by the measurement state signal acquisition means;
A measurement state statistic value calculating means for determining a statistic value of frequency spectrum information of the measurement state signal obtained by the measurement spectrum information calculating means;
A comparison calculation means for comparing and calculating the statistical value obtained by the reference state statistical value calculation means and the statistical value obtained by the measurement state statistical value calculation means;
Determination means for determining the state of the evaluation object from the comparison calculation result obtained by the comparison calculation means;
A diagnostic object evaluation apparatus comprising: display means for displaying the determination result by the determination means to the outside. Signal determining means for determining whether the state reflected signal detected by the state reflected signal detecting means is a stationary signal or an unsteady signal;
If the state reflection signal is determined to be an unsteady signal by the signal discriminating means, the average value of the absolute values of the divided data so that the state reflection signal can be processed approximately as a steady signal: μ and standard deviation: signal dividing means for obtaining the reference state signal and the measurement state signal by dividing the state reflected signal into a short time length that satisfies the following formula: [Equation 19];
(Where i ≠ j, j = 1 to N, M _i and M _j are the numbers of the i-th and j-th divided data, respectively, and t (α, ∞) is the probability density function of the t distribution is the upper probability α (Percentage point for .alpha. = 0.00001 to 0.9.)
The apparatus for evaluating a diagnostic object according to claim 12, further comprising a stationary signal converting means comprising: