JP3495129B2 - Plant abnormality detection device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a plant abnormality detecting device for early and automatically detecting abnormality occurring in plant equipment constituting a plant.

[0002]

2. Description of the Related Art For example, in a large-scale plant such as a power plant using a gas turbine or a steam turbine, in order to detect a failure or a decrease in efficiency of plant equipment constituting a plant such as a rotating machine, a valve, or a pipe, at an early stage. , A plant abnormality detection device is provided. This plant abnormality detection device inputs detection signals from various detectors, processes the detection signals, and automatically detects an abnormality of a monitoring target. The detection signal input to the plant abnormality detection device is not only the process amount such as pressure, flow rate, or temperature that is used for supervisory control operation of the plant, but also the acoustic signal detected by a vibration sensor or microphone, or a television camera. A detection signal from an on-site sensor dedicated to abnormality monitoring, such as an image signal to be displayed, is also input.

On the other hand, in a large-scale plant, an enormous variety of equipment is operated in conjunction with each other, so that the respective detection signals are always associated with each other over time. Therefore, as a judgment method of abnormality detection in the plant abnormality detection device,
Deviation from the normal state of the plant equipment by accumulating a certain number of time series detection data and using statistical hypothesis test or applying a time series model considering such characteristics of the detection signal. The determination method of monitoring the is generally adopted.

FIG. 5 shows a block diagram of a plant abnormality detection apparatus using such a determination method. FIG. 5 is for detecting abnormality of the gas turbine 1 of the gas turbine power plant. In this case, the plant abnormality detection device 2 outputs a detection signal from the tachometer 3 which is a detector for detecting the rotation speed of the gas turbine 1 and a microphone 4 which is a detector for detecting vibration noise around the gas turbine 1. A determination method for detecting abnormality based on a plurality of series of detection signals together with the detection signal is adopted.

In FIG. 5, the state of the gas turbine 1 to be monitored is detected as an acoustic signal by the microphone 4 and is input to the input section 6 of the plant abnormality detection device 2 via the frequency filter 5 at a constant cycle. It Further, the rotation speed signal of the gas turbine 1 is detected by the rotation speed detector 3, and is input to the input unit 6 at a constant cycle like the acoustic signal. The sample data storage unit 7 stores a total of three types of change amounts of the high-frequency sound signal, the low-frequency sound signal, and the rotation speed signal as history data for a certain period. The threshold value calculation unit 8 calculates a threshold value for judging the deviation from the normal state based on the historical data of the respective variation amounts stored in the sample data storage unit 7. Then, the determination unit 9 compares the low-frequency sound signal, the high-frequency sound signal, and the rotation speed signal as the data to be determined with a threshold value, and determines that there is an abnormality when deviating. The output unit 10 reports the determination result to the reporting device 1
Output to 1 and notify the operator.

The threshold value and determination method in this case will be described with reference to FIGS. 6 and 7. First, FIG. 6 is a characteristic diagram showing a correlation characteristic between a low frequency sound signal level and a rotation speed signal in a normal state. The threshold value calculating means 8 calculates a correlation characteristic curve S1 between the low frequency sound signal level and the rotation speed signal in a normal state based on the history data, and calculates a low frequency range S2 shown by hatching in the figure. . Then, when the low-frequency sound signal S3, which is the data to be determined, is within the low-frequency permissible range S2, it is determined to be normal, and when it deviates, it is determined to be abnormal. FIG. 6 shows a case where the low frequency acoustic signal S3 deviates from the low frequency permissible range S2 and is located at a position away from the correlation characteristic curve S1 by the distance A.

Next, FIG. 7 is a characteristic diagram showing a correlation characteristic between a high frequency sound signal level and a rotation speed signal in a normal state. The threshold value calculation means 8 calculates a correlation characteristic curve S4 between the high frequency sound signal level and the rotation speed signal in a normal state based on the history data, and calculates a low frequency range S5 shown by hatching in the figure. . Then, when the low-frequency sound signal S6, which is the data to be determined, is within the low-frequency permissible range S5, it is determined to be normal, and when it deviates, it is determined to be abnormal. FIG. 7 shows a case where the high frequency acoustic signal S6 is within the high frequency permissible range S5 and is located at a position separated from the correlation characteristic curve S4 by a distance B.

That is, the threshold value calculating means 8 calculates the correlation characteristic S1 between the normal rotation speed signal and the low frequency sound signal and the correlation characteristic curve between the rotation speed signal and the high frequency sound signal based on the history data. S4 is calculated, and when the observed low-range acoustic signal S3 and high-range acoustic signal S6 that are the determined data deviate from the allowable ranges S2 and S5 of these correlation characteristics, it is determined to be abnormal. Here, since the observed value of the data to be judged contains noise, the allowable range corresponding to the dead zone equivalent to the noise level is used as a permissible value in the low-range permissible range shown by the diagonal lines in the figure when judging deviation. S2 and a high range allowable level S5 are set. Therefore, the threshold values in this case are the correlation characteristic characteristic curves S1 and S4 and the allowable value ranges S2 and S5.

[0009]

However, in a large-scale plant, due to the complicated structure of the plant equipment and the advanced operation method, many kinds of process state quantities are influenced by each other and fluctuate. There are many. Therefore, as shown in FIG. 5, there are cases where the process state quantity is not limited to three, and it is necessary to analyze the relationship with respect to three or more multidimensional process state quantities and manage many threshold values. Is coming out.

On the other hand, even when the movement of each process state quantity is related to the behavior of the plant, as shown in FIG. 7, there is a difference between the two process state quantities of the high frequency acoustic characteristic and the rotation speed. In the case of having a substantially linear correlation, if only the change ratio between the rotation speed and the high frequency sound signal can be quantified, it is managed as one process state quantity that combines the rotation speed and the high frequency sound signal. It becomes possible to do.

As described above, the same plant behavior is often represented by seemingly different process state quantities. In such a case, an abnormal determination is made by using an unnecessarily large number of process state quantities, and the man-hour required for factor analysis at the time of abnormality increases as the control value increases. In addition, if the process state quantity that is not closely related to the behavior of the plant is excessively caught in the abnormality determination, the reliability of the determination result may be impaired.

An object of the present invention is to manage a numerical value obtained by converting a large number of mutually correlated process state quantities into a small number of combined state quantities for use in abnormality determination, and to relatively speed up a large number of process state quantities. And it is to provide an abnormality detection device capable of highly accurately determining an abnormality.

[0013]

According to a first aspect of the present invention, there are provided an input section for fetching a plurality of process state quantities at a constant cycle through respective detectors, and each process state quantity fetched by the input section. The sample data storage unit that stores the history data of, and the same number of principal components as the process state quantities acquired based on each historical data are obtained, and eigenvectors and eigenvalues are calculated for each principal component corresponding to each principal component. The eigenvalue calculation unit that stores the values, and the contribution rate that selects all the principal elements whose cumulative contribution rate calculated by calculating each contribution rate for each principal component based on the eigenvalue and adding up in descending order of each contribution rate falls within the specified range. The judgment unit and the process state quantity at a constant cycle from the input unit are input as judgment data, and the judgment data and history are displayed on the Cartesian coordinate system with each main component selected by the contribution ratio judgment unit as a parameter. A distance calculation unit that calculates the distance to the data and a determination unit that determines that the distance is abnormal when the distance exceeds a threshold value and outputs the determination result to the notification device via the output unit are provided.

According to the second aspect of the present invention, the distance calculator converts the history data into an approximate pattern showing the characteristics of the history data by using an approximate expression in the distance calculation between the judgment data and the history data, and the judgment data is obtained. Calculate the distance between and the approximate pattern.

According to a third aspect of the present invention, the detector for detecting the data to be judged is a microphone for capturing airborne sound and a tachometer.

According to the present invention, the input section takes in the acoustic signal detected by the microphone as the data to be judged which is different for each frequency band.

[0017]

According to the first aspect of the invention, the input unit takes in each process state quantity of the judged data through the respective detectors at a constant cycle and samples the historical data of each process state quantity of the taken judged data. The data is stored in the data storage unit.
The eigenvalue calculation unit obtains the same number of principal components as the respective process state quantities fetched based on each history data, and calculates an eigenvector and an eigenvalue for each principal component corresponding to each principal component. The contribution rate determination unit calculates each contribution rate for each principal component based on the eigenvalue calculated by the eigenvalue calculation unit, and the cumulative contribution rate accumulated in descending order of each contribution rate falls within the specified range. Select the main component of.

On the other hand, the distance calculation unit, on the Cartesian coordinate system having each main component selected by the contribution rate determination unit as a parameter,
The distance between the judgment data fetched from the input unit and the history data is calculated. When the distance exceeds the threshold value, the determination unit determines that the abnormality is abnormal and outputs the determination result to the notification device via the output unit.

As a result, the principal component of the process state quantity related to the plant equipment to be monitored is calculated, and the abnormality determination is performed based on the principal component having a high contribution rate. Simplification and speedup can be achieved.

According to the second aspect of the present invention, the distance calculation unit converts the history data into an approximate pattern showing the characteristics of the history data by using an approximate expression and calculates the distance between the judgment target data and the approximate pattern. The processing speed can be increased.

According to the third aspect of the invention, since the detector for detecting the data to be judged is the microphone for capturing the airborne sound or the tachometer, it is suitable for detecting the abnormality of the rotating body.

According to the invention of claim 4, the input unit takes in the acoustic signal detected by the microphone as different judgment data for each frequency band, and obtains a plurality of judgment data from signals from the same detector. I have to.

[0023]

FIG. 1 is a block diagram showing an embodiment of the present invention. The plant abnormality detection device 2 of the present invention is different from the conventional plant abnormality detection device 2 shown in FIG. 5 in that the threshold value calculation unit 8 is replaced with a large process state quantity related to the plant equipment to be monitored. The eigenvalue calculation unit 12 for calculating the principal component and the eigenvalue thereof, the contribution rate determination unit 13 for selecting a principal component having a large influence on the plant equipment from the calculated principal components, and the contribution rate determination unit 13 The distance calculating means 14 for calculating the distance between the judgment target data and the history data is provided on the orthogonal coordinate system with each main component selected as a parameter.

That is, the process state quantity is accumulated as history data, the main component of the process state quantity related to the plant equipment to be monitored is calculated based on the history data, and the contribution rate thereof is taken into consideration. Create a Cartesian coordinate system for abnormality determination. And in the abnormality judgment,
With respect to the process state quantity input in a constant cycle, the abnormality determination is performed based on the calculated distance from the history data on the orthogonal coordinate system.

In FIG. 1, the state of the gas turbine 1 to be monitored is detected as an acoustic signal by the microphone 4 and is transmitted through the frequency filter 5 to the plant abnormality detection device 2
Is input to the input unit 6 of the above in a constant cycle. Further, the rotation speed signal of the gas turbine 1 is detected by the rotation speed detector 3, and is input to the input unit 6 at a constant cycle like the acoustic signal.

The microphone 4 captures the sound propagated in the air around the gas turbine 1 which is a plant facility. The acoustic signal detected by the microphone 4 is input to the input section 6.
Then, it is fetched as the data to be judged which is different for each frequency band. That is, it is captured as a high-frequency sound signal and a low-frequency sound signal.

A total of three types of change amounts of the high-frequency sound signal, the low-frequency sound signal, and the rotation speed signal fetched from these detectors are input to the sample data storage unit 7 and the distance calculation unit 14. The sample data storage unit 7 stores history data of a total of three types of process state quantities of a high-frequency sound signal, a low-frequency sound signal, and a rotation speed signal fetched from the detector.

The eigenvalue calculation unit 12 obtains the same number of principal components as the process state quantities based on the history data of the process state quantities stored in the sample data storage unit 7,
An eigenvector and an eigenvalue are calculated for each principal component corresponding to each principal component. Then, the contribution rate determination unit 13 calculates each contribution rate for each principal component based on the eigenvalue, and calculates all the principal components whose cumulative contribution rates integrated in descending order of each contribution rate fall within the range of the specified value. select. That is, the main component of the process state quantity having a large relation to the plant equipment to be monitored is selected.

The distance calculation unit 14 calculates a rectangular coordinate system with each principal component selected by the contribution rate determination unit 13 as a parameter, and the process state quantity fetched from the input unit at a constant cycle on the rectangular coordinate system. That is, the distance between the data to be judged and the history data in the sample data storage unit is calculated. The determination unit 9 is configured to determine that the distance is abnormal when the distance exceeds the threshold value, and output the determination result to the notification device 11 via the output unit 10.

Here, the creation of a rectangular coordinate system for abnormality determination will be described. First, the eigenvalue calculation unit 12 calculates the eigenvalue and eigenvalue vector of the main component of the process state quantity related to the plant equipment to be monitored according to the history data of the process state quantity in the following procedure. (A1) The history data is distributed to 1 for each process state quantity,
Normalize so that the average is zero.

[0031] Dispersion of rotation speed signal; V (x) (= 1) Dispersion of low-frequency acoustic signal; V (y) (= 1) Dispersion of high frequency sound signal; V (z) (= 1) (A2) Calculate the variance / covariance of historical data.

Covariance of the rotational speed signal and the low frequency sound signal;
C (x, y) Covariance of rotation frequency signal and high frequency sound signal; C (x,
z) Covariance of low-range acoustic signal and high-range acoustic signal; C (y,
z) (A3) Find the root of the characteristic equation for the variance-covariance matrix.

In this embodiment, since the number of process state quantities is 3, three roots of λ (λ1, λ) that satisfy the following characteristic equation
2, λ3) is obtained as the eigenvalue of the main component.

[0034]

[Equation 1] Here, the variance-covariance matrix S is shown below.

[0035]

[Equation 2] (A4) The size of the root is discriminated and rearranged.

The roots λ1, λ2, obtained in the procedure (A3),
λ3 is sorted in ascending order, that is, in descending order. Now, let the maximum eigenvalue be α1, the intermediate eigenvalue be α2, and the minimum eigenvalue be α3. The obtained maximum eigenvalue α1 is the eigenvalue of the first principal component, and the variance of the history data becomes the largest in the first principal component direction. Similarly, the intermediate eigenvalue α2 is the eigenvalue of the second principal component, and the minimum eigenvalue α3 is the eigenvalue of the third principal component. In the principal component analysis in which the number of process state quantities is 3, the variance of the history data is the smallest in the third principal component direction. (A5) Obtain an eigenvector for each eigenvalue.

The eigenvector (a1, b) of the maximum eigenvalue α1
1, c1) is calculated by the following formula.

[0038]

[Equation 3] Similar to the above equation, the eigenvector (a2, b) of the intermediate eigenvalue α2
2, c2) and the eigenvector (a3,
Calculate b3, c3).

The eigenvalues and eigenvectors of the principal components of the process state quantity calculated in this way are determined by the contribution rate judging unit 1.
Input to 3. In the contribution rate determination unit 13, first, based on the eigenvalues calculated by the eigenvalue calculation unit 12 in the steps (A1) to (A5), the main component that represents the behavior of the plant is selected by the following procedure. (B1) The contribution rate of each principal component is calculated.

In this embodiment, the contribution rate of each principal component is obtained by the following equation. This contribution rate is a ratio showing how much each principal component reflects the behavior related to plant operation.

Contribution rate h1 of the first principal component h1 = α1 / (α1 + α2 + α3) Contribution rate h2 of the second principal component h2 = α2 / (α1 + α2 + α3) Contribution rate h3 of the third principal component h3 = α3 / (α1 + α2 + α3) (B2) The cumulative contribution rate is compared with a specified value.

Contribution rates are arranged in descending order and integrated in order. The integration is continued until the integrated value falls within the range of the specified value, and the main component for which the integration is performed is selected.

For example, the contribution rate h1 of the first principal component is h1 =
0.6, the contribution rate h2 of the second principal component is h2 = 0.3, the third
It is assumed that the contribution rate h3 of the main component is h3 = 0.1, the upper limit of the specified value is 1.0, and the lower limit thereof is 0.8. in this case,
Since the contribution ratio h1 of the first principal component of the maximum value is 0.6,
This alone is within the specified value range (0.8 ≦ specified value ≦ 1.
Since it does not fall within 0), the contribution ratio h2 of the second largest second principal component is integrated with the contribution ratio h1 of the first principal component. Since the contribution rate h2 of the second principal component is 0.3, the cumulative contribution rate is h
1 + h2 = 0.9, which falls within the specified range. Therefore, in this case, the first principal component and the second principal component are selected. As described above, in the present invention, the main component of the process state quantity that is largely related to the abnormality of the plant equipment, that is, the first main component and the second main component, is selected to perform the abnormality determination.

Next, the distance calculation unit 14 creates an orthogonal coordinate system with each principal component selected by the contribution rate determination unit 13 as a parameter, and on the orthogonal coordinate system, the object captured from the input unit 6 is created. The distance between the determination data and the history data of the sample data storage unit 7 is calculated by the following procedure. (C1) The history data is normalized so that the variance is 1 and the average is 0 for each process state quantity that is taken in every fixed period that is the data to be determined. (C2) The inputted judgment target data X, Y, Z are made dimensionless by the following formula.

X '= X-M (x) / V (x) Y' = Y-M (y) / V (y) Z '= Z-M (z) / V (z) where X' , Y ′, Z ′ are the dimensionless data X, Y, Z, M (x) is the average value of the history data x before dimensionless, and M (y) is the dimensionless data before dimensionless. Average value of history data y, M (z) is the average value of history data z before dimensionlessization, V (x) is the variance of history data x before dimensionlessization, and V (y) is before dimensionlessization. Variance of historical data y, V
(Z) is the variance of the history data z before dimensionlessization. (C3) The combined state quantity is calculated.

Substituting each history data and judgment data normalized in the procedure (C1) and the procedure (C2) into the following equation,
The combined state quantity ur (i) of history data and the combined state quantity uh (i) of non-history data, that is, the data to be judged are calculated using the elements of the eigenvectors as coefficients. Here, i defines the principal component, and when i = 1, it means the first principal component, and when i = 2, it means the second principal component.

Combined state quantities ur (1), uh (1) ur (1) = a1 · x + b1 · y + c1 · z uh (1) = a1 · X ′ + b1 · Y ′ + c1 · Z ′ of the first principal component Two state components ur (2), uh (2) ur (2) = a2.x + b2.y + c2.z uh (2) = a2.X '+ b2.Y' + c2.Z ' FIG. 2 shows a plot of each combined state quantity of the second principal component on the orthogonal coordinate system. The first component is plotted on the horizontal axis and the second principal component is plotted on the vertical axis. In FIG. 2, the history data 15 is indicated by black dots, and the judgment data 1 which is non-history data.
6 is shown by a white dot. The combined state quantity of the first principal component is maximized by the sum of squares of the correlation coefficient between the process state quantities of the history data. Line segments A to E are line segments used in the distance calculation described later. (C4) Calculate the distance.

The distance to the data to be judged is calculated as the combined state quantity calculated in step (C3). In this embodiment, the Euclidean square distance is used in the following equation as the distance calculation method.

[0049]

[Equation 4] Here, L is the number of the selected main components described above. In this example, L = 2.

When the Mahalanobis distance is calculated as the distance calculation method, the following formula may be used.

[0051]

[Equation 5] In this embodiment, as indicated by line segments A to E in FIG. 2 on the Cartesian coordinates of the combined state quantity, the distance between each history data existing in the direction of a certain angle around the judgment data and the judgment data. To calculate. (C5) Calculation of Frequency Distribution Corresponding to the line segments A to E in FIG. 2, the frequency distribution is formed as shown in FIG. 3 with the distance between each history data on the line segment and the judgment target data as a class value. (C6) Selection of minimum distance From the frequency distribution tables in FIG. 3, the one with the largest median value is selected. In this embodiment, B in FIG. 3 corresponds to this.

Next, the judging section 9 compares the minimum distance of the procedure (C6) with the threshold value, and if the minimum distance exceeds the threshold value, judges that it is abnormal. The output unit 10 outputs the determination result to the notification device 11.

As described above, according to this embodiment, only the main components that represent the behavior associated with plant operation are extracted,
Since the difference from the normal time is determined, it is possible to perform highly accurate abnormality determination even on the history data and the determined data that include a lot of noise that is low in relation to the behavior of the plant. In addition, by combining the frequency distribution with the principal component analysis as in this embodiment, it is possible to make a probability-based abnormality determination.

Next, another embodiment of the present invention will be described. In the above-described embodiment, the distance between the judgment target data and the history data is calculated according to the frequency distribution of the history data in the Cartesian coordinate system having the selected main component as a parameter as shown in FIG. As shown in FIG. 4, a large number of historical data 15 are converted into an approximate pattern 17 showing the characteristics of the historical data using an approximate expression, and then the judgment target data 16 is obtained.
It is also possible to calculate the distance between and the approximate pattern 17.

That is, in the distance calculation unit 14, the distance from the history data in the sample data storage unit 7 is calculated by the following procedure on the Cartesian coordinate system with each principal component selected by the contribution rate determination unit 13 as a parameter. calculate. (D1) Similar to the procedure C1, the history data is normalized for each variable so that the variance is 1 and the average is 0. (D2) As in the procedure (C2), the input determination target data is made dimensionless. (D3) Similar to the procedure (C3), the combined state quantity is calculated. (D4) The data of the synthetic state quantity calculated in the conversion procedure (D3) to the approximate pattern is set to at least 2
Convert to an approximate expression by multiplication. In this other embodiment,
The number of main components is L = 2 as in the above embodiment. Therefore, the approximate expression can be expressed as one fitted to the two-dimensional approximate pattern 17 as shown in FIG. (D5) The distance between the approximate pattern 17 and the determination target data 16 is calculated.

The distance between the approximate pattern 17 calculated in D4 and the judged data 16 is calculated. In this embodiment, the minimum distance in the Euclidean square distance between the approximate expression and the data to be judged is obtained as shown by A in FIG.

The judging section 9 compares the obtained distance with a threshold value, and if it exceeds the threshold value, judges that it is abnormal. The output unit 10 outputs the determination result to the notification device 11.

According to this embodiment, when a large amount of history data is included, a high speed abnormality determination can be performed by reducing the calculation amount by performing comparison with the determined data by an approximate expression.

[0059]

As described above, according to the present invention, a plurality of process state quantities are converted into a small number of synthetic variables to judge the difference from the normal time. Highly accurate abnormality determination is possible even for history data and determined data that include a lot of noise that is low in relation to. In other words, since the input process state quantity itself is not fixed and the difference from the normal time is not judged, variations in multiple process state quantities caused by the complicated structure of plant equipment and advanced operation method It is possible to make a highly accurate abnormality determination even with respect to. Further, it is possible to improve the efficiency of the work related to abnormality detection by reducing the number of process state quantities that need to be managed.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention.

FIG. 2 is an explanatory diagram of a Cartesian coordinate system created by a distance calculation unit of the present invention for abnormality determination.

FIG. 3 is an explanatory diagram of distance calculation between the judgment target data and history data in the distance calculation unit of the present invention.

FIG. 4 is an explanatory diagram of another distance calculation by the distance calculation unit of the present invention.

FIG. 5 is a block configuration diagram of a conventional plant abnormality detection device.

FIG. 6 is an explanatory diagram of abnormality determination for a low frequency sound signal by a conventional determination method.

FIG. 7 is an explanatory diagram of abnormality determination for a high frequency sound signal by a conventional determination method.

[Explanation of symbols]

1 gas turbine 2 Plant abnormality detection device 3 rotation speed detector 4 microphones 5 frequency filters 6 Input section 7 Sample data storage 8 threshold 9 Judgment section 10 Output section 11 Notification device 12 Eigenvalue calculator 13 Contribution rate judgment section 14 Distance calculator 15 Historical data 16 Judgment data 17 Approximate pattern

Continuation of front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) G05B 23/02 G01D 21/00

Claims (4)

(57) [Claims] 1. A plant in which a plurality of process state quantities representing the state of plant equipment to be monitored are fetched in a fixed cycle, and whether or not there is an abnormality in the plant equipment is detected based on the fetched process state quantities. In the anomaly detection device, an input unit that captures the plurality of process state quantities through a respective detector at a constant cycle, and sample data that stores history data about each of the process state quantities captured by the input unit. A storage unit, and an eigenvalue calculation unit that calculates the eigenvectors and eigenvalues of each of the principal components in correspondence with each of the principal components by obtaining the same number of principal components as the process state quantities fetched based on the history data. And a cumulative contribution rate calculated by calculating each contribution rate for each principal component based on the eigenvalue and integrating the contribution rates in descending order. Of the main components selected by the contribution rate determining section and the process state quantity for each constant period from the input section as the data to be determined. A distance calculation unit that calculates the distance between the determination target data and the history data on a rectangular coordinate system that is a parameter, and if the distance exceeds a threshold value, it is determined to be abnormal, and the determination result is output to the output unit. A plant abnormality detection device, comprising: a determination unit that outputs the information to a notification device through the plant abnormality detection device. 2. The distance calculation unit converts the history data into an approximate pattern indicating characteristics of history data by using an approximate expression in calculating a distance between the judgment data and the history data, The plant abnormality detection apparatus according to claim 1, wherein a distance between the plant abnormality detection device and the approximate pattern is calculated. 3. A detector for detecting the judged data comprises:
The plant abnormality detection device according to claim 1 or 2, which is a microphone for capturing airborne sound and is a tachometer. 4. The plant abnormality detection device according to claim 3, wherein the input unit captures the acoustic signal detected by the microphone as determination target data that differs for each frequency band.