JP2012194936A - Abnormality diagnostic device for plant equipment - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an abnormality diagnostic device for plant equipment that can detect operation having a nonsteady movement as deterioration or abnormality even when process data is within a prescribed range.SOLUTION: The abnormality diagnostic device for plant equipment includes singular matrix extraction means 13 of generating a matrix from the process data and extracting a left singular matrix close to an average from a left singular matrix obtained by performing singular value decomposition of the generated matrix; feature quantity extraction means 14 of calculating a matrix representing a feature quantity from the extracted left singular matrix and process data in a normal state, and extracting a maximum value and a minimum value of a first row of the matrix as feature quantities in the normal state; feature quantity analysis means 16 of calculating a matrix representing a feature quantity of process data to be diagnosed using the left singular matrix having the process data to be diagnosed extracted, and determining respective elements of the first row of the matrix as feature quantities to be diagnosed; and comparison means 17 of comparing the feature quantity in the normal state with the feature quantities to be diagnosed.

Description

  The present invention relates to an abnormality diagnosis apparatus for plant equipment that diagnoses deterioration and abnormality of equipment used in a power plant, a water treatment plant, an industrial plant, or the like.

  In a conventional plant monitoring and control system, the status of equipment used in a plant is collected as process data at regular intervals, and equipment malfunction is monitored by monitoring whether the collected process data values are within a specified range. Judgment is made (see, for example, Patent Document 1).

JP-A-11-095833 (paragraphs [0041] to [0047])

  In the apparatus of Patent Document 1, when the process data is within a specified range, such as deterioration that the characteristics of equipment used in the plant gradually change with time, or abnormalities in which the process data temporarily varies, There is a problem that cannot be determined as deterioration or abnormality.

  The present invention has been made to solve the above-described problems, and even if the process data is within a specified range, if an unsteady movement is observed in the operation, the deterioration or abnormality is present. It aims at providing the abnormality diagnosis apparatus of the plant apparatus which can be detected as.

The abnormality diagnosis device for plant equipment according to the present invention creates a matrix from a database that stores process data of a plant and one type of process data stored in the database, performs singular value decomposition of the matrix, The singular matrix extraction means for extracting the left singular matrix closest to the average from the obtained left singular matrices and storing the extracted left singular matrix in the database, the extracted left singular matrix and the normal state The feature amount extraction means for calculating a matrix representing the feature amount of the process amount using the process data, extracting the maximum value and the minimum value of the first row of the calculated matrix as the feature amount in the normal state and storing it in the database The process data to be diagnosed is read from the database, and a matrix representing the feature quantity of the process data to be diagnosed is calculated using the extracted left singular matrix The feature amount analyzing means for storing each element of the first row of the calculated matrix in the database as the feature amount of the diagnosis target, the feature amount of the normal state and the feature amount of the diagnosis target are read from the database and compared, and the comparison result is Comparing means for storing in a database.

  Further, the abnormality diagnosis apparatus for plant equipment according to the present invention creates a matrix from a database storing plant process data and a plurality of types of process data stored in the normal state, and performs singular value decomposition of the matrix. A singular matrix extracting means for extracting a left singular matrix closest to the average of the obtained left singular matrices and storing the extracted left singular matrix in a database; and the extracted left singular matrix A matrix representing the feature quantity of the process amount is calculated using multiple types of process data in the normal state, and the maximum value and the minimum value of the first row of the calculated matrix are extracted as the feature quantity in the normal state and stored in the database. Feature quantity extraction means, and multiple types of process data to be diagnosed are read from the database, and multiple types of diagnostic targets are extracted using the extracted left singular matrix A feature amount analyzing means for calculating a matrix representing the feature amount of the process data, and storing each element of the first row of the calculated matrix in the database as a feature amount of the diagnosis target; a feature amount of the normal state; and a feature amount of the diagnosis target And comparing means for reading out the data from the database and comparing them, and storing the comparison result in the database.

The abnormality diagnosis device for plant equipment according to the present invention creates a matrix from a database that stores process data of a plant and one type of process data stored in the database, performs singular value decomposition of the matrix, The singular matrix extraction means for extracting the left singular matrix closest to the average from the obtained left singular matrices and storing the extracted left singular matrix in the database, the extracted left singular matrix and the normal state The feature amount extraction means for calculating a matrix representing the feature amount of the process amount using the process data, extracting the maximum value and the minimum value of the first row of the calculated matrix as the feature amount in the normal state and storing it in the database The process data to be diagnosed is read from the database, and a matrix representing the feature quantity of the process data to be diagnosed is calculated using the extracted left singular matrix The feature amount analyzing means for storing each element of the first row of the calculated matrix in the database as the feature amount of the diagnosis target, the feature amount of the normal state and the feature amount of the diagnosis target are read from the database and compared, and the comparison result is Comparing means stored in the database is provided, so that it is possible to detect abnormalities such as deterioration gradually changing with time and temporary value fluctuations that cannot be detected by simple upper and lower limit checks of process data.

  Further, the abnormality diagnosis apparatus for plant equipment according to the present invention creates a matrix from a database storing plant process data and a plurality of types of process data stored in the normal state, and performs singular value decomposition of the matrix. A singular matrix extracting means for extracting a left singular matrix closest to the average of the obtained left singular matrices and storing the extracted left singular matrix in a database; and the extracted left singular matrix A matrix representing the feature quantity of the process amount is calculated using multiple types of process data in the normal state, and the maximum value and the minimum value of the first row of the calculated matrix are extracted as the feature quantity in the normal state and stored in the database. Feature quantity extraction means, and multiple types of process data to be diagnosed are read from the database, and multiple types of diagnostic targets are extracted using the extracted left singular matrix A feature amount analyzing means for calculating a matrix representing the feature amount of the process data, and storing each element of the first row of the calculated matrix in the database as a feature amount of the diagnosis target; a feature amount of the normal state; and a feature amount of the diagnosis target Compared with reading means from the database for comparison and storing the comparison result in the database, the load on the operator monitoring the plant is reduced and the relationship between multiple types of process data changes. It is also possible to determine characteristic fluctuations that are difficult to detect by monitoring individual process data.

BRIEF DESCRIPTION OF THE DRAWINGS It is a whole system block diagram which concerns on the abnormality diagnosis apparatus of the plant equipment of Embodiment 1 of this invention. BRIEF DESCRIPTION OF THE DRAWINGS It is a system block diagram which concerns on the abnormality diagnosis apparatus of the plant equipment of Embodiment 1 of this invention. It is a structure diagram of a database according to the abnormality diagnosis apparatus for plant equipment according to Embodiment 1 of the present invention. It is function explanatory drawing which concerns on the abnormality diagnosis apparatus of the plant equipment of Embodiment 1 of this invention. It is a structure diagram of a database according to the abnormality diagnosis apparatus for plant equipment according to Embodiment 1 of the present invention. It is function explanatory drawing which concerns on the abnormality diagnosis apparatus of the plant equipment of Embodiment 1 of this invention. It is a structure figure of the database which concerns on the abnormality diagnosis apparatus of the plant equipment of Embodiment 2 of this invention. It is function explanatory drawing which concerns on the abnormality diagnosis apparatus of the plant equipment of Embodiment 2 of this invention. It is function explanatory drawing which concerns on the abnormality diagnosis apparatus of the plant equipment of Embodiment 3 of this invention. It is function explanatory drawing which concerns on the abnormality diagnosis apparatus of the plant equipment of Embodiment 4 of this invention. It is a system block diagram which concerns on the abnormality diagnosis apparatus of the plant equipment of Embodiment 5 of this invention. It is a flowchart figure which concerns on the abnormality diagnosis apparatus of the plant equipment of Embodiment 5 of this invention. It is a flowchart figure which concerns on the abnormality diagnosis apparatus of the plant equipment of Embodiment 6 of this invention.

Embodiment 1 FIG.
In the first embodiment, an abnormality diagnosis apparatus for plant equipment according to the present invention that detects deterioration or abnormality of plant equipment by creating a matrix from time-series data of process data, performing singular value decomposition, and extracting feature quantities. It is about.
Hereinafter, a first embodiment of the present invention will be described with reference to FIGS. FIG. 1 is an overall system configuration diagram relating to an abnormality diagnosis device for plant equipment, FIG. 2 is a system configuration diagram for an abnormality diagnosis device for plant equipment, FIGS. 3 and 5 are structural diagrams of databases, and FIGS. .

First, a system configuration of the plant equipment abnormality diagnosis apparatus according to Embodiment 1 of the present invention will be described.
FIG. 1 shows the overall configuration of a plant control and monitoring system including an abnormality diagnosis device 1 for plant equipment according to Embodiment 1 of the present invention.
The overall system extracts feature quantities from the process data and controls the plant equipment abnormality diagnosis device 1 for detecting deterioration and abnormality of plant equipment and devices, the plant monitoring device 2 for monitoring the state of the plant, and the plant. It is comprised from the controller 3 for performing, and the plant 5 which is a monitoring control object.
The plant equipment abnormality diagnosis apparatus 1, the plant monitoring apparatus 2, and the controller 3 are connected by a control network 4 and can transmit and receive data to and from each other.

The controller 3 has a function of sending control commands to various devices and apparatuses in the plant, and conversely collecting process data such as plant temperature. The plant monitoring device 2 collects process data collected by the controller 3 via the control network 4, accumulates the process data of the plant in the plant monitoring device 2 for plant monitoring, and is provided in the plant monitoring device 2. It can be displayed on a display device such as a liquid crystal monitor.
The plant equipment abnormality diagnosis apparatus 1 collects the process data collected by the controller 3 via the control network 4 and stores it in the plant equipment abnormality diagnosis apparatus 1 as history data of the plant process data. Thus, it has an abnormality diagnosis function for detecting deterioration and abnormality of plant equipment and devices.

FIG. 2 is a diagram illustrating an internal system configuration of the abnormality diagnosis apparatus 1 for plant equipment.
The plant equipment abnormality diagnosis apparatus 1 includes a data collection unit 11 that collects process data via the control network 4, a process quantity database 12 that is a database that stores the process data collected by the data collection unit 11, and functions and operations are later described. Singular matrix extraction means 13, feature quantity extraction means 14, feature quantity analysis means 16, and feature quantity database 15 which is a database for storing extracted and analyzed feature quantities.
Further, the abnormality diagnosis apparatus 1 for plant equipment compares the extracted and analyzed feature quantities to detect the presence or absence of abnormality, and displays the extracted and analyzed feature quantities and comparison results on the display device 19. Display means 18 is provided.

Next, functions and operations of the plant equipment abnormality diagnosis apparatus 1 according to Embodiment 1 of the present invention will be described with reference to the system configuration in FIG. 2, the structure diagram of the database in FIGS.
The data collection means 11 collects process data collected from the plant 5 by the controller 3 via the control network 4 at regular intervals, and stores the collected process data in the process quantity database 12.
The structure of the process quantity database 12 is as shown in FIG. 3, and the collected process data is stored together with the collected time.
In the process data stored in the process amount database 12, names are set individually. In the example of FIG. 3, names such as a boiler temperature, a steam generation amount, and a valve opening degree are set.
In addition to the process data (input data) collected from the plant 5, the process quantity database 12 stores output data that is a control command transmitted from the controller 3 to the plant 5 in the same format.

The following processing is performed on the process data related to the device that you want to diagnose.
First, in order to learn the normal state of plant equipment, the plant equipment abnormality diagnosis apparatus 1 collects data by the data collection means 11 for a certain period after plant operation. The process data collected by the data collection unit 11 is stored in the process quantity database 12.
After data sufficient for normal state learning is stored, the singular matrix extraction means 13 starts normal state learning.

With respect to the process data to be diagnosed, the process data at a certain time t is represented as a (t). In addition, it is assumed that process data up to time R is stored in the process amount database 12 in advance.
The singular matrix extraction means 13 reads the data of the process data to be diagnosed from the process quantity database 12 and creates the following matrix.
Note that the number of rows N and the number of columns P of the matrix are values smaller than the time R, and are freely set within a range in which a matrix can be created for each process data.

  For example, when t = 1, a matrix is created using the process data shown in FIG.

  The data matrix A (t) is decomposed into three matrices by performing singular value decomposition.

The matrix S (t) is an N-row P-column matrix having values only in the diagonal terms, and singular values are stored in the diagonal terms in descending order. U (t) is a left singular matrix with N rows and N columns, and V (t) is a right singular matrix with P rows and P columns, each being an orthogonal matrix.
Here, the subscript T on the upper right indicates transposition of the matrix.
This indicates that the matrix A (t) is obtained by linearly transforming the matrix S (t) V ^T (t) with the matrix U (t).
Since the matrix S (t) is a matrix having elements only in the diagonal terms and arranged in descending order, the first row of the matrix S (t) V ^T (t) shows the characteristics of A (t). It will be the best representation. In the following, the features represented by the second row, the third row, and the lower row become smaller.

  The generation of the matrix A (t) and the singular value decomposition of the matrix A (t) are performed on the entire area of the collected data. In the first embodiment, the matrix A (t) is generated and the singular value decomposition of the matrix A (t) is performed by changing from time t = 1 to time t = R−N−P + 2.

  The average value of R−N−P + 2 left singular matrices U (t) obtained by the singular value decomposition is calculated by the equation (3).

Here, U _AVE is a matrix having the average value of each element of R−N−P + 2 left singular matrices U (t) as an element, and hereinafter, the element in the i-th row and j-th column of U _AVE will be uave _ij. It describes. Similarly, the element in i row and j column of U (t) is described as u _ij (t).

Next, a distance between R−N−P + 2 left singular matrices U (t) and U _AVE having an average value as each element is calculated by Expression (4).

When the distance d (t) to the matrix U _AVE is calculated for the R−N−P + 2 left singular matrices U (t), in order to extract the optimal left singular matrix, R−N−P + 2 Among d (t), the time t at which the value of d (t) becomes the minimum is calculated, and the left singular matrix U (t) corresponding to the calculated time t is extracted for use in abnormality diagnosis. The extracted U (t) is denoted as U.

  In Equation (4), a simple distance between the matrices is calculated, but other distance measurement methods such as using a Mahalanobis distance considering data variation may be used.

  The singular matrix extraction means 13 stores the extracted information of the left singular matrix U (t) in the feature quantity database 15 as the matrix U.

Next, the feature quantity extraction unit 14 extracts a feature quantity in a normal state. The feature quantity extraction unit 14 calculates the value of the matrix S (t) V ^T (t) using the matrix U stored in the feature quantity database 15. This can be calculated as follows by multiplying U ^T (t) from the left of equation (2).

  In the following description, for simplification, a matrix W (t) of N rows and P columns is defined by Expression (6).

  Expression (7) is obtained from Expression (5) and Expression (6).

Since the first row of the matrix S (t) V ^T (t) best represents the feature of A (t), in order to monitor the feature quantity of A (t), S (t) V ^T ( What is necessary is just to monitor the 1st line of t). That is, the first row of the matrix W (t) may be extracted and monitored.

  The feature amount extraction unit 14 extracts the normal state feature amount using the process data from time 1 to time R, which is normal state data. This is done by the following procedure.

The above-described matrix A (t) is created for all areas of collected data. In the first embodiment, the matrix A (t) is created by changing from time t = 1 to time t = R−N−P + 2. For the created matrix A (t) at each time, a matrix W (t) indicating a feature amount is calculated by Expression (7). Since R−N−P + 2 matrices W (t) are obtained from time t = 1 to time t = R−N−P + 2, the first of the obtained R−N−P + 2 matrices W (t) is obtained. The maximum value and the minimum value are extracted from all P elements on the line, that is, (R−N−P + 2) × P elements.
The feature quantity extraction unit 14 stores the maximum value and the minimum value in the feature quantity database 15 as the extracted feature quantities in the normal state. Hereinafter, the extracted maximum value is written as Wmax and the minimum value is written as Wmin.

Next, analysis of process data to be diagnosed for abnormality diagnosis will be described.
The feature quantity analysis unit 16 reads process data to be analyzed from the process quantity database 12. Next, in order from the time R + 1, A (t) is calculated by Expression (1), and then a matrix W (t) indicating a feature value is calculated by Expression (7) using the left singular matrix U. .
The feature quantity analyzing means 16 stores each element of the calculated first row of W (t) in the feature quantity database 15 as the diagnostic target feature quantity at time t together with the time.

The structure of the feature quantity data to be diagnosed stored in the feature quantity database 15 is as shown in the structure diagram of the database in FIG. The numerical values in the first row of W (t) are stored in a matrix with the time.
Wxy represents an element of x rows and y columns of the matrix W (t). In FIG. 5, the description of the matrix U stored in the feature amount database 15, the maximum value Wmax that is the feature amount in the normal state, and the minimum value Wmin is omitted.

  The comparison unit 17 reads the maximum values Wmax and Wmin that are normal state feature amounts from the feature amount database 15, and then constantly monitors the feature amount database 15, and when a feature amount to be diagnosed is newly stored, The stored feature quantity of the diagnosis target is read out, compared with the maximum values Wmax and Wmin, whether the feature quantity of the diagnosis target is within the reference value range, and the comparison result is stored in the feature quantity database 15 To do.

The display means 18 reads the maximum values Wmax and Wmin, which are the feature quantities in the normal state, and the feature quantity to be diagnosed from the feature quantity database 15, and the comparison result performed by the comparison means 17, and displays them on the display device 19. .
When a feature quantity to be diagnosed is newly stored in the feature quantity database 15, the display means 18 reads the stored feature quantity to be diagnosed, displays it on the display device 19, and updates the display data.
Further, the display means 18 displays an alarm on the display device 19, generates an alarm sound, or makes a sound notification to notify the operator if there is a possibility of abnormality as a result of the comparison performed by the comparison means 17. be able to.

  The display on the display device 19 displays the value of the feature value of the diagnosis object as it is, or as an alarm when the feature value of the diagnosis object deviates from the reference value, that is, when it is greater than the maximum value Wmax or less than the minimum value Wmin. Can be displayed. Alternatively, when the feature quantity to be diagnosed deviates from the reference value, the graph color can be changed and displayed.

  The function explanatory diagram of FIG. 6 is a graph display when Wmax is near 3.1, Wmin is near 0.9, and W12 at time R + 21 is larger than Wmax, indicating an abnormality sign. It is an example.

As the reference value to be detected as abnormal, the maximum value Wmax and the minimum value Wmin, which are normal state feature amounts, may be used as they are, or the maximum value Wmax and the minimum value Wmin are multiplied by a coefficient. May be.
Also, instead of the maximum and minimum values of the first row elements of the normal state matrix W (t), the statistical standard deviation and average of the first row elements of the normal state matrix W (t) are calculated. It is also possible to calculate and use the standard deviation multiplied by a coefficient as a reference value, and to compare the difference between the feature value of the diagnosis target and the average value with this reference value.

  A series of similar processes described above are performed on other types of process data to be subjected to abnormality diagnosis, and abnormality diagnosis is performed on the process data to be diagnosed.

In the first embodiment, the process amount database 12 and the feature amount database 15 are separately provided as the beta base, but may be combined into one database.
Further, the data collection means 11, the singular matrix extraction means 13, the feature quantity extraction means 14, the feature quantity analysis means 16, the comparison means 17 and the display means 18 are provided as separate means. Alternatively, the whole may be used as, for example, diagnostic means.
Further, the data collection means 11 is provided in the abnormality diagnosis apparatus 1 for plant equipment, but the process monitoring database 2 stores process data necessary for abnormality diagnosis in the process quantity database 12 without providing the data collection means 11. It is good also as a structure.

  As described above, in the plant equipment abnormality diagnosis apparatus 1 according to the first embodiment, time-series process data is expanded into a matrix, singular value decomposition is performed, and an optimal left singular matrix in normal state and normal The feature amount of the state is extracted, the process data of the diagnosis target is analyzed, the feature amount of the diagnosis target is extracted, and deterioration or abnormality is diagnosed. Therefore, in the plant equipment abnormality diagnosis apparatus 1 according to the first embodiment, for example, deterioration that gradually changes with time, temporary fluctuation of process data that cannot be detected by simple upper and lower limit checks of process data. In addition, there is an effect that it is possible to detect temporary noise and the like.

  Further, the diagnosis method of the plant equipment abnormality diagnosis apparatus 1 according to the first embodiment does not depend on the characteristics and control methods of various devices used in the plant, so the characteristics and control methods of the various devices are modeled for diagnosis processing. There is an advantage that the same method can be applied to any device and control method.

Embodiment 2. FIG.
In the second embodiment, when the feature amount extraction unit 14 extracts the maximum value and the minimum value from each element of the matrix W (t) in the normal state with respect to the abnormality diagnosis apparatus 1 for the plant equipment of the first embodiment, When the feature quantity analysis means 16 extracts a feature quantity from the diagnosis target matrix W (t), a plurality of numbers from the first row of the matrix W (t) to the designated n (n is an integer of 2 or more) rows. The present invention relates to an abnormality diagnosis apparatus for plant equipment that performs abnormality diagnosis for a row.

The abnormality diagnosis apparatus for plant equipment according to the second embodiment is basically the same as the abnormality diagnosis apparatus 1 for plant equipment according to the first embodiment, the system configuration, each component, and the function and operation of the apparatus.
Accordingly, the second embodiment of the present invention will be described with reference to the structure diagram of the database in FIG. 7 and the function explanatory diagram in FIG. 8, focusing on the differences from the plant equipment abnormality diagnosis apparatus 1 according to the first embodiment. .
In the following description, an example in which up to the second row of the matrix W (t) is the monitoring target will be described.

Since the function and operation of the singular matrix extraction unit 13 are the same as those in the first embodiment, the function and operation of the feature amount extraction unit 14 will be described.
The feature amount extraction unit 14 normalizes the maximum value and the minimum value of all the P elements in the first row and the second row of the matrix W (t) in the normal state for each row of the matrix W (t). Extracted as a feature amount of the state. That is, when the target is up to the second line, the result extracted for the elements in the first line is W1max as the maximum value and W1min as the minimum value, and these are stored in the feature value database 15 as the feature values in the normal state. Store. Similarly, the results extracted for the elements in the second row are W2max as the maximum value and W2min as the minimum value, and these are stored in the feature amount database 15 as feature amounts in the normal state.

Next, the function and operation of the feature quantity analysis unit 16 will be described.
The feature quantity analysis unit 16 reads out the process data to be diagnosed from the process quantity database 12, and calculates A (t) for the process data to be diagnosed by the equation (1) in order from time R + 1. Then, using the left singular matrix U, W (t) is calculated by Equation (7). Each element of the calculated first row and second row of W (t) is stored in the feature amount database 15 together with the time as a feature amount to be diagnosed at time t.
The structure of the feature quantity of the diagnosis target stored in the feature quantity database 15 is as shown in FIG. The numerical values of the first and second rows of W (t) are stored in a matrix with the time.
Wxy represents an element of x rows and y columns of the matrix W (t). In FIG. 7, the description of the matrix U stored in the feature amount database 15 and the maximum values W1max and W2max and the minimum values W1min and W2min which are the feature amounts in the normal state is omitted.

  The comparison unit 17 reads the maximum values W1max, W2max, the minimum values W1min, W2min, which are normal state feature amounts, from the feature amount database 15, and then constantly monitors the feature amount database 15 to newly detect the feature amount to be diagnosed. Is stored, the stored feature quantity of the diagnosis target is read out, compared with the maximum values W1max, W2max, the minimum values W1min, W2min, and whether the diagnosis target feature quantity is within the range of the reference value is monitored. The comparison result is stored in the feature amount database 15.

The display means 18 reads the maximum values W1max, W2max, the minimum values W1min, W2min, which are the feature quantities in the normal state, and the feature quantities of the diagnosis target from the feature quantity database 15, and further the comparison results performed by the comparison means 17, and these are read out. It is displayed on the display device 19.
When a feature quantity to be diagnosed is newly stored in the feature quantity database 15, the display means 18 reads the stored feature quantity to be diagnosed, displays it on the display device 19, and updates the display data.
Further, the display means 18 displays an alarm on the display device 19, generates an alarm sound, or makes a sound notification to notify the operator if there is a possibility of abnormality as a result of the comparison performed by the comparison means 17. be able to.

  The display on the display device 19 displays the value of the feature value of the diagnosis object as it is, or when the feature value of the diagnosis object deviates from the reference value, that is, when it is larger than the maximum values W1max and W2max or smaller than the minimum values W1min and W2min. Can be displayed as an alarm. Alternatively, when the feature quantity to be diagnosed deviates from the reference value, the graph color can be changed and displayed.

  The function explanatory diagram of FIG. 8 shows that W1max is near 3.1, W1min is near 0.9, W2max is near 2.5, W2min is near 0.3, and W12 at time R + 21 is greater than W1max. Similarly, it is an example of a graph display when W22 at time R + 21 is larger than W2max and indicates an abnormality sign.

As the reference value to be detected as abnormal, the maximum values W1max, W2max, the minimum values W1min, W2min, which are the feature amounts in the normal state, may be used as they are, or the maximum values W1max, W2max, the minimum values W1min, W2min are used. The value may be multiplied by a coefficient.
Also, instead of the maximum and minimum values of the elements in the first and second rows of the normal state matrix W (t), the first and second rows of the normal state matrix W (t) It is also possible to calculate the statistical standard deviation and average of each element, use the standard deviation multiplied by a coefficient as the reference value, and compare the difference between the feature value of the diagnosis target and the average value with this reference value. it can.

  A series of similar processes described above are performed on other types of process data to be subjected to abnormality diagnosis, and abnormality diagnosis is performed on the process data to be diagnosed.

  In the first embodiment, among the elements of the matrix W (t), only the first row that best represents the feature is the monitoring target. However, the abnormality diagnosis device for plant equipment according to the second embodiment uses the feature quantity. The constituent means such as the extracting means 14, the feature amount analyzing means 16 and the comparing means 17 can handle elements from the first row of the matrix W (t) indicating the normal state and the feature quantity to be diagnosed, and the matrix W Deterioration and abnormality can be diagnosed by using the elements from the first line of (t) as a monitoring target.

  As described above, in the second embodiment, as in the first embodiment, the process data cannot be detected by a simple upper / lower limit check, for example, deterioration that changes gradually with time, and temporary In addition to the effect of detecting typical process data fluctuations and temporary noise, there is an effect of making a more detailed diagnosis.

Embodiment 3 FIG.
Although the abnormality diagnosis apparatus 1 for plant equipment according to the first embodiment has one type of process data as a monitoring target, in this third embodiment, a plant equipment that can analyze a plurality of types of process data as monitoring targets collectively. The present invention relates to an abnormality diagnosis apparatus.

The abnormality diagnosis device for plant equipment according to the third embodiment is basically the same as the abnormality diagnosis device 1 for plant equipment according to the first embodiment, the system configuration, each component, and the function and operation of the device.
Therefore, the third embodiment of the present invention will be described with reference to the functional explanatory diagram of FIG. 9, focusing on the differences from the plant equipment abnormality diagnosis device 1 according to the first embodiment.
In the following description, a case where P types of process data are collectively analyzed will be described as an example.

First, P types of process data stored in the process quantity database 12 will be described.
The time series data of the first process data is a1 (1), a1 (2),..., A1 (R), and the time series data of the second process data is a2 (1), a2 ( 2),..., A2 (R),..., Time-series data of the Pth process data is time-series data as ap (1), ap (2),. Is expressed.
Process data up to time R is stored in advance in the process quantity database 12 by the data collection means 11.

The function and operation of the singular matrix extraction means 13 will be described.
The singular matrix extraction means 13 reads data of P types of process data to be diagnosed from the process quantity database 12 and creates the following matrix.

  For example, when t = 1, the singular matrix extraction means 13 creates a matrix using P types of process data shown in the function explanatory diagram of FIG. In FIG. 9, only the first, second, and Pth process data among the P types of process data are shown, and the other process data are not shown.

  The singular matrix extraction means 13 performs singular value decomposition on the created matrix A (t), performs the same processing as in Embodiment 1, and extracts the left singular matrix U (t).

Thereafter, as in the first embodiment, U (t) is calculated for all areas of the collected data, U _AVE , d (t) is calculated, and finally, a matrix U is calculated, and the feature quantity database 15 To store.

  The feature quantity extraction unit 14 uses the matrix U stored in the feature quantity database 15 and uses a plurality of types of process data from time 1 to time R, which is normal state data, to obtain normal state feature quantities. To extract. That is, the feature amount extraction unit 14 extracts the maximum value Wmax and the minimum value Wmin from all the elements in the first row of the matrix W (t), and stores them in the feature amount database 15 as feature amounts in the normal state. .

The feature quantity analysis unit 16 reads a plurality of types of process data to be diagnosed from the process quantity database 12, calculates A (t), and then uses the left singular matrix U to calculate the feature quantity according to equation (7). The matrix W (t) shown is calculated.
The feature quantity analyzing means 16 stores each element of the calculated first row of W (t) in the feature quantity database 15 as the diagnostic target feature quantity at time t together with the time.

  The comparison unit 17 reads the maximum values Wmax and Wmin that are normal state feature amounts from the feature amount database 15, and then constantly monitors the feature amount database 15, and when a feature amount to be diagnosed is newly stored, The stored feature quantity of the diagnosis target is read out, compared with the maximum values Wmax and Wmin, whether the feature quantity of the diagnosis target is within the reference value range, and the comparison result is stored in the feature quantity database 15 To do.

The display means 18 reads the maximum values Wmax and Wmin, which are the feature quantities in the normal state, and the feature quantity to be diagnosed from the feature quantity database 15, and the comparison result performed by the comparison means 17, and displays them on the display device 19. .
When a feature quantity to be diagnosed is newly stored in the feature quantity database 15, the display means 18 reads the stored feature quantity to be diagnosed, displays it on the display device 19, and updates the display data.
Further, the display means 18 displays an alarm on the display device 19, generates an alarm sound, or makes a sound notification to notify the operator if there is a possibility of abnormality as a result of the comparison performed by the comparison means 17. be able to.

  The display on the display device 19 displays the value of the feature value of the diagnosis object as it is, or as an alarm when the feature value of the diagnosis object deviates from the reference value, that is, when it is greater than the maximum value Wmax or less than the minimum value Wmin. Can be displayed. Alternatively, when the feature quantity to be diagnosed deviates from the reference value, the graph color can be changed and displayed.

As the reference value to be detected as abnormal, the maximum value Wmax and the minimum value Wmin, which are normal state feature amounts, may be used as they are, or the maximum value Wmax and the minimum value Wmin are multiplied by a coefficient. May be.
Also, instead of the maximum and minimum values of the first row elements of the normal state matrix W (t), the statistical standard deviation and average of the first row elements of the normal state matrix W (t) are calculated. It is also possible to calculate and use the standard deviation multiplied by a coefficient as a reference value, and to compare the difference between the feature value of the diagnosis target and the average value with this reference value.

  In the first embodiment, the time series data of one type of process data is monitored and analyzed. In the third embodiment, the singular matrix extraction means 13, the feature quantity extraction means 14, and the feature quantity analysis means 16 When creating the matrix A (t), time series data of a plurality of types of process data is used.

  As described above, in the plant equipment abnormality diagnosis apparatus according to the third embodiment, the operator only needs to monitor the feature amount instead of monitoring a plurality of types of process data. There is an effect that can reduce the load. In addition, because the relationship between time series data of multiple types of process data can be analyzed, it is possible to identify characteristic variations that are difficult to detect by monitoring individual process data, such as when the relationship between multiple types of process data changes. There is an effect.

Embodiment 4 FIG.
In the fourth embodiment, when the feature quantity extraction unit 14 extracts the maximum value and the minimum value from each element of the matrix W (t), the matrix W (t) is compared with the abnormality diagnosis device for plant equipment of the third embodiment. The present invention relates to an abnormality diagnosis apparatus for plant equipment that performs a plurality of rows from the first row of t) to the designated n (n is an integer of 2 or more) rows.

The abnormality diagnosis device for plant equipment according to the fourth embodiment is basically the same as the abnormality diagnosis device for plant equipment according to the third embodiment, the system configuration, each component, and the function and operation of the device.
Therefore, with respect to the fourth embodiment of the present invention, the functional explanatory diagrams of FIGS. 8 and 10 referred to in the description of the second embodiment, with a focus on differences from the plant equipment abnormality diagnosis device 1 according to the third embodiment. It explains using.
In the following description, a case where the target is up to the second row of the matrix W (t) will be described as an example.

Since the function and operation of the singular matrix extraction unit 13 are the same as those in the third embodiment, the function and operation of the feature amount extraction unit 14 will be described.
The feature amount extraction unit 14 extracts the maximum value and the minimum value of all the P elements in the first row and the second row of the matrix W (t) in the normal state for each row of the matrix W (t). To do. That is, when the target is up to the second line, the result extracted for the elements in the first line is W1max as the maximum value and W1min as the minimum value, and these are stored in the feature value database 15 as the feature values in the normal state. Store. Similarly, the results extracted for the elements in the second row are W2max as the maximum value and W2min as the minimum value, and these are stored in the feature amount database 15 as feature amounts in the normal state.

Next, the function and operation of the feature quantity analysis unit 16 will be described.
The feature quantity analysis means 16 reads a plurality of types of process data to be diagnosed from the process quantity database 12 and calculates A (t). Then, using the left singular matrix U, W (t) is calculated, and each element of the first row and the second row of W (t) is used as the feature quantity of the diagnosis target at time t. It is stored in the quantity database 15 together with the time.

  The comparison unit 17 reads the maximum values W1max, W2max, the minimum values W1min, W2min, which are the feature amounts in the normal state, from the feature amount database 15, and then constantly monitors the feature amount database 15 and stores the stored features of the diagnosis target. The amount is read out, compared with the maximum values W1max, W2max, minimum values W1min, W2min, whether or not the feature amount to be diagnosed is within the reference value range, and the comparison result is stored in the feature amount database 15. .

The display means 18 reads the maximum values W1max, W2max, the minimum values W1min, W2min, which are the feature quantities in the normal state, and the feature quantities of the diagnosis target from the feature quantity database 15, and further the comparison results performed by the comparison means 17, and these are read out. It is displayed on the display device 19.
Further, the display means 18 displays an alarm on the display device 19, generates an alarm sound, or makes a sound notification to notify the operator if there is a possibility of abnormality as a result of the comparison performed by the comparison means 17. be able to.

  The display on the display device 19 displays the value of the feature value of the diagnosis object as it is, or when the feature value of the diagnosis object deviates from the reference value, that is, when it is larger than the maximum values W1max and W2max or smaller than the minimum values W1min and W2min. Can be displayed as an alarm. Alternatively, when the feature quantity to be diagnosed deviates from the reference value, the graph color can be changed and displayed.

  The display on the display device 19 may be displayed as a graph as shown in FIG. 8 described in the second embodiment, or as shown in the function explanatory diagram of FIG. The temporal change for each element of (t) may be displayed in a graph.

As the reference value to be detected as abnormal, the maximum values W1max, W2max, the minimum values W1min, W2min, which are the feature amounts in the normal state, may be used as they are, or the maximum values W1max, W2max, the minimum values W1min, W2min are used. The value may be multiplied by a coefficient.
Also, instead of the maximum and minimum values of the elements in the first and second rows of the normal state matrix W (t), the first and second rows of the normal state matrix W (t) It is also possible to calculate the statistical standard deviation and average of each element, use the standard deviation multiplied by a coefficient as the reference value, and compare the difference between the feature value of the diagnosis target and the average value with this reference value. it can.

  In the third embodiment, among the elements of the matrix W (t), only the first row that best represents the feature is the monitoring target. However, the abnormality diagnosis device for plant equipment according to the fourth embodiment uses the feature quantity. From the first row to n (n is an integer equal to or greater than 2) of the matrix W (t) in which the constituent means such as the extraction means 14, the feature quantity analysis means 16 and the comparison means 17 indicate the normal state and the feature quantity to be diagnosed. It is possible to perform the diagnosis of deterioration and abnormality for the elements of the plurality of rows from the first row of the matrix W (t).

  As described above, in the fourth embodiment, in addition to the effects in the third embodiment, there is an effect that a finer diagnosis can be performed.

Embodiment 5 FIG.
In the fifth embodiment, the plant apparatus abnormality diagnosis apparatus 1 according to the first embodiment can automatically determine the optimum size of the matrix created by the equation (1).
The fifth embodiment of the present invention will be described with reference to the system configuration diagram of FIG. 11 and the flowchart of FIG.
In FIG. 11, the same or corresponding parts as those in FIG.

  The system configuration of the plant equipment abnormality diagnosis device 21 is a configuration in which an optimum matrix selection means 22 is further added to the plant equipment abnormality diagnosis device 1 of the first embodiment.

  The function and operation of the optimum matrix selection means 22 will be described based on the flowchart of FIG.

  In order to automatically determine the optimum size of the matrix, the optimum matrix selection means 22 starts processing from step S1, and sets the initial value 2 to the variable p indicating the number of columns of the matrix in step S2. To do. This corresponds to the lower limit value of the number of columns of the optimum matrix size. Similarly, in step S3, an initial value 2 is set to a variable n indicating the number of rows in the matrix. This corresponds to the lower limit of the number of rows of the optimum matrix size.

  Next, a matrix corresponding to the matrix represented by Expression (1) is created. In Embodiments 1 and 2, the size of the matrix is fixed to the number of rows N and the number of columns P, so the matrix to be created is expressed as A (t). In order to make a decision while changing the size of the matrix, n and p indicating the size of the matrix are also used at the same time and expressed as A (n, p, t).

In step S4, it is determined whether or not the matrix of Expression (9) can be created for the set number of rows n and number of columns p. As in the first and second embodiments, when the process data is stored until time R, if the value of n + p−1 is larger than R, it is determined that the expression (9) cannot be created. Then, the process proceeds to step S20. However, R is 5 or more.
In step S20, the column number p is incremented by 1, and the process returns to step S3. If it is determined in step S4 that the expression (9) can be created, the process proceeds to step S5, and the initial value 1 is set to the time t.

  In step S6, a matrix A (n, p, t) is created according to equation (9). In step S7, the created matrix A (n, p, t) is subjected to singular value decomposition.

In addition, what is described as A (t) in the first and second embodiments is similar to that described as A (n, p, t) in the fifth embodiment. What is described as U (t), S (t), and V ^T (t) in the first and second embodiments is referred to as U (n, p, t), S in this embodiment. (N, p, t) and V ^T (n, p, t).

  In step S8, it is checked whether t has reached the upper limit. Since the process data is stored until time R, the upper limit of t is R−n−p + 2. If the upper limit has not been reached, the process proceeds to step S21. In step S21, t is incremented by 1, and the process returns to step S6. A (n, p, t) is created and U (n, p, t) is extracted for the counted time t. . If the upper limit has been reached, the process proceeds to step S9.

  In step S9, the variation of R−n−p + 2 U (n, p, t) calculated by changing t with respect to the set n and p is calculated. Specifically, the following procedure is used.

First, an average value of each element of U (n, p, t) is calculated. A matrix in which the average values of the elements are arranged is described as U _AVE (n, p).

The i row and j column of U (n, p, t) are described as u (n, p, t) _ij , the i row and j column of U _AVE (n, p) are described as uave (n, p) _ij, and each element The variance indicating the variation of is calculated by the equation (12).

The average of the calculated variance uvar (n, p) _ij of each element is calculated, and this is defined as the variance of the matrix U (n, p, t) with respect to the set n, p, and var (n, p) and Describe.

  Next, proceeding to step S10, it is checked whether n has reached the upper limit. Since process data is stored until time R, the upper limit of n is R / 2. If the upper limit has not been reached, the process proceeds to step S22. In step S23, n is incremented by 1, and the process returns to step S4 to repeat the process for the counted number n of rows. If the upper limit has been reached, the process proceeds to step S11.

  In step S11, it is checked whether p has reached the upper limit. Since the process data is stored until time R, the upper limit of p is R / 2. If the upper limit has not been reached, the process proceeds to step S23. In step S23, p is incremented by 1, and the process returns to step S3, and the process is repeated for the counted column number p. If the upper limit has been reached, the process proceeds to step S12.

  The processing up to step S11 is repeated while changing the number of rows n and the number of columns p, and the variance var (n, p) is obtained for each value of n and p.

In step S12, a combination of n and p that minimizes the value is selected from the plurality of var (n, p) calculated up to step S11. For the selected number of rows n and number of columns p, the matrix A (t) shown in Expression (1) is created for the stored R process data, and the singular value is changed while changing the time t. When the matrix U (t) is calculated by performing decomposition, the number of rows and the number of columns with the smallest variation is the number of rows that can be most accurately learned in the normal state learning for the analysis of the process data. The number of columns.
In step S13, the selected number of rows n and number of columns p are stored in the feature quantity database 15, and the process ends in step S14.

  Since the optimum number of rows and columns of the matrix A (t) represented by the equation (1) is selected and stored in the feature quantity database 15, the abnormality diagnosis device for plant equipment according to the first and second embodiments Uses the optimum number of rows and columns stored in the feature amount database 15 and performs an abnormality diagnosis of process data to be diagnosed according to the procedure described in the first and second embodiments.

  As described above, the abnormality diagnosis apparatus for plant equipment according to the fifth embodiment performs singular value decomposition on a matrix while changing the number of rows and columns of the matrix, and the variation in the result of singular value decomposition is minimized. In addition to the effects of the first and second embodiments, the optimum number of rows and columns of the matrix most suitable for singular value decomposition is added. It can be automatically selected, and there is an effect that it is not necessary to manually determine the number of rows and columns of a matrix in consideration of characteristics of plant equipment and process data.

Embodiment 6 FIG.
In the sixth embodiment, it is possible to automatically determine the optimum size of the matrix created by the equation (1) for the abnormality diagnosis apparatus for plant equipment of the third and fourth embodiments. .
The system configuration of the plant equipment abnormality diagnosis device according to the sixth embodiment of the present invention is the same as the system configuration diagram (FIG. 11) according to the fifth embodiment.
With respect to the sixth embodiment of the present invention, the function and operation of the optimum matrix selection means 22 will be described based on the flowchart of FIG.

  In order to automatically determine the optimum size of the matrix, the optimum matrix selection means 22 starts processing from step S31, and sets an initial value 2 to a variable n indicating the number of rows of the matrix in step S32. To do. This corresponds to the lower limit of the number of rows of the optimum matrix size. In step S33, an initial value 1 is set to time t.

  Next, it progresses to step S34 and the matrix equivalent to the matrix shown by Formula (8) and Formula (9) is produced. In the third and fourth embodiments, since the matrix size is fixed to the number of rows N and the number of columns P, the matrix to be created is expressed as A (t). In the present embodiment, In order to make a decision while changing the number of rows of the matrix, n indicating the size of the matrix is also used at the same time, expressed as A (n, t), and the matrix is created by the equation (14).

  Next, in step S35, the created matrix A (t, n) is subjected to singular value decomposition.

In addition, what is described as A (t) in the third and fourth embodiments is similar to that described in the present embodiment as A (n, t). 3. What has been described as U (t), S (t), V ^T (t) in the fourth embodiment, U (n, t), S (n, t) in this embodiment , V ^T (n, t).

  In step S36, it is checked whether t has reached the upper limit. Since the process data is stored until time R, the upper limit of t is R−n + 1. If the upper limit has not been reached, the process moves to step S51, t is incremented by 1, and the process returns to step S34. A (n, t) is created for time t counted up in step S34, and U (n, t) is calculated. If the upper limit has been reached, the process proceeds to step S37.

  In step S37, the variation of R−n + 1 U (n, t) calculated by changing t with respect to the set n is calculated. Specifically, the calculation is performed according to the following procedure.

First, an average value of each element of U (n, t) is calculated. A matrix in which the average values of the elements are arranged is described as U _AVE (n).

The i row and j column of U (n, t) are described as u (n, t) _ij , the i row and j column of U _AVE (n) are described as uave (n) ij, and the variance indicating the variation of each element is expressed by the formula ( 17).

The average of the calculated variance uvar (n) _ij of each element is calculated, this is defined as the variance of the matrix U (n, t) for the set n, and is described as var (n).

  Next, the process proceeds to step S38, where it is checked whether n has reached the upper limit. Since the process data is stored until time R, the upper limit of n is R / P. If the upper limit has not been reached, the process proceeds to step S52. In step S52, n is incremented by 1, and the process returns from step S33 to step S34 to repeat the process for the counted number n of rows. If the upper limit has been reached, the process proceeds to step S39.

  The processing up to step S38 is repeated while changing the number of rows n, and a variance var (n) is obtained for each value of n.

In step S39, n having the minimum value is selected from the plurality of var (n) calculated up to step S38. The selected number n of rows is generated by creating a matrix A (t) represented by the equations (8) and (9) for the R pieces of stored process data and changing the time t. When the matrix U (t) is calculated by performing value decomposition, the number of rows has the smallest variation, and the number of rows that can be most accurately learned in learning of a normal state for process data analysis.
In step S40, the selected number of lines n is stored in the feature quantity database 15, and the process ends in step S41.

  Since the optimum number of rows of the matrix A (t) represented by the equations (8) and (9) is selected and stored in the feature amount database 15, the abnormality of the plant equipment according to the third and fourth embodiments. The diagnosis device performs an abnormality diagnosis of a plurality of types of process data to be diagnosed using the optimum number of rows stored in the feature amount database 15 and according to the procedure described in the third and fourth embodiments. .

  As described above, the abnormality diagnosis apparatus for plant equipment according to the sixth embodiment performs singular value decomposition of a matrix while changing the number of rows of the matrix, and the matrix having the smallest variation in the result of singular value decomposition. Since the optimum matrix selecting means 22 for automatically selecting the number of rows is added, in addition to the effects of the third and fourth embodiments, the number of rows of the most appropriate matrix for the singular value decomposition can be automatically selected. There is an effect that it is not necessary to manually determine the number of rows of the matrix in consideration of the data characteristics.

1,21 Plant equipment abnormality diagnosis device, 2 Plant monitoring device, 3 Controller,
4 control network, 5 plant, 11 data collection means,
12 process quantity database, 13 singular matrix extraction means, 14 feature quantity extraction means,
15 feature quantity database, 16 feature quantity analysis means, 17 comparison means, 18 display means,
19 Display device, 22 Optimal matrix selection means.

Claims (5)

A database for storing plant process data;
A matrix is created from one kind of normal process data stored in the database, singular value decomposition of this matrix is performed, and the left singular matrix closest to the average is extracted from the obtained left singular matrices Singular matrix extraction means for storing the extracted left singular matrix in the database;
The extracted left singular matrix and the normal state process data are used to calculate a matrix representing the feature quantity of the process amount, and the maximum value and the minimum value of the first row of the calculated matrix are used as the normal state feature quantity. Feature quantity extraction means for extracting and storing in the database;
The process data to be diagnosed is read from the database, the matrix representing the feature quantity of the process data to be diagnosed is calculated using the extracted left singular matrix, and each element of the first row of the calculated matrix is diagnosed A feature quantity analyzing means for storing in the database as a target feature quantity, a comparison means for reading out and comparing the feature quantity in the normal state and the feature quantity for the diagnosis from the database, and storing a comparison result in the database;
An abnormality diagnosis device for plant equipment comprising: A database for storing plant process data;
Create a matrix from multiple types of normal process data stored in the database, perform singular value decomposition of this matrix, and extract the left singular matrix closest to the average from the obtained left singular matrices Singular matrix extraction means for storing the extracted left singular matrix in the database;
The extracted left singular matrix and a plurality of types of process data in the normal state are used to calculate a matrix representing the feature quantity of the process amount, and the maximum value and the minimum value of the first row of the calculated matrix are set to the normal state. Feature quantity extraction means for extracting as feature quantity and storing it in the database;
A plurality of types of process data to be diagnosed is read from the database, a matrix representing the feature amount of the plurality of types of process data to be diagnosed is calculated using the extracted left singular matrix, and a first of the calculated matrices is calculated. A feature amount analyzing means for storing each element of the row in the database as a feature amount to be diagnosed;
Comparison means for reading out and comparing the feature quantity in the normal state and the feature quantity of the diagnosis target from the database, and storing a comparison result in the database;
An abnormality diagnosis device for plant equipment equipped with Data collection means for collecting process data from the plant and storing it in the database;
The abnormality of the plant equipment according to claim 1, further comprising: a display unit that reads out the feature amount in the normal state, the feature amount of the diagnosis target stored in the database, and the comparison result and displays the comparison result on a display device. Diagnostic device. The feature amount extraction means extracts the maximum value and the minimum value of each of a plurality of rows from the first row to the n-th row (n is an integer of 2 or more) of the calculated matrix as a feature amount in a normal state, and Stored in
The feature amount analyzing means stores each element of each of a plurality of rows from the first row to the n-th (n is an integer of 2 or more) rows of the calculated matrix in the database as a feature amount to be diagnosed.
The abnormality diagnosis device for plant equipment according to claim 1 or 2. An optimal matrix that creates a matrix from normal state process data stored in the database, performs singular value decomposition of the matrix, and calculates the number of rows and columns of the matrix that minimizes the variation of the obtained left singular matrix The abnormality diagnosis device for plant equipment according to claim 1 or 2, further comprising a selection means.

Images