JP2011090382A - Monitoring system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a monitoring system for automating a series of processing from the monitoring of the omen of failure to failure diagnosis in the object of monitoring. <P>SOLUTION: This monitoring system 1 is provided with: a monitoring device 2 for calculating the Mahalanobis distance of monitoring object data, and for detecting the failrue of a monitoring object 10 based on the Mahalanobis distance; a data processor 3 for extracting monitoring object data (abnormal signal) in which the omen of failure appears and monitoring object data (related signal) related with the abnormal signal from a plurality of monitoring object data, and for generating a prescribed input signal based on those abnormal signal and the related signal; and a failure diagnosis device 4 for performing the failure diagnosis of the monitoring object 10 based on the input signal. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a monitoring system, and more particularly to a monitoring system capable of automating a series of processes from monitoring of a sign of abnormality in a monitoring target to failure diagnosis.

  In various plants such as a gas turbine power plant, a nuclear power plant, and a chemical plant in recent years, a system for monitoring the operation state of the plant using the Mahalanobis distance is adopted. In such a monitoring system, a plurality of monitoring target data is acquired from a plant to be monitored at a predetermined time interval, a Mahalanobis distance is calculated for each of these monitoring target data, and when this Mahalanobis distance exceeds a predetermined threshold value It is determined that there is an abnormality in the operation state of the plant. As such a monitoring system, a technique described in Patent Document 1 is known.

  On the other hand, there is known a device (failure diagnosis device) that performs failure diagnosis of a monitoring target when there is an abnormality in the operation state of the plant. This failure diagnosis apparatus can probabilistically estimate a failure component to be monitored using predetermined input data relating to the monitoring target and a Bayesian network.

  In the conventional technology, the monitoring system and the failure diagnosis apparatus are independent from each other, and a specialized engineer uses them over time to perform failure diagnosis of the monitoring target.

JP 59-68644 A

  An object of the present invention is to provide a monitoring system capable of automating a series of processes from monitoring of a sign of abnormality in a monitoring target to failure diagnosis.

  In order to achieve the above object, a monitoring system according to the present invention is a monitoring system capable of automating a series of processes from monitoring of a sign of abnormality in a monitoring target to failure diagnosis, and acquiring predetermined monitoring target data from the monitoring target. Monitoring means for calculating a Mahalanobis distance of the monitoring target data and detecting an abnormality of the monitoring target based on the Mahalanobis distance, and monitoring target data in which a sign of abnormality appears from the plurality of monitoring target data ( And a data processing means for extracting monitoring target data related to the abnormal signal (hereinafter referred to as a related signal) and generating a predetermined input signal based on the abnormal signal and the related signal. And failure diagnosis means for performing failure diagnosis of the monitoring target based on the input signal.

  When it is determined that there is an abnormality in the monitoring target, this monitoring system extracts an abnormal signal and a related signal from the monitoring target data. Further, an input signal to the failure diagnosis means is generated based on these abnormal signals and related signals. Based on this input signal, the failure diagnosis of the monitoring target is performed. Thereby, there is an advantage that a series of processes from monitoring of a sign of abnormality in the monitoring target to failure diagnosis can be automated.

  The monitoring system according to the present invention further includes means for selecting a period in which the time differential value of the Mahalanobis distance continuously exceeds a predetermined threshold as the target period, and from the monitoring target data in the target period The abnormal signal and the related signal are extracted.

  In this monitoring system, a period in which the time differential value of the Mahalanobis distance continuously exceeds a predetermined threshold is set as a target period, and an abnormal signal and a related signal are extracted from the monitoring target data in this target period. In such a configuration, a period in which a sign of abnormality appears prominently can be selected as the target period, so that useful abnormality signals and related signals can be extracted. This has the advantage of improving the accuracy of fault diagnosis. Also, in such a configuration, when the Mahalanobis distance changes slowly due to aging degradation or environmental changes (for example, noise such as changes in the atmospheric temperature), an erroneous failure that determines that the environmental change is an abnormality of the monitored object There is an advantage that diagnosis is prevented.

  The monitoring system according to the present invention further includes means for selecting a period in which the time differential value of the abnormal signal continuously exceeds a predetermined threshold as the target period, and from the monitoring target data in the target period The abnormal signal and the related signal are extracted.

  In this monitoring system, the abnormal signal and the related signal are extracted based on the time differential value of the abnormal signal. In such a configuration, a period in which a sign of abnormality appears prominently can be selected as the target period, so that useful abnormality signals and related signals can be extracted. This has the advantage of improving the accuracy of fault diagnosis. In addition, such a configuration has an advantage that when the Mahalanobis distance is gradually changed due to aging degradation or environmental change of the monitoring target, erroneous fault diagnosis that determines such environmental change as an abnormality of the monitoring target is prevented.

  The monitoring system according to the present invention further includes means for selecting, as the target period, a period in which the Mahalanobis distance continuously exceeds a predetermined threshold within the target period, and the monitoring in the target period The abnormal signal and the related signal are extracted from the target data.

  In this monitoring system, the target period is selected by (1) a combination of the time differential value D ′ of the Mahalanobis distance D and the Mahalanobis distance D, or (2) a combination of the time differential value of the abnormal signal and the Mahalanobis distance D. The In such a configuration, a more suitable target period is selected, so that there is an advantage that the accuracy of failure diagnosis is further improved.

  In the monitoring system according to the present invention, when it is determined that the monitoring target is abnormal, the abnormal signal and the related signal are extracted from the monitoring target data. Further, an input signal to the failure diagnosis means is generated based on these abnormal signals and related signals. Based on this input signal, the failure diagnosis of the monitoring target is performed. Thereby, there is an advantage that a series of processes from monitoring of a sign of abnormality in the monitoring target to failure diagnosis can be automated.

FIG. 1 is a configuration diagram showing a monitoring system according to an embodiment of the present invention. FIG. 2 is a functional block diagram showing a monitoring device of the monitoring system described in FIG. FIG. 3 is a functional block diagram showing the data processing device of the monitoring system shown in FIG. FIG. 4 is a functional block diagram showing the failure diagnosis apparatus for the monitoring system shown in FIG. FIG. 5 is a flowchart showing the operation of the monitoring system shown in FIG. FIG. 6 is an explanatory diagram showing the operation of the monitoring system shown in FIG. FIG. 7 is an explanatory diagram showing a method for calculating the Mahalanobis distance. FIG. 8 is an explanatory view showing a modification of the monitoring system shown in FIG.

  Hereinafter, the present invention will be described in detail with reference to the drawings. Note that the present invention is not limited to the embodiments. Further, the constituent elements of this embodiment include those that can be replaced while maintaining the identity of the invention and that are obvious for replacement. In addition, a plurality of modifications described in this embodiment can be arbitrarily combined within a range obvious to those skilled in the art.

[Monitoring system]
The monitoring system 1 is a system that monitors the operating state of the monitoring target. More specifically, the monitoring system 1 acquires predetermined monitoring target data from the monitoring target, and uses the Mahalanobis distance of the monitoring target data to determine the monitoring target operating state. This is a monitoring system. The monitoring system 1 uses, for example, various plants such as a gas turbine power plant, a nuclear power plant, and a chemical plant as a monitoring target 10. In this embodiment, as an example, a case where the monitoring system 1 sets a gas turbine power plant as a monitoring target 10 will be described (see FIG. 1). The monitoring system 1 can monitor not only a single monitoring target 10 but also a plurality of monitoring targets 10 (not shown).

  The monitoring system 1 includes a monitoring device 2, a data processing device 3, and a failure diagnosis device 4 (see FIG. 1). These devices 2 to 4 are configured by, for example, a PC (Personal Computer). For example, in this embodiment, the monitoring device 2, the data processing device 3, and the failure diagnosis device 4 are independent from each other, and these are connected so as to be capable of information communication (see FIG. 1). ). The monitoring device 2, the data processing device 3, and the failure diagnosis device 4 can be installed independently, and are installed in either a gas turbine power plant or a remote monitoring center.

  The monitoring system 1 is not limited to this, and may be configured by a single device having a program for realizing the functions of the devices 2 to 4 (not shown). Further, when there are a plurality of monitoring targets 10, the monitoring devices 2 are respectively installed for these monitoring targets 10, and these monitoring devices 2 are connected to the data processing device 3 and the failure diagnosis device 4, respectively. (Not shown).

  The monitoring device 2 is a device that acquires predetermined monitoring target data from the monitoring target 10, calculates the Mahalanobis distance D, and detects an abnormality in the operating state of the monitoring target 10 based on the Mahalanobis distance D. The calculation of the monitoring target data and the Mahalanobis distance D will be described later. The monitoring device 2 includes a Mahalanobis distance calculation unit 21, an abnormality determination unit 22, an alarm signal generation unit 23, and a storage unit 24. The Mahalanobis distance calculation unit 21 calculates the Mahalanobis distance D of the acquired monitoring target data. The abnormality determination unit 22 determines whether there is an abnormality in the operating state of the monitoring target 10 based on the Mahalanobis distance D. The alarm signal generator 23 generates a predetermined alarm signal when there is an abnormality in the operation state of the monitoring target 10. The storage unit 24 stores a plurality of pieces of monitoring target data acquired from the monitoring target 10, data necessary for calculation processing of the Mahalanobis distance D (for example, a reference data group related to the Mahalanobis unit space described later), and an abnormal operation state of the monitoring target 10. It is a recording medium that stores data necessary for determination (for example, a threshold k of Mahalanobis distance D for determining whether or not there is an abnormality in the driving state).

  When there is an abnormality in the operation state of the monitoring target 10, the data processing device 3 monitors the monitoring target data (abnormal signal) in which a sign of abnormality appears from the plurality of monitoring target data and the monitoring target data related to the abnormal signal (related Signal), and generates an input signal (abnormal signal data and related signal data) to the failure diagnosis apparatus based on these abnormal signals and related signals and the Mahalanobis distance D. The data processing device 3 includes a target signal extraction unit 31, a target period selection unit 32, an abnormal signal data calculation unit 33, and a related signal data calculation unit 34. The target signal extraction unit 31 extracts an abnormal signal and a related signal from the plurality of acquired monitoring target data. The target period selection unit 32 selects a period (target period) used for calculating abnormal signal data and related signal data from all sampling periods of the monitoring target data. The abnormal signal data calculation unit 33 generates an input signal (abnormal signal data) to the failure diagnosis device 4 based on the abnormal signal and the Mahalanobis distance D. The related signal data calculation unit 34 generates an input signal (related signal data) to the failure diagnosis apparatus based on the related signal and the Mahalanobis distance D.

  The failure diagnosis device 4 is a device that performs failure diagnosis of the monitoring target 10 based on abnormal signal data and related signal data. The failure diagnosis apparatus 4 includes a failure diagnosis unit 41, a Bayesian network 42, a database 43, a diagnosis result generation unit 44, and a display unit 45. The failure diagnosis unit 41 performs failure diagnosis of the monitoring target 10 using the Bayesian network 42 based on the target period, the abnormal signal data, and the related signal data. The Bayesian network 42 is a graphical model that stochastically describes the causal relationship between the abnormal signal and the related signal and the failed part of the monitoring target 10. The database 43 is a recording medium that stores data necessary for failure diagnosis. The diagnosis result generation unit 44 generates a result of failure diagnosis performed by the failure diagnosis unit 41. The display unit 45 is, for example, a PC monitor, and displays a failure diagnosis result generated by the diagnosis result generation unit 44.

  The monitoring target 10 is a gas turbine power plant, and includes a gas turbine 11 and a generator 12. The gas turbine 11 includes a compressor C, a combustor B, and a turbine T. In this gas turbine power plant, first, the compressor C compresses the air taken in from the air intake port to generate compressed air. Next, the combustor B injects fuel into the compressed air to generate high-temperature and high-pressure combustion gas. Next, the turbine T converts the thermal energy of the combustion gas into the rotational energy of the rotor to generate a driving force. Then, this driving force is transmitted to the generator 12 connected to the rotor, and the generator 12 operates to generate power.

[Monitored data]
The monitoring target data is measurement data related to the operating state of the monitoring target 10 and is composed of a plurality of state quantities. These monitoring target data include, for example, control signals for controlling plant equipment (for example, the opening degree of the fuel control valve), output values of various sensors (for example, a generator output sensor, a turbine rotational speed sensor, a fuel temperature, etc. Sensor, fuel flow sensor, air compressor inlet temperature sensor and outlet temperature sensor, blade path temperature sensor, exhaust gas temperature sensor, nitrogen oxide concentration sensor, fuel pressure fluctuation sensor, bearing vibration sensor, bearing metal temperature sensor, rotor cooling air Output value of a temperature sensor, a disk cavity temperature sensor, etc.).

[Mahalanobis distance]
The Mahalanobis distance D is a conceptual distance of the monitoring target data with respect to a predetermined reference data group (Mahalanobis unit space) (see FIG. 7). For example, FIG. 7 shows a correlation diagram in which the horizontal axis is the output of the generator 12 and the vertical axis is the blade path temperature of the gas turbine 11. In the figure, a plurality of reference data (measurement data) is acquired in advance during normal operation of the monitoring object 10, and a reference data group in which the Mahalanobis distance D = 1 is configured by these reference data. As shown in the figure, it can be seen that when the output of the generator 12 increases, the blade path temperature increases accordingly. Further, it can be seen that each reference data is within a specific range due to the correlation between the output of the generator 12 and the blade path temperature, although there are variations due to differences in atmospheric conditions and operating conditions.

  Here, when an abnormality occurs in the operating state of the monitoring target 10, the monitoring target data at that time appears at a position outside the reference data group, and the Mahalanobis distance D increases. Therefore, by calculating the Mahalanobis distance D between the monitoring target data and the reference data group and comparing the Mahalanobis distance D with a predetermined threshold k, it can be determined whether or not the operating state of the monitoring target 10 is abnormal. .

  For example, in this embodiment, the monitoring target 10 is a gas turbine power plant, and monitoring target data for a predetermined period during normal operation is acquired in advance as reference data. The Mahalanobis unit space is calculated from this reference data group. Then, the Mahalanobis distance D of the current monitoring target data is calculated by setting the Mahalanobis distance D in the Mahalanobis unit space to D = 1. Then, the Mahalanobis distance D and a predetermined threshold value k are compared to determine whether the monitoring target 10 is operating abnormally.

  The Mahalanobis distance D is generally calculated as follows. First, the total number of a plurality of state quantities (monitoring target data) indicating the state of the monitoring target 10 is u, each state quantity is assigned to a variable X, and this state quantity is defined by variables X1 to Xu (u is An integer of 2 or more). Next, in the operation state of the monitoring target 10 as a reference, a total of v state quantities of the variables X1 to Xu are respectively acquired (v is an integer of 2 or more).

  Next, the average value Mi and the standard deviation σi (the degree of variation of the reference data) of the variables X1 to Xu are calculated by the following formulas (1) and (2), respectively. In Equations (1) and (2), i is the number of items (the number of state quantities; an integer), i = 1 to u, and indicates a value corresponding to variables X1 to Xu. j is any value (integer) of j = 1 to v, and means that the number of each state quantity is v. For example, if 60 state quantities are acquired, v = 60. The standard deviation is the square root of the expected value obtained by squaring the difference between the state quantity and its average value.

  Next, using the calculated average value Mi and standard deviation σi, the original variables X1 to Xu are converted into x1 to xu by the following mathematical formula (3) (state quantity standardization). That is, the state quantity of the monitoring target 10 is converted into a random variable having an average of 0 and a standard deviation of 1. In Equation (3), j is any value (integer) from j = 1 to v, meaning that the number of each state quantity is v.

Next, in order to perform analysis using data in which the variables are standardized to mean 0 and standard deviation 1, the correlation between the variables X1 to Xu, that is, the covariance matrix (correlation matrix) R indicating the correlation between the variables, The inverse matrix R ⁻¹ of the variance matrix R is calculated by the following mathematical formula (4). In Equation (4), k is the number of items (number of state quantities), and here k = u. Further, i and p are values in each state quantity, and here, i = 1 to u and p = 1 to u.

Next, the Mahalanobis distance D is calculated using the following formula (5). In Equation (5), j is a value (integer) of j = 1 to v, and means that the number of each state quantity is v. K is the number of items (the number of state quantities), and here k = u. Further, a11 to akk are coefficients of the inverse matrix R- ¹ of the covariance matrix R in Expression (4).

  In the Mahalanobis unit space, the average value of the Mahalanobis distance D is D≈1. Further, in the operation state (normal operation state) of the monitoring target 10 serving as a reference, the Mahalanobis distance D is generally within the range of D ≦ 3. However, when an abnormality occurs in the monitoring target 10, the Mahalanobis distance D increases. Therefore, by comparing the Mahalanobis distance D and the predetermined threshold value k, it can be determined whether or not the operating state of the monitoring target 10 is abnormal. The threshold value k is, for example, k> 3, and can be set as appropriate according to the type of monitoring target data.

[Fault diagnosis for monitoring target]
In this monitoring system 1, failure diagnosis of the monitoring target 10 is performed as follows (see FIGS. 5 and 6).

  In step ST1, a plurality of pieces of monitoring target data are acquired from the monitoring target 10 when the monitoring target 10 is in operation. For example, in this embodiment, the monitoring device 2 acquires these monitoring target data in real time at a predetermined measurement interval, and stores them in the storage unit 24 together with information on the measurement time. Note that the measurement interval of the monitoring target data can be arbitrarily set. After this step ST1, the process proceeds to step ST2.

  In step ST2, the Mahalanobis distance D is calculated for each of the plurality of monitoring target data acquired in step ST1. For example, in this embodiment, the Mahalanobis distance calculation unit 21 of the monitoring device 2 reads the monitoring target data acquired at each time and information necessary for the calculation process of the Mahalanobis distance D from the storage unit 24, and these The Mahalanobis distance D is calculated for each monitoring target data. After this step ST2, the process proceeds to step ST3.

  In step ST3, based on the Mahalanobis distance D calculated in step ST2, it is determined whether or not there is an abnormality in the operating state of the monitoring target 10. For example, in this embodiment, the abnormality determination unit 22 of the monitoring device 2 compares the Mahalanobis distance D calculated in step ST2 with a predetermined threshold k, and when D> k, The operating state is determined to be abnormal (positive determination). If an affirmative determination is made in step ST3, the process proceeds to step ST4, and if a negative determination is made, the process ends.

  Note that when the operating state of the monitoring target 10 is abnormal (affirmative determination in step ST3), the alarm signal generator 23 of the monitoring device 2 generates a predetermined alarm signal. And the monitoring apparatus 2 transmits this warning signal, the some monitoring object data acquired in step ST1, and the Mahalanobis distance D calculated in step ST2 to the data processing apparatus 3. Then, based on these pieces of information, the data processing device 3 performs steps ST4 to ST7.

  In step ST4, an abnormal signal and a related signal are extracted from the plurality of monitoring target data acquired in step ST1 (target signal extraction step ST4). The abnormal signal refers to monitoring target data in which a sign of abnormality appears among a plurality of monitoring target data. The related signal refers to other monitoring target data related to the abnormal signal. For example, in this embodiment, the target signal extraction unit 31 of the data processing device 3 has a high contribution ratio to the desired characteristics of the Mahalanobis distance D among the plurality of monitoring target data acquired in step ST1. Are extracted as abnormal signals. Further, the target signal extraction unit 31 extracts other monitoring target data related to the abnormal signal as a related signal. In addition, the mutual relationship (relationship between an abnormal signal and a related signal) between certain monitoring target data and other monitoring target data is defined in advance. After this step ST4, the process proceeds to step ST5.

  In step ST5, a target period is selected (target period selection step ST5). The target period is a period used in an abnormal signal data calculation step ST6 and a related signal data calculation step ST7, which will be described later, and is selected from all sampling periods of the monitoring target data. For example, in this embodiment, the target period selection unit 32 of the data processing device 3 calculates the time differential value D ′ of the Mahalanobis distance D for the abnormal signal extracted in step ST4 and the time differential value D ′ is calculated. A period continuously exceeding the predetermined threshold k ′ is selected as the target period (see FIG. 6). The threshold value k ′ can be set as appropriate according to the type of monitoring target data. After this step ST5, the process proceeds to step ST6.

  In step ST6, abnormal signal data is calculated for the abnormal signal extracted in step ST4 (abnormal signal data calculating step ST6). The abnormal signal data is data calculated based on an abnormal signal (monitoring target data in which a sign of abnormality appears), and is used as an input signal to the failure diagnosis device 4. For example, in this embodiment, the abnormal signal data calculation unit 33 of the data processing device 3 performs the absolute value, integral value, differential value, average value, maximum value, minimum value, and standard deviation of the abnormal signal in the target period of step ST5. Is calculated as abnormal signal data. After step ST6, the process proceeds to step ST7.

  In step ST7, related signal data is calculated for the related signal extracted in step ST4 (related signal data calculating step ST7). The related signal data is data calculated based on the related signal (other monitoring target data related to the abnormal signal) and is used as an input signal to the failure diagnosis device 4. For example, in this embodiment, the related signal data calculation unit 34 of the data processing device 3 performs the absolute value, integral value, differential value, average value, maximum value, minimum value, and standard deviation of the related signal in the target period of step ST5. Is calculated as related signal data. After step ST7, the process proceeds to step ST8.

  Of the abnormal signal data and related signal data, the calculation and input to the failure diagnosis device 4 may be omitted for some numerical values with clear data characteristics.

  In step ST8, failure diagnosis of the monitoring target 10 is performed based on the target period selected in step ST5, the abnormal signal data calculated in step ST6, and the related signal data calculated in step ST7. For example, in this embodiment, the failure diagnosis unit 41 performs failure diagnosis of the monitoring target 10 using the Bayesian network 42 based on the target period, the abnormality signal, and the related signal. In this failure diagnosis, information on the Bayesian network 42 and the database 43 is used, and the failure part of the monitoring target 10 is estimated probabilistically. After step ST8, the process proceeds to step ST9.

  In step ST9, the result of the failure diagnosis performed in step ST8 is output. For example, in this embodiment, the diagnosis result generation unit 44 generates the result of the failure diagnosis performed in step ST8, and the display unit 45 displays the result of the failure diagnosis. After this step ST9, the process is terminated.

[Specific example of failure diagnosis for monitoring target]
Next, a specific example of failure diagnosis of the monitoring target 10 will be described. Here, the case where the monitoring object 10 is a gas turbine power plant (refer FIG. 1) is demonstrated as an example.

  In the monitoring system 1, the monitoring device 2 acquires a plurality of pieces of monitoring target data during operation of the gas turbine power plant that is the monitoring target 10 (step ST1) (see FIG. 5). These monitored data are the cavity temperature of the gas turbine power plant, the blade path temperature, the amount of displacement of the rotor shaft, and the opening of various valves. Specifically, the monitoring target data is arranged along a plurality of cavity temperature sensors arranged between the rotor and the stator, a plurality of blade path temperature sensors arranged in the circumferential direction of the combustion gas outlet, and the circumferential direction of the rotor. Further, it is obtained as an output value from a plurality of rotor shaft displacement sensors and opening sensors installed in various valves. Further, the monitoring device 2 calculates the Mahalanobis distance D (step ST2), and determines whether or not there is an abnormality in the operating state of the monitoring target 10 based on the Mahalanobis distance D (step ST3). In a normal operation state, the Mahalanobis distance D of all the monitoring target data and the predetermined threshold k are in a relationship of D ≦ k (negative determination in step ST3).

  Here, it is assumed that an abnormality has occurred in the blade path temperature among these monitoring target data (see FIG. 6). For example, it is assumed that the blade path temperature rapidly rises from the temperature T0 during normal operation to the temperature T1 in a short time. As a result, the Mahalanobis distance D of the blade path temperature increases. When the Mahalanobis distance D and the predetermined threshold k are in a relationship of D> k, the monitoring device 2 determines that the operation state of the monitoring target 10 is abnormal (affirmative determination in step ST3) and generates an alarm signal. To do.

  Then, the data processing device 3 extracts the blade path temperature as an abnormal signal (monitoring target data in which an abnormality sign appears), and relates other cavity temperatures, rotor shaft displacements, and various valve openings. It extracts as a signal (target signal extraction step ST4). Further, the data processing device 3 calculates the time differential value D ′ of the Mahalanobis distance D of the abnormal signal (blade path temperature), and this time differential value D ′ continuously exceeds a predetermined threshold k ′. -Tb is selected as a target period (target period selection step ST5).

  Further, the data processing device 3 calculates abnormal signal data (absolute value, integral value, differential value, average value, maximum value, minimum value, and standard deviation of the blade path temperature) of the abnormal signal in the target period ta to tb ( Abnormal signal data calculation step ST6), and related signal data of related signals in the target period ta to tb (absolute value, integral value, differential value, average value, maximum value, minimum value, and standard deviation of other monitoring target data) Is calculated (related signal data calculation step ST7).

  Next, the failure diagnosis apparatus 4 performs failure diagnosis of the monitoring target 10 based on the target periods ta to tb, the abnormal signal data, and the related signal data (step ST8). In this failure diagnosis, information on the Bayesian network 42 and the database 43 is used, and the failure part of the monitoring target 10 is estimated probabilistically. Then, the failure diagnosis device 4 displays the diagnosis result on the display unit 45 (step ST9). As a result, the supervisor can acquire a sign of occurrence of an abnormality in the gas turbine power plant and an estimation result of the failed part. The failure diagnosis apparatus 4 generally has a learning function and can improve the accuracy of failure diagnosis by accumulating diagnosis results.

[effect]
As described above, the monitoring system 1 calculates the Mahalanobis distance D of the monitoring target data and detects the abnormality of the monitoring target 10 based on the Mahalanobis distance D, and the monitoring target data. A data processing means (data processing device 3) for extracting an abnormal signal and a related signal from the signal and generating a predetermined input signal (abnormal signal data and related signal data) based on the abnormal signal and the related signal, and the input signal And a failure diagnosis means (failure diagnosis device 4) for performing a failure diagnosis of the monitoring target 10 based on (see FIG. 1). In such a configuration, when it is determined that the monitoring target 10 is abnormal (positive determination in step ST3), an abnormal signal and a related signal are extracted from the monitoring target data (target signal extraction step ST4). Further, input signals (abnormal signal data and related signal data) to the failure diagnosis means are generated based on these abnormal signals and related signals (abnormal signal data calculation step ST6 and related signal data calculation step ST7). Then, based on this input signal, failure diagnosis of the monitored object 10 is performed (step ST8) (see FIG. 5). Thereby, there is an advantage that a series of processes from monitoring of a sign of abnormality in the monitoring target 10 to failure diagnosis can be automated.

  Further, in this configuration, the process of extracting the abnormal signal and the related signal from the monitoring target data (target signal extraction step ST4) is automated, so that the accuracy of extracting the abnormal signal and the related signal is improved. This has the advantage of improving the accuracy of fault diagnosis. For example, in a configuration in which the monitor extracts the abnormal signal and the related signal while calculating the input signal to the failure diagnosis device while viewing the trend chart of a large number of monitoring target data, the monitor needs a high level of skill. It is difficult to improve the accuracy of fault diagnosis.

  The monitoring system 1 further includes means (data processing device 3) for selecting a period (target period) in which the time differential value D ′ of the Mahalanobis distance D continuously exceeds a predetermined threshold k ′, and An abnormal signal and a related signal in the target period are extracted (target signal extraction step ST4 and target period selection step ST5) (see FIGS. 5 and 6). That is, a period in which the time differential value D ′ of the Mahalanobis distance D continuously exceeds the predetermined threshold k ′ is set as a target period, and an abnormal signal and a related signal are extracted from the monitoring target data in this target period. In such a configuration, a period in which a sign of abnormality appears prominently can be selected as the target period, so that useful abnormality signals and related signals can be extracted. This has the advantage of improving the accuracy of fault diagnosis. Further, in such a configuration, when the Mahalanobis distance D changes gradually due to aging degradation or environmental change of the monitoring target 10 (for example, noise such as a change in atmospheric temperature), the environmental change is determined as an abnormality of the monitoring target 10. There is an advantage that erroneous fault diagnosis is prevented.

[Modification]
The monitoring system 1 is not limited to this, and includes a means (data processing device 3) for selecting a period in which the time differential value of the abnormal signal continuously exceeds a predetermined threshold k ′ as a target period, and The abnormal signal and the related signal may be extracted from the monitoring target data in the target period (target signal extraction step ST4 and target period selection step ST5) (see FIGS. 5 and 8). That is, the abnormal signal and the related signal may be extracted based on the time differential value of the abnormal signal. Even in such a configuration, since a period in which a sign of abnormality appears remarkably can be selected as the target period, a useful abnormality signal and related signals can be extracted. This has the advantage of improving the accuracy of fault diagnosis. In addition, such a configuration has an advantage that when the Mahalanobis distance D changes gradually due to aging degradation or environmental change of the monitoring target 10, an erroneous failure diagnosis that determines such environmental change as an abnormality of the monitoring target 10 is prevented. .

  For example, in the modification shown in FIG. 8, an abnormality (a rapid increase in temperature) appears in the blade path temperature among a plurality of monitoring target data, and this blade path temperature is extracted as an abnormal signal (target signal extraction step ST4). ). Further, a time differential value of this abnormal signal (blade path temperature) is calculated (abnormal signal data calculation step ST6), and a period in which the time differential value of this abnormal signal continuously exceeds a predetermined threshold k ′ is a target period. It is selected as ta-tb (target period selection step ST5). Thus, a period in which a sign of abnormality appears prominently is selected as a target period, and abnormal signal data and related signal data are calculated based on the abnormal signal and the related signal in this target period (abnormal signal data calculation step) ST6 and related signal data calculation step ST7).

  The above configuration further includes means (data processing device 3) for selecting, as the target period, a period in which the Mahalanobis distance D continuously exceeds the predetermined threshold k within the target period, and The abnormal signal and the related signal may be extracted from the monitoring target data in the target period (target signal extraction step ST4 and target period selection step ST5) (see FIGS. 5, 6, and 8). That is, the target period may be selected by (1) a combination of the time differential value D ′ of the Mahalanobis distance D and the Mahalanobis distance D, or (2) a combination of the time differential value of the abnormal signal and the Mahalanobis distance D. . In such a configuration, a more suitable target period is selected, so that there is an advantage that the accuracy of failure diagnosis is further improved. For example, when starting a gas turbine power plant, noise tends to appear in a lot of monitoring target data. Accordingly, (1) selecting the target period in consideration of both the time differential value D ′ of the Mahalanobis distance D and the Mahalanobis distance D, or (2) both the time differential value of the abnormal signal and the Mahalanobis distance D. Accordingly, it is possible to effectively prevent erroneous diagnosis due to noise at the time of starting the plant.

  As described above, the monitoring system according to the present invention is useful in that it can automate a series of processes from monitoring of a sign of abnormality in a monitoring target to failure diagnosis.

DESCRIPTION OF SYMBOLS 1 Monitoring system 2 Monitoring apparatus 21 Mahalanobis distance calculation part 22 Abnormality determination part 23 Alarm signal generation part 24 Storage part 3 Data processing apparatus 31 Target signal extraction part 32 Target period selection part 33 Abnormal signal data calculation part 34 Related signal data calculation part 4 Failure diagnosis apparatus 10 Monitoring target 11 Gas turbine 12 Generator 41 Failure diagnosis unit 42 Bayesian network 43 Database 44 Diagnosis result generation unit 45 Display unit

Claims (4)

A monitoring system that can automate a series of processes from monitoring for signs of abnormality in the monitoring target to fault diagnosis,
Monitoring unit that obtains predetermined monitoring target data from the monitoring target, calculates a Mahalanobis distance of the monitoring target data, and detects an abnormality of the monitoring target based on the Mahalanobis distance; and an abnormality from the plurality of monitoring target data Monitoring target data (hereinafter referred to as an abnormal signal) and monitoring target data related to the abnormal signal (hereinafter referred to as a related signal) are extracted and predetermined based on the abnormal signal and the related signal. A monitoring system comprising: a data processing unit that generates an input signal; and a failure diagnosis unit that performs a failure diagnosis of the monitoring target based on the input signal.   Means for selecting a period in which the time differential value of the Mahalanobis distance continuously exceeds a predetermined threshold as a target period, and the abnormal signal and the related signal are extracted from the monitoring target data in the target period The monitoring system according to claim 1.   Means for selecting as a target period a period in which the time differential value of the abnormal signal continuously exceeds a predetermined threshold, and the abnormal signal and the related signal are extracted from the monitoring target data in the target period The monitoring system according to claim 1.   Means for selecting a period within the target period and the Mahalanobis distance continuously exceeding a predetermined threshold as the target period, and the abnormal signal and the related signal from the monitoring target data in the target period The monitoring system according to claim 2 or 3, wherein is extracted.

Images