JPH1020925A - Plant diagnostic device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a plant diagnostic device which automatically and speedily retrieves the start point and the propagation route of abnormality when inconvenience occurs and provides highly precise information on an abnormality event name, the cause and the countermeasure method. SOLUTION: The standard deviation (RMS) of n-pieces of observation data at observation time including present time in respective frequency bands dividing an observation signal is obtained in an RMS operation part 5. When an abnormal signs generation judgment part 7 judges that RMS exceeds a first threshold decided by multiplying RMS of normal time data by a prescribed positive number, an inconvenience generated day and time estimation part 8 estimates day and time when RMS exceeds a second threshold which is separately set by multiplying RMS of normal time data by a prescribed positive number based on the time shift of RMS.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a plant for monitoring an operating state based on a process signal of a large-scale plant such as a nuclear power plant, a thermal power plant, and a chemical plant to detect and diagnose an abnormality at an early stage. It relates to a diagnostic device.

[0002]

2. Description of the Related Art Generally, in a large-scale plant such as a nuclear power plant, an attempt has been made to detect an abnormality or a sign of an abnormality in the plant at an early stage based on a large number of observation process signals (observation signals). There is a plant diagnostic device in practical use in the department. In these plant diagnostic apparatuses, the following two monitoring diagnostic methods are generally used.

[0003] The first monitoring / diagnosing method statistically processes a time-dependent characteristic change of an observation signal using observation data of a past fixed time, and for example, obtains a standard deviation (hereinafter referred to as RMS) to obtain an object. When the observed value of the time including the time becomes different from the above-described statistical result, it is determined that an abnormality has occurred. For example, if the current standard deviation value is three times the normal value, it is determined to be abnormal.

[0004] The second monitoring and diagnostic technique is a technique called noise analysis, which converts time-series observation data into a frequency domain using a fast Fourier transform (FFT) or the like to obtain a plurality of frequency wave bands, For example, it is divided into low frequency, medium frequency, and high frequency, and when the frequency dependency of the power spectrum density (PSD) of the observed signal in each frequency band is different from the reference frequency dependency at the normal time, it is determined that an abnormality has occurred. Is made. For example, when the square root (frequency dependence) of the current PSD addition amount is three times the value in the normal state, it is determined that an abnormality has occurred.

[0005] A plant diagnostic apparatus employing both of these monitoring and diagnostic methods determines abnormally the time change of the statistic of the observation signal including all frequency components and the difference from the normal data of the spectrum in a plurality of frequency bands. This is effective for early detection of abnormalities.

[0006]

However, in these monitoring and diagnosing methods, a small abnormality (sign of abnormality) that hardly affects the stable operation of the plant in each frequency band causes an abnormal condition (temporary stop) of the plant. The disadvantage is that it is not possible to track the growth process and to predict future growth.

SUMMARY OF THE INVENTION It is an object of the present invention to automatically and promptly search for a starting point and a propagation path of an abnormality when a problem occurs, and to provide highly accurate information on an abnormal event name, its cause, and a countermeasure. It is to obtain a plant diagnostic device that can perform the above.

[0008]

According to a first aspect of the present invention, there is provided an RMS which divides an observation signal into a plurality of predetermined frequency bands and calculates a standard deviation of n pieces of observation data at the observation time including the present time in the frequency band. A calculating unit, an abnormal sign occurrence determining unit that determines that an abnormal sign has occurred when the standard deviation exceeds a first threshold value, and a standard deviation based on a time transition of the standard deviation when the occurrence of the abnormal sign is detected. Has a defect occurrence date and time estimating unit that estimates the date and time exceeding the second threshold value and outputs the estimated date and time as the defect occurrence date and time.

According to the first aspect of the present invention, the standard deviation of n pieces of observation data at the observation time including the present time in each frequency band obtained by dividing the observation signal is obtained by the RMS calculation unit, and the standard deviation is set to the first threshold. When it is determined by the abnormal symptom occurrence determining unit that it has exceeded, the failure occurrence date and time estimating unit estimates the date and time when the standard deviation exceeds a second threshold value that is separately set based on the time transition of the standard deviation.

According to a second aspect of the present invention, in the first aspect of the present invention, the first threshold value and the second threshold value are determined by multiplying the standard deviation of the normal data by different predetermined positive numbers. It was done.

According to the second aspect of the present invention, when the standard deviation exceeds a first threshold determined by multiplying the standard deviation of the normal data by a predetermined positive number, it is determined that an abnormal symptom is indicated.
The expected date and time when the standard deviation exceeds a second threshold separately set by multiplying the standard deviation of the normal time data by a predetermined positive number is estimated as the date and time when the failure occurred.

According to a third aspect of the present invention, in the first aspect of the present invention, after the abnormal sign occurrence determining section detects the occurrence of the abnormal sign, the RMS calculating section shortens the cycle for calculating the standard deviation,
The failure occurrence date and time estimating unit is configured to successively estimate the date and time exceeding the second threshold value from the time transition of the standard deviation while updating the date and time.

According to a third aspect of the present invention, in addition to the operation of the first aspect, after the standard deviation exceeds the first threshold value, the period for calculating the standard deviation is shortened, and the time transition from the time transition to the first period is calculated. The estimation is sequentially performed while updating the date and time exceeding the threshold value of 2.

According to a fourth aspect of the present invention, in the first aspect of the present invention, the RMS calculation unit calculates the standard deviation in a frequency band of interest by performing a fast Fourier transform in a predetermined observation cycle. The square root of the addition amount of the power spectrum density is calculated, and the standard deviation is calculated.

In the invention of claim 4, in addition to the effect of the invention of claim 1, attention has been paid to calculating the square root of the addition amount of the power spectrum density calculated by performing fast Fourier transform in a predetermined observation period. Calculate the standard deviation in the frequency band.

According to a fifth aspect of the present invention, in the first aspect of the invention, when calculating the standard deviation in a frequency band of interest, the RMS calculation unit converts a plurality of observation signals with a band-pass filter to the observation signals. The standard deviation is calculated by dividing the frequency band into and calculating various statistics.

According to a fifth aspect of the present invention, in the first aspect of the present invention, the observed signal is divided into a plurality of frequency bands by a band-pass filter, and various statistics are calculated. Calculate the standard deviation at.

According to a sixth aspect of the present invention, in place of the standard deviation of the first aspect of the present invention, a value obtained by dividing the standard deviation by an average value in a predetermined time is used. The second threshold value is a value obtained by multiplying a value obtained by dividing a standard deviation of normal data by an average value at a predetermined time and another predetermined positive number.

According to the sixth aspect of the present invention, the value obtained by dividing the standard deviation by the average value at the predetermined time is obtained by multiplying the value obtained by dividing the standard deviation of the normal data by the average value at the predetermined time by an arbitrary positive number. When the determined first threshold is exceeded, it is determined that an abnormal sign has occurred, and the date and time exceeding the second threshold separately set by multiplying the standard deviation of the normal data by a predetermined positive number occurs. Estimate the date and time.

The invention of claim 7 is the first to sixth aspects of the present invention.
In the invention of the above, when the standard deviation exceeds a second threshold value, a failure occurrence determination unit that determines that a failure has occurred, and when it is determined that a failure has occurred, a signal that has caused the occurrence of the abnormality and a transient And an abnormal propagation path estimating unit for estimating a change propagation path.

According to the seventh aspect of the present invention, in addition to the functions of the first to sixth aspects, when the standard deviation actually exceeds the second threshold value, the qualitative analysis of the process signal including the observation signal is performed. By extracting the behavior and comparing and referring to the physical causal relationship between the signals created in advance, the signal that is the starting point of the sudden change or the transient change and the propagation path of the sudden change or the transient change are estimated.

According to an eighth aspect of the present invention, based on the seventh aspect of the present invention, the abnormal condition is compared with a previously stored knowledge base based on the power spectrum density, the signal that has caused the occurrence of the abnormality and the propagation path of the transient change. It has a neural network that provides event names, their causes, and how to deal with them.

According to an eighth aspect of the present invention, in addition to the operation of the seventh aspect, when the standard deviation exceeds a second threshold value, the neural network receives the power spectrum density distribution, the origin of the abnormality, and its propagation path as inputs. To provide the user with the name of the abnormal event, its cause, and a countermeasure against the event registered in advance by using.

According to a ninth aspect of the present invention, in the invention of the eighth aspect, when an event unknown to the neural network occurs, or when a difference between the result of the determination by the neural network and the actual abnormal event name and its cause occurs. Further, there is provided a learning means for learning an abnormal event name, its cause, and a coping method for the unknown event, and accumulating it as a new knowledge base.

According to the ninth aspect of the present invention, in addition to the operation of the eighth aspect, when an event unknown to the neural network occurs, or when the difference between the decision result by the neural network and the actual event name or its cause is determined. When this occurs, it is accumulated and learned as a knowledge base to improve the accuracy of determination when a similar event occurs.

[0026]

Embodiments of the present invention will be described below. FIG. 1 is a block diagram of a plant diagnostic apparatus according to an embodiment of the present invention. After the data from the plant 1 is collected in the data collection device 2, the plant diagnosis device 3 diagnoses whether an abnormality has occurred in the plant. Then, the diagnosis result is displayed on the display device 15.

The FFT operation unit 4 of the plant diagnosis device 3
An observation signal from the data collection device 2 is input at a predetermined cycle, and a fast Fourier calculation process is performed. That is, the observation signal is divided into a plurality of predetermined frequency bands, and the observation data n at the observation time including the present time in the frequency band is obtained.
Get. For example, time-series data a of a process signal of a nuclear power plant, such as a neutron flux and a core flow rate, which are constantly monitored, are input at fixed time intervals, and the frequency of each signal obtained by transforming the frequency domain by a fast Fourier (FFT) or the like is obtained. Get the properties. That is, a pattern signal b of the power spectrum density (PSD) is calculated. As shown in FIG. 2, PSD (k) in the frequency band k (f1 to fk) is standardized such that the maximum value is 1 on the vertical axis.

The PSD obtained by the FFT operation unit 4 is RMS
The data is input to the arithmetic unit 5, where the standard deviation (RMS) is calculated. RMS is the standard deviation of n pieces of observation data at the observation time Δt including the current time t in the frequency band k. Then, the RMS calculation unit 5 calculates the RMS (k) in the frequency band k of interest at time t by integrating the signal PSD (k, i) with respect to the N frequency components i and taking the square root thereof. Desired. That is, the RMS calculation unit 5 calculates the standard deviation RMS (k) in the frequency band k according to the following equation (1).
Ask for.

[0029]

(Equation 1)

Here, the number of frequencies for which the PSD is calculated in the frequency band k is i (i = 1 to N), and the number i of the frequencies is the resolution required for performing a diagnosis by noise analysis. I am satisfied.

In the above description, the RMS in each frequency band k
Is calculated by integrating the PSD of the observation signal s by FFT, but the R in the frequency band k of interest is calculated.
MS (k) can also be calculated by applying a filtering circuit (bandpass filter) and a statistic calculation circuit to the time-series observation signal s. The RMS (k) calculated in this manner is input to the abnormal symptom occurrence determination section 7.

The abnormal symptom occurrence judging section 7 includes an RMS calculating section 5
And the standard deviation of the normal data stored in the normal RMS storage unit 6 to determine whether an abnormal sign has occurred in the observation signal.

That is, the abnormal symptom occurrence determining section determines whether or not the standard deviation RMS (k) of n pieces of observation data in the calculation cycle Δt including the current time t in the frequency band k has exceeded the first threshold value. When it is determined that the value exceeds the first threshold, it is determined that an abnormal town has occurred. The first threshold is determined by multiplying the standard deviation NRMS (k) calculated from the past normal observation data by an arbitrary positive number a.

When the abnormal sign occurrence judging section 7 judges that an abnormal sign has occurred, the malfunction occurrence date / time estimating section 8 is started. The failure occurrence date and time estimating unit 8 tracks the time change of the standard deviation RMS (k) after it is determined that an abnormal symptom has occurred, and determines the occurrence of a failure from the time transition represented by the change rate. A date and time exceeding a set second threshold is predicted. The prediction result is displayed on the display device 15. The second threshold value is a standard deviation N of the normal data.
It is determined by multiplying RMS (k) by an arbitrary positive number b. Further, the positive numbers a and b determined when setting the first threshold value and the second threshold value may be a <b or a> b.

Here, when the occurrence of an abnormal sign is detected by the abnormal sign occurrence determining section 7, the standard deviation RMS in RMS is used.
Is shortened. That is, after the standard deviation RMS (k) exceeds the first threshold value, the period for calculating the standard deviation RMS is shortened after the standard deviation RMS (k) exceeds the first threshold value in order to improve the accuracy of estimating the date and time of occurrence of the failure. Is sequentially updated while updating the date and time exceeding the threshold value.

Here, the failure date and time estimating unit 8 is shown in FIG.
As shown in (a), b = tan θ1 = (ΔRMS (t, k) / Δt) as a time change rate d of RMS (k) when the standard deviation RMS (k) exceeds the first threshold value. Is used to predict the date and time that will exceed the second threshold. At this time, after exceeding the first threshold, FIG.
As shown in (2), by setting the time period for calculating the standard deviation RMS (k) to Δt ′ (Δt ′ <Δt), the estimated failure occurrence date and time determined by exceeding the second threshold is updated.

Next, the fault occurrence judging section 9 detects that the standard deviation has exceeded the second threshold value. If the standard deviation actually exceeds the second threshold value, the fault occurrence judging section 9 determines that the standard deviation has exceeded the second threshold value. 9 activates the abnormal discourse route estimation unit 10.

When it is determined that a failure has occurred, the abnormal propagation path estimating section 10 estimates the signal that has become the starting point of the occurrence of the abnormality and the propagation path of the transient change. That is, the standard deviation is set to the second threshold value by extracting the increase and decrease of the process signal including the observation signal, and other qualitative behaviors, and comparing and extracting the qualitative behavior with the physical causal relationship between the signals created in advance. Is estimated, the signal that has become the starting point of the sudden or transient change when the threshold value is exceeded, and the propagation path of the sudden or transient change are estimated.

As an estimation method by the abnormal propagation path estimation unit 10, for example, an estimation method disclosed in Japanese Patent Application No. 6-91820 (Japanese Patent Application Laid-Open No. 7-294695) by the same applicant as the present application is applied. . FIG. 4 shows a configuration diagram of the abnormal propagation path estimating unit 10. Abnormal propagation path estimating unit 10
Is the observation signal s read by the data collection device 2.
A change pattern extracting unit 10a for extracting an initial qualitative change, a causal relationship database 10b for recording a qualitative causal relationship between observation signals s created in advance, and an initial pattern extracted by the change pattern extracting unit 10a. Is compared with the qualitative causal relationship data between the observation signals recorded in the causal relationship database 10b, there is a causal relationship that is assumed to have caused a sudden change or a transient change in the plant. Propagation route estimation means 1 for estimating the route
0c and the signal located at the most upstream of the estimated route is extracted as the starting point of sudden change or transient change, and narrowing down the cause candidate of sudden change or transient change is performed based on the signal that has become the starting point of change and its behavior. And a starting signal extracting means 10d for performing the starting point signal extraction.

When it is determined that a failure has occurred, the abnormal propagation path estimating unit 10 extracts a failure origin signal. That is, the change pattern extraction unit 10a extracts an initial qualitative change of the observation signal s read by the data collection device 2. The propagation path estimating means 10c includes:
The initial qualitative change extracted by the change pattern extracting means 10a is compared with the qualitative causal relationship data between the observation signals recorded in the causal relationship database 10b, and the sudden change and the transient change of the plant are propagated. A causal route that is assumed to have been performed is estimated. Then, the starting point signal extracting means 10d extracts the signal located at the most upstream of the estimated route as the starting point of the sudden change or the transient change, and performs the sudden change or the transient based on the signal which is the starting point of the change and its behavior. Narrow down potential change causes. As a result, the starting point of the defect and its propagation path are extracted.

Next, the neural network 11 preliminarily detects an abnormal state based on the power spectrum density from the FFT operation unit 4, the signal from the abnormal propagation path estimating unit 10 that has caused the occurrence of the abnormality, and the propagation path of the transient change. Compared with the knowledge base stored in the cause coping method storage unit 12, the name of the abnormal event, its cause, and a coping method are provided.

When an event unknown to the neural network 11 occurs, or when there is a difference between the result of the determination by the neural network 11 and the actual abnormal event name and its cause, the output interpreting unit 13 sets the learning unit to the learning unit. 14 is started. The learning unit 14 learns the name of the abnormal event, its cause, and a countermeasure for the unknown event, and accumulates it in the abnormal cause countermeasure storage unit 12 as a new knowledge base.

Further, the determination of occurrence of an abnormality by the neural network 11, the abnormality cause coping unit 12, the output interpreting unit 13, and the learning unit 14 can be performed, for example, by the same applicant as in the present application (Japanese Patent Application No. 1-228411). The determination method disclosed in Japanese Unexamined Patent Publication No. 3-92795) is applied. That is, the observation signal s, which is the time series data of the process signal of the nuclear power plant such as the neutron flux and the core flow rate that are constantly monitored, is
The frequency characteristic (PSD pattern signal) of each signal obtained by transforming the frequency domain by FFT or the like at regular time intervals is used as an input.

As shown in FIG. 2, PSD in frequency band k
k is normalized such that the maximum value on the vertical axis is 1, and the input to the neural network 11 divides the frequency domain as far as the resolution required for diagnosis can be obtained. Then, the standardized PSD value (Pi) at each frequency division point (fi) is used. Neural net 11
Identifies the given pattern for PSDk and R
Time change rate d of MS (k) and abnormal propagation path estimating unit 10
In addition to the information on the propagation path and the starting point of the abnormality from, the cause of the abnormality and the method of coping with the abnormality are provided using known information learned in advance.

FIG. 5 is a configuration diagram of the neural network 11. In the input layer of the neural network, each frequency division point (f
Normalized PSD value (Pi) in i), RMS
The change in time (k), the failure detection signal, the start signal of the abnormality, and the like are input, and the output layer calculates which abnormal event corresponds to it. For example, if the output pattern is as shown in FIG. 6, the event 2 is an abnormal event that has occurred.

In this way, as shown in FIGS. 5 and 6, the name of the generated event, its cause, and a method of coping with it are provided by using the neural network 11. That is,
If the fault event is event 2 of a known event, the neurons in the output layer are excited, the name of the event is immediately identified, and the cause of the abnormality and its countermeasure are provided from the knowledge base of the countermeasure for abnormality storage unit 12. You. This diagnosis result is displayed on the display device 1
5 is displayed.

On the other hand, when the malfunction event is the unknown event X, the output neuron for the unknown event X is determined, the PSDk pattern of the unknown event X, the time change of RMS (k), the starting point of the abnormality, and the propagation path of the abnormality. Is given a correct answer pattern that excites this neuron, and learning is performed to adjust the weight of the network to the maximum.

In the above description, the determination of occurrence of an abnormal symptom and the prediction of the date and time of occurrence of a malfunction are performed using RMS (k). However, the characteristic that RMS (k) increases as the magnitude (gain) increases. With respect to the signal having the same, the value obtained by dividing the standard deviation RMS (k) of n pieces of observation data at the observation time Δt including the present time t in the frequency band k by the average value AVE (k) at Δt is the standard value of the normal data. When the difference exceeds a first threshold value a × NRMS (k) / AVE (k) determined by multiplying a value obtained by dividing the deviation NRMS (k) by AVE (k) by an arbitrary positive number a, the normal state A second threshold value b separately set by multiplying the standard deviation of the data by an arbitrary positive number b
A method of estimating the date and time exceeding × NRMS (k) / AVE (k) based on the time transition of the standard deviation RMS (K) / AVE (k) is effective.

[0049]

As described above, according to the present invention, it is possible to estimate the date and time when an abnormal symptom has occurred and to estimate the date and time when the abnormality will grow into a defect in the future. Can be repaired,
The operation rate and reliability of the plant can be improved.
In addition, when a failure occurs, even an inexperienced technician who is not familiar with the transient characteristics of the plant by the qualitative model can estimate the starting point and propagation path of the abnormality in a short time. Furthermore, the name of the trouble event can be known by the neural network, and the best choice can be made as a measure for the cause by the knowledge base.

[Brief description of the drawings]

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

FIG. 2 is an explanatory diagram showing a normalized power spectrum density distribution PSD (k) in a frequency band k according to the embodiment of the present invention.

FIG. 3 is an explanatory diagram showing a method of obtaining an estimated failure occurrence date and time from a standard deviation change rate at the time of detection of an abnormal sign in a failure occurrence date and time estimation unit according to the embodiment of the present invention.

FIG. 4 is a configuration diagram of an abnormal propagation path estimation unit according to the embodiment of the present invention.

FIG. 5 is a configuration diagram of a neural network according to the embodiment of the present invention.

FIG. 6 is an explanatory diagram showing an output pattern of the neural network according to the embodiment of the present invention.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 plant 2 data collection device 3 plant diagnostic device 4 FFT calculation unit 5 RMS calculation unit 6 RMS storage unit under normal condition 7 abnormal symptom occurrence determination unit 8 failure occurrence date / time estimation unit 9 failure occurrence determination unit 10 abnormality propagation path estimation unit 11 neural network 12 Error cause countermeasure storage unit 13 Output interpretation unit 14 Learning unit 15 Display device

Claims (9)

[Claims] An RMS operation unit that divides an observation signal into a plurality of predetermined frequency bands and calculates a standard deviation of n pieces of observation data in an observation time including the present time in the frequency band; An abnormal sign occurrence determining unit that determines that an abnormal sign has occurred when the threshold value is exceeded, and the standard deviation exceeds a second threshold based on a time transition of the standard deviation when the occurrence of the abnormal sign is detected. A plant diagnosis apparatus comprising: a failure occurrence date and time estimating unit that estimates a date and time and outputs the date and time as a failure occurrence date and time. 2. The method according to claim 1, wherein the first threshold value and the second threshold value are determined by multiplying a standard deviation of normal data by another predetermined positive number. The plant diagnostic apparatus according to the above. 3. After the abnormal sign occurrence judging section detects the occurrence of the abnormal sign, the RMS calculating section shortens the cycle for calculating the standard deviation, and the malfunction occurrence date and time estimating section determines the time of the standard deviation. The plant diagnostic apparatus according to claim 1, wherein the estimation is performed sequentially while updating the date and time exceeding the second threshold value from the transition. 4. The RMS calculation unit calculates a square root of an addition amount of a power spectrum density calculated by performing a fast Fourier transform in a predetermined observation cycle when calculating a standard deviation in a frequency band of interest. The plant diagnostic apparatus according to claim 1, wherein the standard deviation is calculated. 5. The RMS calculation unit calculates a standard deviation in a frequency band of interest, divides the observation signal into a plurality of frequency bands by a band-pass filter, and calculates various statistics. The plant diagnostic apparatus according to claim 1, wherein the standard deviation is calculated by performing the calculation. 6. The standard deviation may be a value obtained by dividing the standard deviation by an average value at a predetermined time, and the first threshold value and the second threshold value may be a standard deviation of normal data. 2. The plant diagnostic apparatus according to claim 1, wherein a value obtained by dividing by an average value at a predetermined time is multiplied by another predetermined positive number. 7. A malfunction occurrence judging unit for judging that a malfunction has occurred when the standard deviation exceeds a second threshold value, and a starting point of the occurrence of the malfunction when it is determined that the malfunction has occurred. The plant diagnostic apparatus according to any one of claims 1 to 6, further comprising: an abnormal propagation path estimating unit that estimates a signal and a propagation path of a transient change. 8. Based on the power spectrum density, the signal from which the occurrence of the abnormality has started, and the propagation path of the transient change, a comparison is made with a knowledge base stored in advance to determine the name of the abnormal event, its cause, and a countermeasure. The plant diagnostic apparatus according to claim 7, further comprising a neural network to be provided. 9. When an event unknown to the neural network occurs, or when there is a difference between the result of determination by the neural network and an actual abnormal event name and its cause, an abnormal event name corresponding to the unknown event 9. The plant diagnostic apparatus according to claim 8, further comprising learning means for learning the cause and the coping method thereof, and accumulating the learned information as a new knowledge base.