JPH064789A - Method and device for monitoring abnormality of equipment - Google Patents

Abstract

PURPOSE:To improve detection accuracy by detecting many monitoring signals from equipment difficult to grasp a normal value, extracting the relevance between those many monitoring signals, and detecting the initial sign of the equipment abnormality. CONSTITUTION:Among the many monitoring signals, the only component required for the abnormality monitoring of the equipment is extracted from the monitoring signal changing at a high speed (steps 1-5). The monitoring signal corresponding to the object, explanation variable name list stored in advance is read and the regression coefficient between the monitoring signals is calculated (steps 6-12). By calculating the allowable range of the regression coefficient to discriminate whether or not the regression coefficient is within the allowable range. If it is outside the range, a warning is displayed (Steps 13-16). Then, the time depending change rate of the regression coefficient is calculated and the allowable time wise change rate is also calculated to discriminate whether or not the time depending change rate of the recursion coefficient is within the allowable value. In the case it is outside the allowable value, the warning is displayed (steps 17-20).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus abnormality monitoring method and apparatus for monitoring whether an operating state of an apparatus used in a plant is normal or abnormal,
In particular, the present invention relates to an apparatus abnormality monitoring method and apparatus that uses not only a monitoring signal for the apparatus itself but also a monitoring signal for related equipment such as a surrounding coolant loop.

[0002]

2. Description of the Related Art For example, there is an abnormality monitoring method or apparatus for detecting the vibration or the rotation speed of equipment used in a nuclear power plant and monitoring the initial sign that the equipment is abnormal. As an example of the above-mentioned abnormality monitoring method, Japanese Patent Laid-Open No. 62-14210
There is an abnormality diagnosis analysis method described in Japanese Patent Publication No. In this abnormality diagnosis analysis method, for example, the change in the observation signal of the equipment used in the nuclear power plant is expressed as a vector, and this vector is analyzed to diagnose the initial sign of the abnormality phenomenon of the equipment. Further, as an example of the abnormality monitoring device, there is an abnormality inspection device for a rotating machine described in Japanese Patent Laid-Open No. 1-234083. This abnormality inspection device was made to speed up the inspection of rotating machinery.The regression coefficient between the rotation speed and sound pressure was obtained by regression analysis, and the rotation speed and sound pressure of the rotating machine were calculated from the obtained regression coefficient. Predict normal characteristics of. Then, it is configured to determine whether the device is normal or abnormal depending on how much the rotational speed and the sound pressure deviate from the normal characteristics.

Further, as another example, JP-A-61-119
There is a movable body motion characteristic evaluation device described in Japanese Patent No. 9114. In this operation characteristic evaluation device, a regression coefficient of the drive signal and the displacement amount of the movable body is obtained by regression analysis,
The obtained regression coefficient is compared with a predetermined normal value. Then, it is configured to determine whether the operation of the movable body is normal or abnormal based on the comparison result. Further, as another example, there is a control system monitoring device described in JP-A-60-19207. In this control system monitoring device, a control system model is prepared in advance, and regression coefficients of this control system model are obtained by regression analysis. The control system is configured to determine whether the control system is normal or abnormal depending on how much the regression coefficient of the actual control system deviates from the regression coefficient of the control system model.

[0004]

By the way, in a large-scale plant such as a nuclear power plant, it is important to detect the early sign of abnormality of the equipment used. Therefore, it is desirable to detect as many monitoring signals as possible from the equipment being used and its associated equipment and to detect the early abnormal sign early based on these many monitoring signals. Therefore, the above-mentioned Japanese Patent Laid-Open No. 62-14
It is conceivable to improve the detection accuracy of the early sign of abnormality by applying the abnormality diagnosis analysis method described in Japanese Patent No. 210. However, in a nuclear power plant or the like, the dynamic characteristics are complicated, and the normal value itself of the monitoring signal changes for each plant and for each periodic inspection, making it difficult to grasp the normal value. Therefore, even if the above-mentioned abnormality diagnosis analysis method is applied to a nuclear power plant as it is, improvement in detection accuracy of the initial abnormality sign cannot be expected.

Further, the above-mentioned Japanese Patent Laid-Open Nos. 1-234083, 61-199114, and 60-1.
It is conceivable to apply regression analysis as described in Japanese Patent No. 9207 to improve the detection accuracy of abnormal early signs in a nuclear power plant. However, the above-mentioned Japanese Patent Laid-Open Nos. 1-234083 and 61-199114
In the regression analysis described in Japanese Patent Laid-Open No. 60-19207 and Japanese Patent Laid-Open No. 60-19207, the number of variables is small and it is not suitable for processing a large number of monitoring signals. Furthermore, in a nuclear power plant, the characteristics of many monitoring signals are not all the same. For example, the vibration signal and the rotation speed signal of the recirculation pump fluctuate at high speed, and changes in temperature, flow rate, etc. are slow. How to perform regression analysis on these monitoring signals having different characteristics is not described in the above publication, and it has been difficult to improve the detection accuracy of the early sign of abnormality.

An object of the present invention is to detect a large number of monitoring signals from a device for which it is difficult to grasp a normal value, extract the relationship between these many monitoring signals, detect an initial sign of abnormalities in the device, and detect the detection accuracy. To realize an abnormality monitoring method and an abnormality monitoring device.

[0007]

In order to achieve the above object, the present invention is configured as follows. In the device abnormality monitoring method, a step of detecting a large number of characteristics of the monitored device or the monitored device and its peripheral equipment and capturing the detected large number of characteristics as a large number of monitoring signals Regression analysis, step of calculating regression coefficient, step of executing validity evaluation of the calculated regression coefficient, judgment whether the calculated regression coefficient is within a predetermined normal range, and a predetermined normal range If it is not within the range, it is judged that an abnormality has occurred in the device, or the time change rate of the regression coefficient is calculated, and it is judged whether or not the calculated time change rate exceeds a predetermined normal change rate, A step of determining that an abnormality has occurred in the device when the rate of change exceeds a predetermined normal rate.

Further, in the device abnormality monitoring method, a step of detecting a large number of characteristics of the monitored apparatus or the monitored apparatus and its peripheral equipment, and taking in the detected large number of characteristics as a large number of monitoring signals, and a large number of taken in signals. Of the monitoring signals of, the step of separating and extracting only the component necessary for monitoring from the high-speed characteristic monitoring signal showing the characteristics that change at high speed,
Multiple regression analysis is performed on the monitoring signals other than the high-speed characteristic monitoring signal and the components necessary for monitoring the high-speed characteristic monitoring signal, or multiple regression analysis is performed on the components necessary for monitoring the high-speed characteristic monitoring signal and regression is performed. A step of calculating a coefficient, a step of executing a validity evaluation of the calculated regression coefficient, and a judgment of whether or not the calculated regression coefficient is within a predetermined normal range,
If it is not within the predetermined normal range, it is judged that an abnormality has occurred in the device, or the time change rate of the regression coefficient is calculated and whether the calculated time change rate exceeds the predetermined normal change rate. And a step of determining that an abnormality has occurred in the device when the rate of change exceeds a predetermined normal change rate.

Preferably, the predetermined normal range of the regression coefficient is calculated as a function of the operating state by multiple regression analysis at the time of trial operation of the device, and the normal range is set according to the operating state of the device. Further, preferably, after extracting only the components necessary for monitoring from the high-speed characteristic monitoring signal, the extracted components are averaged, and the components subjected to the averaging are used for calculating the regression coefficient.

In the device abnormality monitoring device, a detection means for detecting a large number of characteristics of the monitored device or the monitored device and its peripheral equipment as a large number of monitoring signals, and a multiple regression analysis of the detected large number of monitoring signals, Correlation extraction means for calculating the regression coefficient and performing validity evaluation of the calculated regression coefficient, and determining whether the calculated regression coefficient is within a predetermined normal range, and is not within the predetermined normal range Occasionally, it is judged that an abnormality has occurred in the equipment, or the time change rate of the regression coefficient is calculated, and it is judged whether or not the calculated time change rate exceeds a predetermined normal change rate. And an abnormality determining means for determining that an abnormality has occurred in the device when the rate is exceeded.

Further, in the device abnormality monitoring apparatus, a detecting means for detecting a large number of characteristics of the monitored device or the monitored device and its peripheral equipment as a large number of monitoring signals, and a high speed of the detected large number of monitoring signals. Separates only the components required for monitoring from the high-speed characteristic monitoring signal that shows changing characteristics,
Component separation means to extract, the output signal from the component separation means, and a multiple regression analysis of the monitoring signal other than the high-speed characteristic monitoring signal, or multiple regression analysis only the output signal from the component separation means, Correlation extraction means for calculating the regression coefficient and performing validity evaluation of the calculated regression coefficient, and determining whether the calculated regression coefficient is within a predetermined normal range, and is not within the predetermined normal range Occasionally, it is judged that an abnormality has occurred in the equipment, or the time rate of change of the regression coefficient is also calculated, and it is judged whether the calculated time rate of change exceeds a predetermined normal change rate. An abnormality determining unit that determines that an abnormality has occurred in the device even when the rate is exceeded.

[0012] Preferably, the predetermined normal range of the regression coefficient is calculated as a function of the operating state by multiple regression analysis at the time of trial operation of the equipment, and is stored in the storage means.
The predetermined normal range of the stored regression coefficient is set as a normal range according to the operating state of the device. Further, preferably, the component separating means has an averaging processing part for averaging the signals separated into the components, and is configured to supply the signal averaged by the averaging part to the correlation extracting means. Furthermore, preferably, the signal averaged by the averaging unit is supplied to the correlation extracting unit and the abnormality determining unit, and the abnormality determining unit outputs the output signal from the averaging unit and the monitoring signal other than the high-speed characteristic monitoring signal. It is configured to determine the operating state of the equipment from. Further, preferably, the component separating means has a frequency component analyzing section for analyzing the frequency component of the high speed characteristic monitoring signal, and the averaging processing section comprises a low pass filter.
Further, preferably, the component separation means has a plurality of bandpass filters, the averaging processing unit, the average response processing unit for adding the output signals of the bandpass filter, and the effective value of the output signal of the bandpass filter. An effective value calculation unit that calculates,
And an output selection unit that selects and outputs the output signal from the average response processing unit and the output signal from the effective value calculation unit.

Further, preferably, the component separating means is analyzed by a coherence analyzing unit for analyzing a frequency band having a strong correlation and a frequency band having a low correlation between two signals of the high speed characteristic monitoring signal, and analyzed by the coherence analyzing unit. A correlation band determining unit that determines a frequency band having a strong correlation and a frequency band having no correlation based on the frequency band, and a pass band selecting unit that passes a signal in a predetermined frequency band among output signals from the correlation band. , And the averaging processing section is composed of an effective value calculating section for calculating an effective value of the output signal from the pass band selecting section. Further, preferably, the device for which abnormality monitoring is performed is a rotating device, and the component separating means executes signal processing with reference to a predetermined rotation angle of the device.

Further, preferably, the abnormality determining means is a mode setter for selecting and setting the learning mode and the monitoring mode, and the regression coefficient calculated when the learning mode is set by the mode setter. And a large number of monitoring signals are taken in, and the regression width and fluctuation range data relating to the normal fluctuation range of a large number of monitoring signals are calculated and stored. When the monitoring mode is set, the stored fluctuation range data is output. When the monitoring mode is set by the fluctuation range data learning device and the mode setting device, the regression coefficient and the normal range of a large number of monitoring signals are calculated based on the fluctuation range data, and the actual regression coefficient and the large number of An abnormality detector that determines whether or not the monitoring signal is within the normal range, and determines that an abnormality has occurred in the device when the monitored signal is not within the normal range.

Further, preferably, the abnormality discriminating means is supplied with the calculated regression coefficient and a large number of monitoring signals, and separates the main monitoring parameter and the dependent monitoring parameter from the large number of monitoring signals and outputs them. Means, a mode setter for selecting and setting the learning mode and the monitoring mode, and a mode setting device, when the learning mode is set, in correspondence with each of the plurality of quantized values of the main monitoring parameter. The fluctuation range data for the normal fluctuation range of each dependent monitoring parameter is calculated and stored, and when the monitoring mode is set, the fluctuation range of the dependent monitoring parameter stored corresponding to the actual quantized value of the main monitoring parameter. When the monitoring mode is set by the monitoring parameter learning device that outputs data and the mode setting device, based on the fluctuation range data An abnormality detector that calculates the normal range of the dependent monitoring parameters, determines whether the actual dependent monitoring parameters are within the above normal range, and determines that an abnormality has occurred in the device when not within the normal range, Equipped with.
Preferably, the fluctuation range data is the average value and standard deviation of each dependent monitoring parameter.

[0016]

[Operation] Multiple detected signals are subjected to multiple regression analysis,
The regression coefficient is calculated. The validity evaluation of the calculated regression coefficient is executed, and the validity evaluation removes the regression coefficient meaningless for the equipment abnormality monitoring. Then, an abnormality occurs in the device depending on whether the regression coefficient useful for equipment abnormality monitoring is within the predetermined normal range or whether the time change rate of the regression coefficient exceeds the predetermined normal change rate. It is determined whether or not. Further, the high-speed characteristic monitor signal of the detected many monitor signals may include a component unnecessary for monitoring. In this case, of the high-speed characteristic monitor signal, unnecessary components for monitoring are removed, and only essential components are extracted. The regression coefficient of the extracted component and the monitor signal other than the high-speed characteristic monitor signal is calculated by the multiple regression analysis. Whether the calculated regression coefficient is within the specified normal range or whether the time rate of change of the regression coefficient exceeds the specified normal change rate is determined, and whether or not a device abnormality has occurred Is judged. As a result, the detection accuracy of the abnormal initial sign of the device, which is difficult to grasp the normal value, is improved.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is an operation flowchart of an embodiment of a device abnormality monitoring method according to the present invention, and FIG. 2 is a schematic configuration diagram of an embodiment of an equipment abnormality monitoring device according to the present invention. The examples of FIGS. 1 and 2 are examples in which the present invention is used for abnormality monitoring of a recirculation pump loop of a nuclear power plant.

First, in FIG. 2, the heat generated in the core 102 inside the nuclear reactor 101 is transferred to the cooling water flowing inside the core 102, and the cooling water boils and the main steam pipe 10
A turbine / generator (not shown) is passed through 3 to obtain electric power. The steam that has rotated the turbine / generator becomes water again and is guided to the reactor 101 through the water supply pipe 104. The heat generation amount of the core 102 changes depending on the flow rate of the recirculation pump loop including the recirculation pump 105, the recirculation pipe 106, and the jet pump 107. In order to maintain the purity of the cooling water passing through the reactor core 102, the piping, etc., the cooling water is branched from the recirculation piping 106, purified by the reactor water purification device 108, and returned to the water supply piping 104. In this embodiment, as described above, the abnormality of the recirculation pump loop that controls the heat generation amount of the core 102 is monitored. For monitoring, a vibration measuring device 121, an acoustic measuring device 122, a rotation speed measuring device 123, a bearing temperature measuring device 133, etc. for the recirculation pump 105 are installed. Further, various measuring devices such as the flow rate measuring device 132 for knowing the operating state of the recirculation pump loop are provided. In order to monitor the water quality having information on the operating state of the recirculation pump loop, a sensor is arranged in the reactor water purification device 108 and the electric conductivity is measured by the electric conductivity measuring device 131. The vibration measuring device 121, the acoustic measuring device 122, and the rotation speed measuring device 123.
Is a monitoring signal measuring device that changes at a high speed, and the bearing temperature measuring device 133, the flow rate measuring device 132, and the conductivity measuring device 131 are measuring signals that change at a low speed. It should be noted that measurement devices other than the illustrated measurement device are also arranged.

Next, one embodiment of the abnormality monitoring method of the present invention will be described with reference to FIGS. In step 1 of FIG. 1, a data decimation number is set from the monitoring signal of the recirculation pump loop, which changes rapidly. In step 2, the components necessary for abnormality monitoring are separated from the components (for example, frequency components) of the monitoring signal that change rapidly. Next, in step 3, the components necessary for the separated abnormality monitoring are averaged to suppress random time fluctuations. Then, in step 4, for example, a selection process such as selection of a predetermined number of large components among the components required for abnormality monitoring is performed. The monitoring signal that changes at high speed contains a large number of frequency components, and also contains components unnecessary for monitoring. The frequency components unnecessary for this monitoring are removed by the above steps 2, 3, 4 and only the essential components are extracted. Next, in step 5, it is determined whether or not data processing has been performed by the set thinning number. If the data processing for the set thinning number has not been completed, the process returns to step 2. If the data processing for the set thinning number has been completed, the process proceeds to step 6. In this step 6, the purpose and explanatory variable name (multiple regression equation) list stored in an appropriate storage means is read.

Here, the multiple regression equation will be described. First,
The multiple regression equation is represented by the following equation (1). Y = A0 + A1X1 + A2X2 + ... AqXq ---
-(1) where Y is an objective variable, X1 to Xq are explanatory variables, and A0 to
Aq is a regression coefficient. Now, for example, assuming that the vibration is the objective variable Y, the rotational speed is the explanatory variable X1, the flow rate is the explanatory variable X2, the temperature is the explanatory variable X3, and the residual is Ei, the equation (1) becomes the following equation (2). Yi = A0 + A1 X1 i + A2 X2 i + A3 X3 i + E
i --- (2) The regression coefficients A0, A1, A2 and A3 are determined so that the sum of squares of the residual Ei is minimized. The sum of squares of the residual Ei is
It is expressed by the following equation (3).

[0021]

[Equation 1]

In order to minimize the residual Ei in the above equation (3), the following equation (4) must be satisfied.

[0023]

[Equation 2]

By the way, the equation (3) becomes the following equation (5).

[0025]

[Equation 3]

When the equation (4) is solved using the equation (5), the following equation (6) is obtained.

[0027]

[Equation 4]

The values of the regression coefficients A0, A1, A2 and A3 can be calculated by solving the above equation (6).
By the way, in the above equation (6), the variance of the product of two variables,
So the covariance appears. Here, the covariance is defined. The following expression (7) shows normal dispersion, and the following expression (8)
Indicates the covariance.

[0029]

[Equation 5]

Here, n is the number of data and Xb is the average value of X. In equation (8), the covariance is XiYi, and in equation (6), X1 i and X2 i, and X1 i and X are
3 i, X2 i and X3 i, Yi and X1, Yi and X2 i, Y
i and X3 i. After calculating these covariances, the regression coefficient is calculated.

In the above description, the supervisory signal itself was made to correspond to each variable, but it is also possible to perform multiple regression analysis with the power of the supervisory signal, exponential function, etc. as new explanatory variables.

Now, in step 6, when a list of objective variable names and explanatory variable names (including a monitoring signal name that changes rapidly and a monitoring signal name that changes slowly) is read, step 7
Proceed to and read the data corresponding to the list. Next, in step 8, all covariances of the read objective variable and explanatory variable are calculated. Then, in step 9, the regression coefficient is calculated according to the list. Next, in step 10, the partial correlation coefficient is calculated according to the list. However, the partial correlation coefficient is a correlation when the objective variable Y and the explanatory variables X1 to X2, X3, ... Xq are ignored in the equation (1). That is, it is a correlation coefficient of only Y and X1.

Subsequently, in step 11, it is determined whether or not all the calculated partial correlation coefficients are equal to or more than the set values. If the partial correlation coefficient is less than or equal to the set value, the process proceeds to step 12, and the explanatory variable corresponding to the partial correlation coefficient less than or equal to the set value is deleted from the list. That is, steps 11 and 12
In, the validity evaluation of the calculated regression coefficient is executed. This is because the reliability of the regression coefficient is low when the partial correlation coefficient is less than or equal to the set value. When the process of step 12 is completed, the process returns to step 9.

In step 11, if all the partial correlation coefficients are equal to or more than the set values, the process proceeds to step 13. And
In this step 13, the time variation of each regression coefficient is obtained, and the average value and fluctuation width of each regression coefficient are calculated. Next, in step 14, the allowable range of each regression coefficient is calculated from the preset fluctuation width coefficient, the average value, and the fluctuation width. Then, in step 15, it is judged whether or not there is a regression coefficient exceeding the allowable fluctuation width. If there is a regression coefficient that exceeds the allowable fluctuation width, the process proceeds to step 16, an alarm indicating that fact is issued by an appropriate display means, and the process proceeds to step 17. If there is no regression coefficient exceeding the allowable fluctuation width in step 15, the process proceeds to step 17 without passing through step 16.

Next, in step 17, the time change rate of the regression coefficient is calculated. Subsequently, in step 18, the allowable change rate of the regression coefficient is calculated from the fluctuation width of the regression coefficient and the preset allowable change rate coefficient. Then, in step 19, it is determined whether or not the rate of change of the regression coefficient exceeds the calculated allowable rate of change. If the rate of change of the regression coefficient exceeds the allowable rate of change, in step 20, an alarm is displayed to that effect. If the rate of change of the regression coefficient does not exceed the allowable rate of change in step 19, the process proceeds to step 21 and it is determined whether the monitoring process is completed. If the process is finished, the process is ended, and if not, the process returns to step 1 and the monitoring process is started again.

As described above, according to the device abnormality monitoring method of the embodiment of the present invention, a large number of monitoring signals are detected from a device whose normal value is difficult to be grasped, and these multiple monitoring signals are subjected to multiple regression. Anomalies are monitored by determining not only whether the regression coefficient obtained by analysis deviates from the prescribed range, but also whether the rate of change of the regression coefficient deviates from the prescribed rate of change. ing. Since it is not necessary to consider a physical model for the combination of the objective variable and the explanatory variable in the multiple regression analysis, the normal value of the regression coefficient can be easily grasped at the time of trial operation of the plant. Therefore, it is possible to realize an abnormality monitoring method capable of detecting with high accuracy an abnormal initial sign of a device whose normal value is difficult to grasp. Also,
According to the above-described embodiment, the monitoring signal that changes at high speed is subjected to component separation processing, averaging processing, and selection processing to extract only the components that are essential for monitoring, and the extracted component and the monitoring signal that changes at low speed. Is performing multiple regression analysis. Therefore,
Since unnecessary components are not used in the multiple regression analysis, it is possible to further improve the detection speed and detection accuracy of the initial abnormal sign of the device.

In the above embodiment, the allowable range and the allowable rate of change of the regression coefficient are calculated as a function of the operating state by using the multiple regression analysis at the time of test operation of the device, and are calculated according to the operating state of the device. An allowable range may be set. By doing so, the initial sign of abnormality of the device can be detected with higher accuracy. Further, depending on the applied plant or the like, when the validity evaluation of the calculated regression coefficient is executed, unnecessary components in the high speed change monitoring signal may be removed. In this case, the component separation process can be omitted. Further, it may be configured to judge whether or not there is an abnormality based on whether or not the calculated regression coefficient is within a predetermined normal range or only over a predetermined normal change rate. Good. Further, when the monitoring signal is only the rapid change monitoring signal, the regression coefficient is calculated from only the rapid change monitoring signals.

Next, an embodiment of the device abnormality monitoring apparatus according to the present invention will be described. In FIG. 2, the output signals from the measuring devices 121, 122, 123 and the like of the monitoring signals that change at high speed are supplied to the component separation processing device 200. And
In this component separation processing device 200, the measuring device 12
Of the components of the signals supplied from 1, 122, 123, etc., only the components necessary for monitoring are extracted. The extracted components are supplied to the averaging processing device 300, and the time variation of the above components is suppressed. The signal component whose time variation is suppressed is supplied to the correlation extraction processing device 400. The correlation extraction processing device 400 is also supplied with output signals from the measuring devices 131, 132, 133, etc., which measure the monitor signal changing at a low speed. In the correlation extraction processing device 400, multiple regression signals are subjected to multiple regression analysis, regression coefficients are calculated, and the validity of the calculated regression coefficients is determined. Then, the calculated regression coefficient is supplied to the abnormality determination device 500,
The initial abnormal sign of the device is discriminated from the regression coefficient itself and from the rate of change of the regression coefficient.

The component separation processing device 200, the averaging processing device 300, the correlation extraction processing device 400, and the abnormality determination device 500, which are the main parts in the example of FIG. 2, will be described below. The component separation processing device 200 includes a frequency component separation type processing device, a frequency band separation type processing device, and a correlation component separation type processing device.

FIG. 3 shows a component separation processing device 200.
It is a block diagram of the example using the frequency-component separation-type processing apparatus.
In FIG. 3, the component separation processing device 200 is a Fourier transform unit 210 for frequency analysis, an amplitude / phase component transform unit 211 for calculating amplitude and phase from a real number part and an imaginary number part obtained as a frequency analysis result, and useful for monitoring. The average processing apparatus 300 includes an output selection unit 212 that selects only the appropriate components.
Is composed of a low-pass filter unit 310 that averages the monitoring components. A monitoring signal that changes at high speed, that is, a high frequency signal such as vibration is input to the Fourier transform unit 210, and frequency analysis is performed. Then, the output signal from the Fourier transform unit 210 is converted into an amplitude and a phase by the amplitude / phase conversion unit 211. The phase indicates a delay from a certain time reference, and the rotation reference pulse of the recirculation pump 105 is used as the time reference here. That is, the input data to be frequency-analyzed is fetched in synchronization with the generation timing of the rotation reference pulse, so that the rotation reference pulse is automatically used as the time reference. And
The output signal from the amplitude / phase conversion unit 211 is supplied to the output selection unit 212, and a component useful for predetermined monitoring, that is, for example, in vibration, an integral multiple component of the rotation frequency, a fractional component, and a mechanical system. The natural frequency of is selected. The selected component is output to the low pass filter unit 310. It is also possible to omit the output selection unit 212 and configure the amplitude / phase conversion unit 211 to perform the selection processing. In that case, since the amplitude / phase conversion process only needs to be performed on the selected frequency component, the amount of signal processing is reduced. Here, in the selection process, it is set to select an integer multiple component of the rotation frequency, a component of fractional order, and the natural frequency of the rotating shaft. Low-pass filter section 31
The cutoff frequency of 0 is set with reference to the flow rate measuring device 132 and the temperature measuring device 133. It is based on the knowledge that in multiple regression analysis, the closer the response to a change in the state of the plant is, the smaller the residual error in the regression analysis is. With such a device configuration, it is possible to convert a high frequency signal into a low frequency signal and reduce information unnecessary for monitoring.
Then, the output signal of the low-pass filter unit 310 is supplied to the correlation extraction processing device 400.

FIG. 4 shows a component separation processing device 200.
Bank filter unit 22 which is a frequency band separation type processing device
It is a block diagram of the example using 0. This bank filter unit 2
20 is composed of a number of bandpass filters having different center frequencies. Further, the averaging processing device 300 includes an average response processing unit 320 that adds output signals of the bank filter unit 220 for each same angle in synchronization with rotation, and the bank filter unit 22.
RMS value calculation unit 321 for calculating the RMS value of the output signal of 0
And an output selection unit 322 that selects and outputs an output signal according to a predetermined algorithm. The above algorithm is, for example, an algorithm that outputs a predetermined number of signals from the larger one of the output signals of the average response processing unit 320. Bank filter unit 22
0 has various center frequencies, and even if the band with the best SN ratio in the monitor signal is unknown, the band with a good SN ratio can be selected by the output selection algorithm. Therefore, the optimum SN ratio can be dynamically monitored. As will be described later, when the rotation synchronization component is detected by the average response method, if the sound level at a certain angle is higher than other angles,
It can be judged that the rotation synchronization component has occurred, and the optimum frequency band can be dynamically selected.

FIG. 5 shows a component separation processing device 200.
It is a block diagram of the example at the time of using a correlation component separation type | mold processing apparatus. In this example, a large number of fast-changing supervisory signals are combined every two signals, and a frequency region having a high correlation and a frequency region having a low correlation between the combined two signals are determined based on the coherence analysis result. Then, the effective value of each band is output. In FIG. 5, the coherence analysis unit 230 quantifies the strength of the correlation between two signals of the input signal for each frequency, and the correlation band determination unit 231 follows a predetermined correlation strength determination range to determine a frequency band having a strong correlation. Select a frequency band that has no correlation with. The pass band selection unit 233 includes a band pass filter. The signal delay unit 232 includes a storage unit that delays the input signal by the time required for the coherence analysis, and supplies the input signal to the passband selection unit 233 after the passband is determined by the coherence analysis. The output signals of the pass band selection unit 233 are two signals that have passed the frequency band having a correlation and two signals that have passed the frequency band having no correlation. These signals are processed by the effective value calculation unit 331 as the averaging processing device 300.
Is converted to an effective value. By performing a regression analysis of correlated signals, a highly accurate regression coefficient can be obtained. In addition, since the non-correlated component excludes information of an event that is dominant in a strongly correlated signal, it is effective for monitoring events other than the strongly correlated event. The output signal of the RMS calculator 331 is
It is supplied to the correlation extraction processing device 400.

FIG. 6 is an operation flowchart of the correlation extraction processing device 400 using multiple regression analysis. In step 30 of FIG. 6, a list of predetermined objective variable names and explanatory variable names is read. Next, in step 31, the time series data of each variable of a predetermined time width is read according to the read list. Then, in step 32, the covariance of all combinations of variables is calculated,
In step 33, the regression coefficient is calculated. And
In step 34, a partial correlation coefficient is calculated based on the calculated covariance data. In step 35, it is judged whether or not all the partial correlation coefficients exceed a predetermined set value, and if there is less than the set value, step 36
Proceed to. In step 36, the explanatory variable corresponding to the partial correlation coefficient equal to or less than the set value is deleted, and the process returns to step 33.
The validity of the regression coefficient calculated by these steps 35 and 36 is evaluated. In step 35,
If all partial correlation coefficients are above the set value, step 3
7, the regression coefficient, the partial correlation coefficient, and the residual are output to the abnormality determination device 500. It is also possible to determine whether the partial correlation coefficient is equal to or larger than the set value, and output only the regression coefficient corresponding to the partial correlation coefficient equal to or larger than the set value. By evaluating this partial correlation coefficient, only the regression coefficient that has reached the required accuracy can be extracted. In addition, the validity evaluation method of the regression coefficient, there is also a method using a non-partial correlation coefficient, the present invention,
The method of performing the validity evaluation is not limited to the method of using the partial correlation coefficient. For example, F value (variance ratio), predicted sum of squares, A
It is also possible to use IC (Akaike's information amount standard), degree-of-freedom adjusted contribution rate, and the like.

FIG. 7 is an operation flowchart of the abnormality discriminating apparatus 500. In FIG. 7, the abnormality determination processing device 50
There are two main functions of 0, one is the storage of the normal regression coefficient at the time of test operation, and the other is the abnormality monitoring during normal operation. First, in step 40, the data of the current operating state of the monitored device, that is, the values of the main monitoring signals are read and set as the operating state. The name of this main monitoring signal is predetermined. For example, it is the number of revolutions of the recirculation pump, the branch flow rate to the system branching the flow rate, and the like. Next, in step 41, the current regression coefficient calculated by the correlation extraction processing device 400 is read. The operator can select either the normal value storage mode or the abnormality monitoring mode, and in step 42, determine what the set mode is. If the normal value storage mode is selected, the routine proceeds to step 43, where the normal value of the regression coefficient for each predetermined main operating state is stored in an appropriate storage means, and the routine returns to step 40. On the other hand, if the abnormality monitoring mode is selected, the routine proceeds to step 44, where the normal values of the regression coefficient corresponding to the operating state and the rate of change of the regression coefficient are read, and the routine proceeds to step 45. In step 45, it is determined whether or not the rate of change of the regression coefficient with time deviates from the normal rate of change. If it deviates, the process proceeds to step 48, and an alarm is displayed to that effect by an appropriate display means. Also, step 4
In step 5, if the time rate of change of the regression coefficient does not deviate from the normal rate of change, the process proceeds to step 46. Then, it is determined whether the regression coefficient is within the normal range. If it is not within the normal range, step 48 is proceeded to, and an alarm is displayed. On the other hand, if it is within the normal range, the process proceeds to step 47, and a display indicating that it is normal is displayed. In this way, the normal value of the regression coefficient is stored and the abnormality is detected using the regression coefficient. Since a regression equation that does not depend on the physical model is used to store the normal value, it is possible to grasp the normal value among a plurality of monitoring signals relatively easily. When the above embodiment is applied to a plant in which it is easy to predict the normal value of the plant, the predicted normal value may be stored.

FIG. 8 shows an example in which an abnormality of the bearing lubrication system is detected by an embodiment of the equipment monitoring apparatus of the present invention. However, the abnormality is detected by the above-described frequency component separation type processing. FIG. 8A shows a state of component separation by frequency analysis. FIG. 8B shows a state in which temporal variation is suppressed by averaging, Af1 is a change in frequency component, and T is a change in bearing temperature. Also, FIG. 8 (C)
Indicates the change in the regression coefficient a. Regarding the normal value of the regression coefficient itself, the value at the time of trial operation of the plant and equipment is treated as the normal value. Bearing temperature and vibration are within the normal range by independent monitoring. By monitoring the regression coefficient that combines them, it is possible to detect abnormalities that could not be detected in the past. By extracting the rotational frequency component that is particularly sensitive to axis bending from the vibration and averaging it,
The absolute value of the residual in the regression analysis is reduced by suppressing the steady fluctuation more, and the regression coefficient with higher accuracy is obtained. Further, by adding the ambient temperature as an element of the correlation analysis, a more accurate regression coefficient can be obtained.

FIG. 9 shows an example in which rubbing vibration is detected by an embodiment of the device monitoring apparatus of the present invention. However, the abnormality is detected by the above-described frequency band separation type processing. FIG. 9A shows the component separation state by the bank filter. FIG. 9B shows the extraction state of the protrusion degree by the average response method. Further, FIG. 9C shows the change of the regression coefficient a between the protrusion degree and the vibration level. The acoustic sensor signal attached to the recirculation pump is separated for each frequency band and averaged in synchronization with the rotation reference pulse.
A protrusion degree P and a rotation frequency component Af1 are obtained. Then, the data in the frequency range with a large protrusion is extracted, the time variation of the protrusion and the regression coefficient of the vibration amplitude are obtained, and the abnormality is detected. In this example, it is detected that the rotary shaft and the casing are in contact with each other in the vicinity of the bearing in synchronization with the rotation. From the correlation between vibration and sound generation, more reliable abnormality detection is possible. In other words, a large protrusion indicates that sound is generated in synchronization with the rotation, and a change in the vibration of the rotation frequency component means that some unbalance has occurred or the rotation axis is bent. It becomes information. However, the degree of protrusion also changes due to electrical noise synchronized with the rotation of the acoustic measurement system, and the vibration may also change due to changes in the fluid temperature and the characteristics of the bearing support system. However, the simultaneous fluctuation of the vibration component of the protrusion degree and the rotation frequency indicates that the rotating shaft comes into contact with the casing, and a part of the rotating shaft is locally heated and thermally deformed.

As can be expected from the above two examples, the regression coefficient may not be obtained in a normal operating state because the correlation between a plurality of monitoring signals is small. Only when an abnormality occurs does the relationship emerge. For example, this is the case when a part of the rotary shaft of the pump comes into contact with the casing, vibration increases, and metal powder melts into the fluid to lower the electrical conductivity. In such a case, the abnormality monitoring apparatus of the present invention can detect an abnormality as a change in the regression coefficient. In the abnormality monitoring device of one embodiment of the present invention, the process executed by each device such as the component separation processing device 200 may be executed by a microcomputer. In addition, when the number of signals to be taken in is large and real-time processing is difficult, a DSP (digital signal processor) can be used.

As described above, according to the abnormality monitoring apparatus of the embodiment of the present invention, a large number of monitoring signals are detected from a device whose normal value is difficult to grasp, and the correlation extraction processing apparatus 400 detects a large number of monitoring signals. Not only whether the calculated regression coefficient deviates from the specified range by multiple regression analysis, but also whether the rate of change of the regression coefficient deviates from the specified change rate is abnormal. Judgment is made by the discriminating device 500 to monitor the abnormality. Since it is not necessary to consider a physical model for the combination of the objective variable and the explanatory variable in the multiple regression analysis, the normal value of the regression coefficient can be easily grasped at the time of trial operation of the plant. Therefore, it is possible to realize an abnormality monitoring device capable of detecting with high accuracy an abnormal initial sign of a device whose normal value is difficult to grasp. Further, according to one embodiment of the present invention, when calculating the regression coefficient, the important components of the fast-changing supervisory signal are separated and averaged, and the frequency is increased to the same frequency band as the slow-varying supervisory signal. Bandwidth is being reduced. With this, it is only necessary to calculate the regression coefficients of a large number of monitoring signals in the same frequency band without calculating unnecessary components for monitoring.
The calculation accuracy and the calculation speed can be improved, and the detection accuracy of the abnormal initial sign of the device can be further improved. In addition, since only the components necessary for monitoring are used in the calculation of the regression coefficient and the amount of data to be processed is minimized, an abnormality monitoring device can be used without using a large data storage device. realizable.

Depending on the plant to be applied and the like, it is conceivable that unnecessary components of the high speed change monitoring signal may be removed by executing the validity evaluation of the calculated regression coefficient. In this case, the component separation processing device 200 can be omitted. Further, the abnormality determination device 500 determines whether or not the calculated regression coefficient is abnormal only based on whether or not the calculated regression coefficient is within a predetermined normal range or exceeds a predetermined normal change rate. It may be configured as follows. Further, when the monitoring signal is only the rapid change monitoring signal, the regression coefficient is calculated from only the rapid change monitoring signals.

FIG. 10 is a schematic configuration diagram of another embodiment of the device abnormality monitoring apparatus according to the present invention. Note that this example
This is an example when applied to the abnormality monitoring of the turbine / generator system of a nuclear power plant. In FIG. 10, high-temperature, high-pressure steam is guided to a turbine 110 through a main steam pipe 103, the turbine 110 is rotated, and power is generated by a generator 111 directly connected to the turbine 110. The steam emitted from the turbine 110 passes through the condenser 112 to become water, which is then sent to the water supply system. Measured quantities for abnormality monitoring include generator output, vibration, rotation speed, bearing temperature and other lubrication system instrumentation, vacuum degree of condenser 112, main steam flow rate, and the like. These measured amounts are supplied to the component separation processing device 200 through the rotation speed measuring device 140, the vibration measuring device 141, the vapor flow measuring device 150, the vacuum degree measuring device 151, and the like. The component separation processing device 200, the averaging processing device 300, and the correlation extraction processing device 400 are the same as those in the example of FIG. A feature of this embodiment is an abnormality determination device 550. This abnormality determination device 550 has a function of performing comprehensive abnormality determination by using the value of the regression coefficient, the monitoring signal level, and the level of the signal that has been component-separated and averaged.

The abnormality determination by the abnormality determination device 550 will be described below. For example, the output of the generator can be increased by increasing the flow rate of steam in the main steam pipe 103. Therefore, in monitoring using only the regression coefficient, if the relationship between the steam flow rate and the generator output is constant,
Even if the steam flow rate exceeds the rated value, it may not be judged as abnormal. Therefore, by providing a monitoring system that monitors the steam flow rate itself, it is possible to prevent the steam flow rate from exceeding the rated value. Further, since the change range of the signal that has been component-separated and averaged is narrower than the change range of the overall level, it is possible to narrow the range of the normal value of the signal and improve the abnormality detection accuracy. That is, for example, in an oil whirl or oil whip that occurs in a steam turbine or a generator, a vibration abnormality can be detected early by setting the normal value range of the natural frequency component of vibration to be sufficiently narrow. Further, in the case of so-called self-excited vibration, the influence of the exciting force is unlikely to appear in other monitoring signals. Therefore, it is possible to accurately detect the abnormality by using the level of the monitoring signal itself. in this way,
By performing abnormality detection using three types of signals, that is, the value of the regression coefficient, the level of the monitoring signal, and the level of the signal that has been component-separated and averaged, it is possible to improve the detection accuracy of the initial sign of abnormality of the device. . According to the example of FIG. 10, in addition to the effect similar to that of the example of FIG. 2, the abnormality detection is performed by using the three types of signals as described above, and thus the detection accuracy of the abnormality initial sign of the device is high. Can be further improved.

FIG. 11 is a schematic configuration diagram of still another embodiment of the device abnormality monitoring apparatus according to the present invention. It should be noted that this example is an example applied to the abnormal sound monitoring of the steam generator of the fast breeder reactor. In FIG. 11, as a typical abnormal sound, there is a sound caused by water leaking into sodium and resulting sodium water reaction. The high-temperature sodium passes through the sodium inlet pipe 701, the periphery of the heat transfer pipe 707, and returns from the sodium outlet pipe 702 to the pump side. Inside the container body 705, a heat transfer tube 707 through which the water supply supplied from the water supply inlet pipe 703 passes is arranged,
The heat transfer tube 707 is supported by the heat transfer tube support structure 706. An acoustic sensor 710 is arranged on the outer wall of the container body 705. In addition, sodium outlet temperature measuring device 1
61, a steam temperature measuring device 162 that measures the steam temperature of the steam outlet pipe 704, an acoustic measuring device 163, a feed water flow rate measuring device 164, a sodium flow rate measuring device 165, and a hydrogen in sodium measuring device 166 are installed. The output signal of the acoustic measurement device 163 passes through the component separation processing device 240, the averaging processing device 350, and is supplied to the correlation extraction processing device 410. The output signals of other measuring devices are
It is directly supplied to the correlation extraction device 410. The output signal of the correlation extraction processing device 410 is supplied to the abnormality determination device 510, and the abnormality determination device 510 determines the abnormality from the value of the regression coefficient and its change rate.

The correlation extraction device 410 calculates the regression coefficient by the following autoregressive equation (9). Yi = A0 + A1 X1 + A2 X2 + A3 X3 + A4 X4
+ A5 X5 --- (9) where Yi is the effective value of the i-th bank filter output of the acoustic signal, X1 is the sodium outlet temperature, X2 is the steam temperature, X3 is the feed water flow rate, X4 is the sodium flow rate, and X5 is the inside of sodium. Hydrogen amount, A0, A1, A2, A3, A4, A
5 is a regression coefficient corresponding to each monitoring signal.

The acoustic signal changes due to the sodium water reaction and the flow of steam and sodium. Further, the amount of hydrogen in sodium also increases in the sodium-water reaction in sodium.
Therefore, if there is a correlation between the acoustic level and the amount of hydrogen in sodium, it is considered that the sodium water reaction has occurred. Therefore, in the abnormality determination device 510, by monitoring the value of the regression coefficient A5, the rate of change, and the partial correlation coefficient, high-sensitivity abnormality detection can be performed only by acoustic monitoring. High-sensitivity anomaly detection is impossible only by monitoring the hydrogen amount alone. The frequency band of the acoustic signal is selected by the bank filter in the component separation processing device 240, and the abnormality determination is executed by the bank filter output having the largest partial correlation coefficient to further improve the sensitivity.

As described above, according to the example of FIG.
In addition to the effect equivalent to that of the example shown in (1), the following effect can be obtained. That is, it becomes possible to monitor only the relationship between the hydrogen concentration in sodium and the change in the acoustic level, the detection accuracy of the sodium water reaction detection sensitivity can be improved, and the abnormality initial sign detection accuracy of the device abnormality monitoring device can be improved. Moreover, since the regression coefficient of the component having the largest correlation coefficient among the components of the acoustic signal is adopted as the representative value, the abnormality initial sign detection accuracy of the device abnormality monitoring device can be further improved.

FIG. 12 is a schematic configuration diagram of still another embodiment of the device abnormality monitoring apparatus according to the present invention, in which components equivalent to those in FIG. 2 are designated by the same reference numerals. Then, in the example of FIG. 12, the abnormality determination device 500 of the example of FIG.
Instead of this, a learning-type abnormality determination device 1000 is provided. The configuration other than this learning-type abnormality determination device 1000 is the same as the example of FIG.

In FIG. 12, the learning type discriminating apparatus 1000 is shown.
Is composed of a signal distributor (signal separator) 1010, a monitoring parameter pattern selector 1220, a monitoring parameter learning device 1230, a monitoring mode setting device 1250, an abnormality detector 1260, and an alarm display device 1280. ing. The signal distributor 1010 includes a correlation extraction processing device 400,
Of the signals supplied from the vibration measurement device 121, the acoustic measurement device 122, the rotation speed measurement device 123, etc., the main monitoring parameter (for example, the rotation speed signal) and the dependent monitoring parameters (regression coefficient, vibration signal, flow rate signal, bearing temperature). Signal, etc.) and. The monitoring mode setting device 1250 outputs a signal indicating the learning mode or the monitoring mode according to the operation of the operator through the operation panel or the like. The main monitoring parameter from the signal distributor 1010 is supplied to the monitoring parameter pattern selector 1220, and the main monitoring parameter value is quantized (for example, 1175 rpm is rounded to 1170 rpm). Then, the quantized value number is output from the monitoring parameter pattern selector 1220.

The supervisory parameter learning device 1230 is supplied with the dependent supervisory parameters from the signal distributor 1010.
Then, when the signal supplied from the monitoring mode setting unit 1250 indicates the learning mode, the monitoring parameter learning unit 1230 normally changes the dependent monitoring parameters according to the quantized value number supplied from the pattern selector 1220. Learn width data. In addition, when the signal supplied from the monitoring mode setting unit 1250 indicates the monitoring mode, the monitoring parameter learning unit 1230 may use the pattern selector 12
According to the quantized value number supplied from 20, the learned representative value of the dependent monitoring parameter is supplied to the abnormality detector 1260.

The anomaly detector 1260 is also supplied from the signal distributor 1010 with actual slave monitoring parameters at the present time. The abnormality detector 1260 operates when the signal from the mode setter 1250 indicates the monitoring mode,
The learned representative value of the dependent monitoring parameter is compared with the current actual dependent monitoring parameter. As a result of comparison, if the actual dependent monitoring parameter at the present time is outside the normal range,
It is judged that there is an abnormality, and a signal indicating the judgment result is displayed on the alarm indicator 128.
Supply to 0. The alarm indicator 1280 is the abnormality detector 12
According to the determination result of 60, a predetermined warning is displayed.

FIG. 13 shows a monitoring parameter learning device 1230.
It is a block diagram of. In FIG. 13, the learning device 1230 is
A storage unit 1231 for storing the learning data, a data switching unit 1232 for changing the output destination of the learning data according to the monitoring mode, and an accumulator 1233 for updating the representative value of the dependent monitoring parameter are provided. There is. Note that, here, the learning is an operation for determining the variation range of the dependent monitoring parameter under normal conditions. FIG. 14 shows the data flow inside the learning device 1230 in the monitoring mode. However, i is the main monitoring parameter quantized value number, j is the dependent monitoring parameter number, N (i) is the number of data, S
(I, j) is the monitoring parameter simple sum, and S2 (i, j) is the monitoring parameter square sum. XB (i, j) is the average value of the monitoring parameters, σ (i, j) is the standard deviation, Xma
x (i, j) is the maximum value of the monitoring parameter, and Xmin (i, j
j) is the minimum value of the monitoring parameter, k is the input order of the monitoring parameter, and XjG is the current value of the dependent monitoring parameter.

In FIG. 14, the memory 1231 stores, in accordance with the quantization number, the learned data number N (i), which is the representative value of the dependent monitoring parameters, and the monitoring parameter simple sum S.
(I, j) sum of squares S2 (i, j), average value XB (i, j)
j), standard deviation σ (i, j), maximum value Xmax (i,
j) and the minimum value Xmin (i, j) are stored. Every time a certain period of time elapses, the dependent parameter representative value storage area of the storage unit 1231 is selected according to the quantized value number i, and the storage content of the selected storage area is supplied to the accumulator 1233. Accumulator 1
At 233, a new representative value is calculated from the representative value supplied from the storage unit 1231 and the current value XjG of the parameter. Then, the contents of the storage area corresponding to the quantized value number i are updated. By repeating the above-described operation in the learning mode, the range of the dependent monitoring parameter at the time of normal operation of the device is stored in the storage unit 1231 for each value of the main monitoring parameter.

FIG. 15 is an operation flowchart of the data processing of the learning-type abnormality determination device 1000. In step 50 of FIG. 15, it is determined whether or not it is the learning or monitoring time. If it is the learning or monitoring time, in step 51, the main monitoring parameter is fetched and AD conversion is performed. Then, in step 52, the main monitoring parameter is quantized, and in step 53, the contents of the storage area corresponding to the quantized value number i are read. Then, in step 54, the value of the dependent monitoring parameter at the current time is AD-converted. Next, in step 55, it is determined whether the currently set mode is the learning mode or the monitoring mode. If it is the learning mode, the routine proceeds to step 56, where after the representative value update calculation is executed, the contents of the storage area are updated, and the routine returns to step 50.

If the monitor mode is set in step 55, the process proceeds to step 57. This step 57
In the following, using the predetermined allowable width coefficient a, the following equation (1
0) and (11), the maximum allowable value Max (j) and the minimum allowable value Mi for each dependent monitoring parameter are learned from the learning data.
Calculate n (j). Max (j) = XB (i, j) + a · σ (i, j)
--- (10) Min (j) = XB (i, j) -a.sigma. (I, j)
--- (11) Here, j is from 1 to the number J of dependent monitoring parameters. Next, in step 58, it is determined whether the dependent parameter value at the current time is within the allowable range, that is, between the maximum allowable value Max (j) and the minimum allowable value Min (j) calculated in step 57. To judge. If even one of the dependent monitoring parameters at the current time is outside the allowable range, step 6
Going to 0, the abnormal alarm display processing is performed. Also, step 5
In 8, if all the dependent monitoring parameters at the current time are within the allowable range, the process proceeds to step 59 to display that the device is normal. Then, the process returns to step 50.

In the example shown in FIG. 12, the normal range of the monitoring parameters is learned in the learning mode during the operation confirmation test after the end of the periodic inspection. Then, during normal operation, the abnormality of the device is monitored in the monitoring mode. Among the monitoring parameters, for example, the conductivity and the like may be different between the value immediately after the periodic inspection and the value during normal operation. Therefore,
The normal range may be partially modified in the learning mode. In addition, the monitoring parameter may be newly or added during the operation of the device so that learning can be performed.

FIG. 16 shows an example of abnormality detection in the example of FIG. 12, in which the main monitoring parameter is the number of revolutions, and the subordinate monitoring parameters are regression coefficient, vibration, flow rate, bearing temperature, conductivity and the like. Here, the main monitoring parameter is set to the number of revolutions because the number of revolutions in the recirculation pump loop is likely to control the fluctuation of various measured amounts. Also,
In FIG. 16, α is the normal range set by the conventional monitoring method, and β is the normal range set by the above embodiment. The circle γ is the value of the monitoring parameter at the current time. Furthermore, the monitoring parameter value is shown in percentage with reference to the maximum value of each monitoring parameter detected in advance. As is apparent from FIG. 16, the normal range according to the above embodiment is narrower than the conventional normal range. Thereby, the abnormality of the device can be detected with high accuracy. The vibration and conductivity regression coefficients are not significant when the equipment is normal. However, in FIG. 16, the regression coefficient of vibration and conductivity has a significant value, and in this case, it is determined that the device is abnormal.

In FIG. 16, the monitoring parameter γ is within the conventional normal range α. Therefore, it can be understood that in the past, such a slight device abnormality could not be detected. In the above-mentioned embodiment, if even one value of the monitoring parameter is out of the normal range, it is judged that an abnormality has occurred in the device, so that an unexpected abnormality can be detected.

In the above learning, when a plurality of main monitoring parameters, for example, the rotational speed and the reactor pressure are set, the main monitoring is also performed on the regression coefficient that changes nonlinearly with the rotational speed and the reactor pressure. It becomes possible to learn for each parameter, which is particularly effective for monitoring plant equipment with high non-linearity.

As described above, according to the example of FIG.
In addition to the effects similar to those of the above example, the following effects can be obtained. Since the learning method adopts the one that can be updated momentarily such as the average value as the representative value of the monitoring parameter, the storage device for storing the representative value has a small capacity and the hardware is small. It is possible to realize a device abnormality monitoring device with improved detection sensitivity.

Further, since the learning result is set to the representative value of the monitoring parameters, that is, the average value or the standard deviation, it is easy to evaluate the learning result and it is easy to judge whether or not the correction is necessary. Therefore, it is possible to realize a device abnormality monitoring device which is flexible and has improved operation performance. Furthermore, since the normal range of the monitoring parameters during operation is automatically learned, it is not necessary to carry out the study work for setting the normal range and the work necessary for test operation, which improves the economic efficiency of equipment abnormalities. A monitoring device can be realized.

Since the normal range including the relationship between the monitoring parameters, that is, the regression coefficient is grasped, it is possible to realize the device abnormality monitoring apparatus with further improved abnormality detection sensitivity. Further, when the abnormality is detected, the allowable range for determining the normal range is set to be proportional to the standard deviation of the monitoring parameter, so that the allowable range of the monitoring range is automatically set according to the magnitude of the normal fluctuation. Therefore, the abnormality detection sensitivity can be adjusted only by changing the allowable width coefficient a, and the device abnormality monitoring apparatus with further improved operation performance can be realized. Also, for example, a regression coefficient that changes non-linearly with respect to rotational speed and reactor pressure can be learned for each main monitoring parameter, and it can be applied to plants with high non-linearity, realizing a device abnormality monitoring device with a wide range of application. it can.

In the example of FIG. 12 described above, the average value is used as one of the representative values of the dependent monitoring parameters, but the median value, the maximum value, or the minimum value may be used instead of the average value. . Further, in the abnormality detection, when the difference in the pattern distance between the observed monitoring parameter pattern and the learning parameter pattern exceeds the set range, it may be determined to be abnormal.

Further, the learning type abnormality discriminating apparatus 1000 may be configured as follows. That is, the abnormality determination device 1000 includes a mode setting device for selecting and setting the learning mode and the monitoring mode, a fluctuation range data learning device,
And an abnormality detector. When the learning mode is set by the mode setter, the fluctuation range data learning device takes in the calculated regression coefficient and a large number of monitoring signals, and relates to these regression coefficients and the normal fluctuation range of the large number of monitoring signals. The fluctuation range data is calculated and stored. Furthermore, the fluctuation range data learning device outputs the stored fluctuation range data when the monitoring mode is set. When the monitoring mode is set, the anomaly detector calculates the regression coefficient and the normal range of many monitoring signals based on the fluctuation range data, and the actual regression coefficient and many monitoring signals are within the normal range. Or not. When it is not within the normal range, it is determined that an abnormality has occurred in the device.

Although the above-mentioned example is an example applied to a nuclear power plant, the present invention is not limited to the nuclear power plant and can be applied to an abnormality monitoring method and apparatus for equipment such as other plants.

[0074]

As described above, according to the device abnormality monitoring method of the present invention, a large number of monitoring signals are detected from a device whose normal value is difficult to be grasped, and these multiple monitoring signals are subjected to multiple regression analysis. The validity of the obtained regression coefficient is evaluated, and the regression coefficient useful for monitoring is extracted. Then, the abnormality is monitored by determining whether the extracted regression coefficient deviates from a predetermined range or whether the rate of change of the regression coefficient deviates from a predetermined rate of change. Therefore, it is possible to realize an abnormality monitoring method capable of detecting with high accuracy an abnormal initial sign of a device whose normal value is difficult to grasp. Also, of the monitoring signals that change at high speed, only the components essential for monitoring are extracted,
Multiple regression analysis is performed using the extracted components and the monitoring signals other than the high-speed characteristic monitoring signal. Therefore, the detection speed and detection accuracy of the abnormal initial sign of the device can be further improved.

Further, according to the device abnormality monitoring apparatus of the present invention, a large number of monitoring signals are detected from a device whose normal value is difficult to be grasped, and the correlation extracting means performs multiple regression analysis on the large number of monitoring signals to perform regression. The coefficient is calculated, the validity of the calculated regression coefficient is evaluated, and the regression coefficient useful for monitoring is extracted. And
Whether the extracted regression coefficient deviates from the predetermined range,
Alternatively, whether or not the rate of change of the regression coefficient deviates from a predetermined rate of change is determined by the abnormality determining means, and the abnormality is monitored. Therefore, it is possible to realize an abnormality monitoring device capable of detecting with high accuracy an abnormal initial sign of a device whose normal value is difficult to grasp.

Further, according to the abnormality monitoring device of the present invention,
The component separation means separates and extracts a significant component of the monitor signal that changes at high speed. As a result, when calculating the regression coefficient, components unnecessary for monitoring are not calculated, so that the calculation speed and calculation accuracy can be improved, and the detection accuracy of the abnormal initial sign of the device can be further improved. In addition, since only the components necessary for monitoring are used in the calculation of the regression coefficient and the amount of data to be processed is minimized, an abnormality monitoring device can be used without using a large data storage device. realizable.

[Brief description of drawings]

FIG. 1 is an operation flowchart of an embodiment of a device abnormality monitoring method according to the present invention.

FIG. 2 is a schematic configuration diagram of an embodiment of an apparatus abnormality monitoring device according to the present invention.

FIG. 3 is a block diagram of an example of a component separation processing device and an averaging processing device of the example of FIG.

FIG. 4 is a block diagram of another example of the component separation processing device and the averaging processing device of the example of FIG.

5 is a block diagram of still another example of the component separation processing device and the averaging processing device of the example of FIG. 2. FIG.

FIG. 6 is an operation flowchart of the correlation extraction processing device in the example of FIG.

7 is an operation flowchart of the abnormality determination device in the example of FIG.

FIG. 8 is an example of device abnormality detection by the apparatus of FIG. 2;

FIG. 9 is another example of device abnormality detection by the apparatus of FIG. 2;

FIG. 10 is a schematic configuration diagram of another embodiment of the device abnormality monitoring apparatus according to the present invention.

FIG. 11 is a schematic configuration diagram of still another embodiment of the device abnormality monitoring device according to the present invention.

FIG. 12 is a schematic configuration diagram of still another embodiment of the device abnormality monitoring apparatus according to the present invention.

13 is a configuration diagram of a monitoring parameter learning device in the apparatus of the example of FIG.

FIG. 14 is a diagram showing a flow of internal data of a monitoring parameter learning device in a monitoring mode.

FIG. 15 is an operation flowchart of data processing of the learning-type abnormality determination device.

16 is a diagram showing an example of abnormality detection when the main monitoring parameter is the number of revolutions in the example of FIG.

[Explanation of symbols]

121, 141 Vibration measuring device 122, 163 Acoustic measuring device 123, 140 Rotation speed measuring device 131 Electric conductivity measuring device 132, 164 Flow rate measuring device 133 Bearing temperature measuring device 150 Steam flow measuring device 151 Vacuum degree measuring device 161 Sodium outlet temperature measurement Apparatus 162 Steam temperature measuring apparatus 165 Sodium flow rate measuring apparatus 166 Hydrogen in sodium measuring apparatus 200, 240 Component separation type processing apparatus 210 Fourier transforming section 211 Amplitude / phase converting section 212 Output selecting section 220 Bank filter section 230 Coherence analyzing section 231 Correlation band Discrimination unit 232 Signal delay unit 233 Pass band selection unit 300, 350 Averaging processing device 310 Low pass filter unit 320 Average response processing unit 321, 331 Effective value calculation unit 322 Output selection unit 400, 410 Correlation extraction Processing device 500, 510, 550 Abnormality discriminating device 1000 Learning type anomaly discriminating device 1010 Signal distributor 1220 Monitoring parameter pattern selector 1230 Monitoring parameter learning device 1231 Storage device 1232 Data switching device 1233 Accumulator 1250 Monitoring mode setting device 1260 Abnormality detector 1280 alarm indicator

Continuation of front page (51) Int.Cl. ⁵ Identification number Reference number within the agency FI technical display location // G05B 23/02 302 N 7208-3H (72) Inventor Mitsuharu Okabe 3-1, Saiwaicho, Hitachi, Ibaraki No. 1 Hitachi Ltd., Hitachi Plant (72) Inventor Kunio Akutagawa 3-1-1, Saiwaicho, Hitachi City, Ibaraki Hitachi Ltd., Hitachi Plant

Claims (30)

[Claims] 1. A step of detecting a large number of characteristics of a device to be monitored or a device to be monitored and its peripheral equipment, and taking in the detected large number of characteristics as a large number of monitoring signals, and performing a multiple regression analysis on the large number of acquired monitoring signals Then, the step of calculating the regression coefficient, the step of executing the validity evaluation of the calculated regression coefficient, and the step of judging whether the calculated regression coefficient is within the predetermined normal range or not are judged within the predetermined normal range. If not, a step of determining that an abnormality has occurred in the device, and a method for monitoring the abnormality of the device. 2. The apparatus abnormality monitoring method according to claim 1, wherein a time change rate of the regression coefficient is calculated, and it is determined whether or not the calculated time change rate exceeds a predetermined normal change rate. The abnormality monitoring method for a device, further comprising the step of determining that an abnormality has occurred in the device when the normal change rate is exceeded. 3. The device abnormality monitoring method according to claim 1, wherein the predetermined normal range of the regression coefficient is calculated as a function of the operating state by multiple regression analysis at the time of a trial operation of the device, and the operating state of the device. An abnormality monitoring method for a device, wherein a normal range is set according to. 4. A step of detecting a large number of characteristics of the monitored device or the monitored device and its peripheral equipment, and capturing the detected large number of characteristics as a large number of monitoring signals, and a multiple regression analysis of the large number of captured monitoring signals. Then, the step of calculating the regression coefficient, the step of executing the validity evaluation of the calculated regression coefficient, the step of calculating the time change rate of the regression coefficient, and whether the calculated time change rate exceeds the predetermined normal change rate And a step of determining that an abnormality has occurred in the device when the rate of change exceeds a predetermined normal change rate. 5. A step of detecting a large number of characteristics of the monitored device or the monitored device and its peripheral equipment, and capturing the detected large number of characteristics as a large number of monitoring signals; Multiple regression is performed on the step of separating and extracting only the components necessary for monitoring from the high-speed characteristic monitoring signal indicating the changing characteristics, the monitoring signals other than the high-speed characteristic monitoring signal, and the components necessary for monitoring the high-speed characteristic monitoring signal. The step of calculating the regression coefficient by performing multiple regression analysis on the components necessary for the analysis or the monitoring of the high-speed characteristic monitoring signal, the step of executing the validity evaluation of the calculated regression coefficient, and the calculated regression coefficient Is determined to be within a predetermined normal range, and if it is not within the predetermined normal range, it is determined that an abnormality has occurred in the device. Vessel abnormality monitoring method. 6. The apparatus abnormality monitoring method according to claim 5, wherein a time change rate of the regression coefficient is calculated, and it is determined whether or not the calculated time change rate exceeds a predetermined normal change rate. The abnormality monitoring method for a device, further comprising the step of determining that an abnormality has occurred in the device when the normal change rate is exceeded. 7. The device abnormality monitoring method according to claim 6, wherein the predetermined normal range of the regression coefficient is calculated as a function of the operating state by multiple regression analysis at the time of test operation of the device, and the operating state of the device. An abnormality monitoring method for a device, wherein a normal range is set according to. 8. The device abnormality monitoring method according to claim 5 or 6, wherein after extracting only the components necessary for monitoring from the high-speed characteristic monitoring signal, the extracted components are averaged, and this averaging is performed. An abnormality monitoring method for a device, wherein the processed component is used to calculate a regression coefficient. 9. A step of detecting a large number of characteristics of the monitored device or the monitored device and its peripheral equipment, and capturing the detected large number of characteristics as a large number of monitoring signals; Multiple regression is performed on the step of separating and extracting only the components necessary for monitoring from the high-speed characteristic monitoring signal indicating the changing characteristics, the monitoring signals other than the high-speed characteristic monitoring signal, and the components necessary for monitoring the high-speed characteristic monitoring signal. Analyze or perform multiple regression analysis of components necessary for monitoring the high-speed characteristic monitoring signal to calculate regression coefficient, execute validity evaluation of the calculated regression coefficient, and time of regression coefficient The rate of change is calculated to determine whether the calculated rate of change over time exceeds a predetermined normal rate of change. If the rate of change exceeds a predetermined normal rate of change, it is determined that an abnormality has occurred in the device. Abnormality monitoring method of the device, characterized in that it comprises the steps of disconnection, the. 10. The device abnormality monitoring method according to claim 9, wherein after extracting only the component necessary for monitoring from the high-speed characteristic monitoring signal, the extracted component is averaged, and this averaging process is performed. The abnormality monitoring method for equipment, wherein the component is used to calculate the regression coefficient. 11. A detection means for detecting a large number of characteristics of a monitored device or a monitored device and its peripheral equipment as a large number of monitoring signals, and a multiple regression analysis of the detected large number of monitoring signals to calculate a regression coefficient. At the same time, the correlation extraction means that executes the validity evaluation of the calculated regression coefficient, and whether or not the calculated regression coefficient is within the predetermined normal range. An abnormality monitoring device for equipment, comprising: an abnormality determining means for determining that an error has occurred. 12. The apparatus for monitoring abnormality of a device according to claim 11, wherein the abnormality determining means also calculates a time change rate of the regression coefficient and determines whether the calculated time change rate exceeds a predetermined normal change rate. An abnormality monitoring device for an apparatus, characterized by determining that an abnormality has occurred in the apparatus even when the rate of change exceeds a predetermined normal change rate. 13. The device abnormality monitoring device according to claim 11, wherein the predetermined normal range of the regression coefficient is calculated as a function of the operating state by multiple regression analysis at the time of test operation of the device, and is stored by the storage means. The device abnormality monitoring device is characterized in that a predetermined normal range of the stored regression coefficient is set as a normal range according to an operating state of the device. 14. The abnormality monitoring device for an apparatus according to claim 11, wherein the abnormality determining means sets the learning mode by a mode setter for selecting and setting a learning mode and a monitoring mode. When it is, the calculated regression coefficient and a large number of monitoring signals are taken in, and the fluctuation range data regarding the normal fluctuation range of these regression coefficients and a large number of monitoring signals is calculated and stored, and when the monitoring mode is set, When the monitoring mode is set by the fluctuation range data learning device that outputs the stored fluctuation range data and the mode setting device, the regression coefficient and the normal range of many monitoring signals are calculated based on the fluctuation range data. Then, it is judged whether the actual regression coefficient and many monitoring signals are within the above normal range, and if they are not within the normal range, it is judged that an abnormality has occurred in the equipment. An abnormality detector for equipment, comprising: an abnormality detector. 15. The abnormality monitoring device for equipment according to claim 11, wherein the abnormality determining means is supplied with the calculated regression coefficient and a large number of monitoring signals, and the main monitoring parameter and the dependent monitoring parameter are supplied from these multiple monitoring signals. When the learning mode is set by the mode setting device for selecting and setting the signal separation means for separating and outputting, and the learning mode and the monitoring mode, a plurality of main monitoring parameters are set. Calculates and stores fluctuation range data relating to the normal fluctuation range of each dependent monitoring parameter corresponding to each quantized value, and when the monitoring mode is set, corresponds to the actual quantized value of the main monitoring parameter. When the monitoring mode is set by the monitoring parameter learning device that outputs the fluctuation range data of the stored dependent monitoring parameters and the mode setting device Calculates the normal range of the dependent monitoring parameter based on the fluctuation range data and determines whether the actual dependent monitoring parameter is within the normal range.If it is not within the normal range, the device is abnormal. An abnormality monitoring device for equipment, comprising: an abnormality detector that determines that an abnormality has occurred. 16. The device abnormality monitoring device according to claim 15, wherein the fluctuation range data is an average value and a standard deviation of each dependent monitoring parameter. 17. A detection means for detecting a large number of characteristics of a monitored device or a monitored device and its peripheral equipment as a large number of monitoring signals, and a multiple regression analysis of the detected large number of monitoring signals to calculate a regression coefficient. At the same time, the correlation extraction means for executing the validity evaluation of the calculated regression coefficient and the time change rate of the regression coefficient are calculated, and it is judged whether the calculated time change rate exceeds a predetermined normal change rate, An abnormality monitoring device for equipment, comprising: abnormality determination means for determining that an abnormality has occurred in the equipment when the normal change rate of is exceeded. 18. A detection means for detecting a large number of characteristics of a monitored device or a monitored device and its peripheral equipment as a large number of monitoring signals, and a high-speed characteristic of the detected large number of monitoring signals, which shows a rapidly changing characteristic. Multiple regression analysis is performed on the component separation means that separates and extracts only the components necessary for monitoring from the monitoring signal, the output signal from the component separation means, and the monitoring signals other than the high-speed characteristic monitoring signal, or the component separation is performed. Only the output signal from the means is subjected to multiple regression analysis to calculate the regression coefficient, and the correlation extraction means for executing the validity evaluation of the calculated regression coefficient and the calculated regression coefficient are within the predetermined normal range. An abnormality monitoring device for an apparatus, comprising: an abnormality determining unit that determines whether or not an abnormality has occurred in the apparatus when it is not within a predetermined normal range. 19. The apparatus for monitoring abnormality of a device according to claim 18, wherein the abnormality determining means also calculates a time change rate of the regression coefficient and determines whether the calculated time change rate exceeds a predetermined normal change rate. An abnormality monitoring device for an apparatus, characterized by determining that an abnormality has occurred in the apparatus even when the rate of change exceeds a predetermined normal change rate. 20. The device abnormality monitoring device according to claim 18, wherein the predetermined normal range of the regression coefficient is calculated as a function of the operating state by multiple regression analysis at the time of test operation of the device, and is stored by the storage means. The device abnormality monitoring device is characterized in that a predetermined normal range of the stored regression coefficient is set as a normal range according to an operating state of the device. 21. The apparatus abnormality detection device according to claim 20, wherein the component separating means has an averaging processing unit for averaging the signals resulting from the component separation, and the signals averaged by the averaging unit are: Supply to the correlation extraction means and the abnormality determination means,
An abnormality monitoring device for equipment, wherein the abnormality determining means is configured to determine the operating state of the equipment from an output signal from the averaging unit and a monitoring signal other than the high speed characteristic monitoring signal. 22. The apparatus abnormality detection device according to claim 18 or 19, wherein the component separating means has an averaging processing unit for averaging the signals resulting from the component separation, and is averaged by the averaging unit. An abnormality monitoring apparatus for equipment, wherein the abnormality signal is configured to be supplied to the correlation extracting means. 23. The apparatus abnormality monitoring device according to claim 22, wherein the component separating means has a frequency component analyzing section for analyzing the frequency component of the high-speed characteristic monitoring signal, and the averaging processing section comprises a low-pass filter. A device abnormality monitoring device characterized by the above. 24. The equipment abnormality monitoring device according to claim 23, wherein the equipment to be monitored is a rotating equipment, and the frequency component analyzing section of the component separating means comprises the rotation frequency of the equipment and its fractional and multiple order components. A device abnormality monitoring device characterized by extracting 25. The abnormality monitoring device for equipment according to claim 22, wherein the component separating means has a plurality of bandpass filters, and the averaging processing section adds an output signal of the bandpass filters. And an RMS value calculation unit that calculates the RMS value of the output signal of the bandpass filter, and an output selection unit that selects and outputs the output signal from the average response processing unit and the output signal from the RMS value calculation unit. A device abnormality monitoring device characterized by the above. 26. The abnormality monitoring device for equipment according to claim 22, wherein the component separating means is one of the high-speed characteristic monitoring signals.
A coherence analysis unit that analyzes a frequency band with strong correlation and a frequency band with low correlation between two signals, and a frequency band with strong correlation and a frequency band with no correlation based on the frequency band analyzed by the coherence analysis unit Correlation band discriminating unit, and a pass band selection unit that passes a signal of a predetermined frequency band of the output signal from the correlation band, the averaging processing unit, the output signal from the pass band selection unit. An apparatus for monitoring abnormality of an apparatus, comprising an effective value calculation unit for calculating an effective value. 27. The device abnormality monitoring apparatus according to claim 22, wherein the device for which abnormality monitoring is performed is a rotating device, and the component separating means executes signal processing with reference to a predetermined rotation angle of the device. An abnormality monitoring device for equipment. 28. A detection means for detecting a large number of characteristics of the monitored device or the monitored device and its peripheral equipment as a large number of monitoring signals, and a high-speed characteristic of the detected large number of monitoring signals, which shows a rapidly changing characteristic. Multiple regression analysis is performed on the component separation means that separates and extracts only the components necessary for monitoring from the monitoring signal, the output signal from the component separation means, and the monitoring signals other than the high-speed characteristic monitoring signal, or the component separation is performed. Only the output signal from the means is subjected to multiple regression analysis to calculate the regression coefficient, and the correlation extraction means for executing the validity evaluation of the calculated regression coefficient and the time change rate of the regression coefficient are calculated to calculate the calculated time. An abnormality determining means for determining whether or not the change rate exceeds a predetermined normal change rate, and determining that an abnormality has occurred in the equipment when the change rate exceeds the predetermined normal change rate. Abnormality monitoring apparatus of the equipment to be. 29. The apparatus abnormality detection device according to claim 28, wherein the component separation means has an averaging processing unit for averaging the signals resulting from the component separation, and the signals averaged by this averaging unit are: An abnormality monitoring device for equipment, which is configured to be supplied to a correlation extracting means. 30. The acoustic characteristics of the monitored device and the amount of hydrogen in the solution used for the monitored device or its peripheral equipment,
Including detection means for detecting a large number of characteristics of the equipment to be monitored and its peripheral equipment as a large number of monitoring signals, and a multiple regression analysis of the detected large number of monitoring signals with the acoustic characteristics as a target variable to calculate regression coefficients. Correlation extraction means for calculating the validity of the calculated regression coefficient as well as calculating the regression coefficient for the above hydrogen amount among the calculated regression coefficients, and determining whether or not the regression coefficient is within a predetermined normal range. An abnormality monitoring device for equipment, comprising: abnormality determining means for determining that an abnormality of water leakage into the aqueous solution has occurred when the abnormality is not within the normal range.