JP2012014222A - Sensor state determination device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a technique capable of determining states of plural sensors measuring a physical quantity without a reference model such as a regression model.SOLUTION: A sensor state determination device 10 comprises data accumulation means 101, sensor state determination means 102 and determination result notification means 103, and determines states of plural sensors 20 measuring an identical physical quantity. The data accumulation means 101 accumulates time-series data of the physical quantity measured by the plural sensors 20. The sensor state determination means 102 calculates a statistic of a relative error, which is a difference in the time-series data during a predetermined period between a pair of two sensors 20, for each pair of the sensors, and narrows the sensors down to a sensor 20 in an abnormal state on the basis of the statistic of each pair of the sensors. The determination result notification means 103 notifies a result narrowed by the sensor state determination means 102 externally.

Description

  The present invention relates to a sensor state determination device that determines the states of a plurality of sensors that measure the same physical quantity.

  Conventionally, in power plants such as nuclear power plants, preventive maintenance has been performed on the equipment. As one of the preventive maintenance, there is so-called TBM (Time Based Maintenance) in which the equipment is periodically inspected within a period in which the continuous operation of the plant does not exceed a certain period (for example, 13 months). In the TBM, when the inspection is performed, after the operation of the plant is stopped, an overhaul inspection and a function confirmation are performed on most of the devices, and thus the operation rate of the plant is lowered. Furthermore, since equipment that has not deteriorated is disassembled and assembled, excessive inspection may cause equipment failure.

  In order to solve these problems, in recent years, while the plant is in operation, the state of the plant is monitored using the measurement data of the sensors installed in each plant facility, and only equipment that is suspected of deterioration is required as necessary. The introduction of so-called CBM (Condition Based Maintenance), which is an overhaul and inspection, has been studied. The introduction of this CBM is expected to optimize plant maintenance plans and improve work quality.

  In this CBM, one physical quantity may be monitored by a plurality of the same type of sensors installed in the plant equipment. The reason for installing a plurality is to allow monitoring of plant facilities to continue by using other normal sensors even when one sensor fails due to some cause such as drift. In such a CBM using a plurality of sensors, the state of the sensor is monitored on the basis of measurement data measured by the plurality of sensors in order to realize improvement of the plant operating rate and maintenance of plant safety, drift, etc. Online monitoring technology is used to notify the operator of sensors in abnormal states.

  Because of such advantages, various inventions have been made for online monitoring. In the technique disclosed in Patent Document 1, it is determined that the time series data of the monitored plant is normal / abnormal, and the linear or nonlinear regression model is determined from the time series data of normal times in other plants. Prediction data to be predicted is generated, and based on a prediction error that is a difference between the prediction data and time-series data of the monitoring target plant. More specifically, when the prediction error exceeds a threshold value (a value that is a constant multiple of the deviation of the prediction error or a predetermined value), the measurement data is abnormal, that is, the sensor that outputs it is abnormal. It is determined.

  Further, in the technique disclosed in Patent Document 2, for a redundant sensor group installed in a monitoring target plant, a sequential establishment ratio test, specifically a likelihood ratio test is performed for changes in time series data of each sensor. Do. Then, it is determined that there is a drifted sensor when changes in time-series data of the sensors do not match.

Japanese Patent No. 4046309 JP 2009-175870 A

  However, in a technique using a regression model such as the technique of Patent Document 1, since the reliability of drift detection depends on the accuracy of the regression model, a prediction error inevitably occurs. Therefore, a normal sensor may be erroneously determined to be abnormal.

  Therefore, the present invention has been made in view of the above problems, and is a technology that can determine the states of a plurality of sensors that measure physical quantities without using a reference model such as a regression model. The purpose is to provide.

  The sensor state determination apparatus according to the present invention is a sensor state determination apparatus that determines the states of a plurality of sensors that measure the same physical quantity, and that accumulates time-series data of the physical quantities measured by the plurality of sensors. Accumulating means is provided. Then, a relative error statistic that is a difference between the time-series data within a predetermined period in the two sensors constituting the sensor pair is calculated for each of the plurality of sensor pairs, and an abnormality is detected based on the statistic of each of the sensor pairs. Sensor state determination means for narrowing down the state sensors, and determination result notification means for notifying the outside of the result of narrowing down the sensor state determination means.

  According to the present invention, the state of the sensor is determined based on the statistical amount of the relative error without using a standard model such as a regression model. Therefore, since it is not necessary to be affected by the prediction error that occurs when using the model, it is possible to suppress a normal sensor from being erroneously determined to be abnormal.

1 is a block diagram illustrating a configuration of a sensor state determination device according to Embodiment 1. FIG. It is a figure which shows operation | movement of the sensor state determination apparatus which concerns on Embodiment 1. FIG. 1 is a block diagram illustrating a configuration of a sensor state determination device according to Embodiment 1. FIG. 1 is a block diagram illustrating a configuration of a sensor state determination device according to Embodiment 1. FIG. 1 is a block diagram illustrating a configuration of a sensor state determination device according to Embodiment 1. FIG. It is a figure which shows operation | movement of the sensor state determination apparatus which concerns on Embodiment 1. FIG. It is a block diagram which shows the structure of the sensor state determination apparatus which concerns on Embodiment 2. FIG. 6 is a flowchart illustrating an operation of the sensor state determination device according to the second embodiment. It is a block diagram which shows the structure of the sensor state determination apparatus which concerns on Embodiment 2. FIG. It is a block diagram which shows the structure of the sensor state determination apparatus which concerns on Embodiment 3. FIG. It is a block diagram which shows the structure of the sensor state determination apparatus which concerns on Embodiment 3. FIG. It is a figure which shows operation | movement of the sensor state determination apparatus which concerns on Embodiment 3. FIG.

<Embodiment 1>
FIG. 1 is a block diagram showing a configuration of a sensor state determination device 10 according to Embodiment 1 of the present invention. The sensor state determination device 10 is connected to K (K ≧ 2) sensors 20 that measure the same physical quantity, and can determine the states of the K sensors 20 (hereinafter, referred to as “sensor states”). In this description, one of the K sensors 20 may be referred to as a sensor 20-k (k = 1,..., K)). In the sensor state determination apparatus 10 according to the present embodiment, it is possible to narrow down the sensors 20 in an abnormal state such as drift among the K sensors 20 without using a regression model.

  The sensor state determination device 10 is composed of, for example, a CPU and a display device (both not shown). Under the control of the CPU, a data storage unit 101, a sensor state determination unit 102, and a determination result notification unit 103. Are formed as functional blocks.

  Before describing each component of the sensor state determination apparatus 10, the outline thereof will be described.

  The data storage unit 101 receives measurement values of physical quantities measured by the K sensors 20 and generates and stores time-series data in which the measurement values are arranged in the order of measurement. Then, the data storage unit 101 transmits time-series data for a certain period to the sensor state determination unit 102.

  The sensor state determination means 102 forms a plurality of sensor pairs by pairing two sensors 20 out of K sensors 20. The sensor state determination unit 102 calculates, for each formed sensor pair, a difference between time series data within a certain period in the two sensors 20 forming the sensor pair as a relative error, and statistically analyzes a distribution using the relative error as a sample. Calculate the amount. Then, the sensor state determination unit 102 performs a hypothesis test based on the statistic of each sensor pair, and narrows down the sensors 20 in the abnormal state.

  The determination result notification unit 103 notifies the outside, for example, an operator of the result narrowed down by the sensor state determination unit 102.

  Next, each component of the sensor state determination apparatus 10 and the sensor 20 will be described in detail.

Each of the K sensors 20 measures the vapor pressure in the nuclear power plant at a plurality of time points as the same physical quantity, and transmits the measured values at the plurality of time points to the data storage unit 101. In the following description, a measured value related to the vapor pressure at time t _i (i = 1, 2,...) Of the sensor 20-k is represented as X _k (t _i ).

The data storage unit 101 receives the measurement values X _k (t _i ) from the K sensors 20 and stores time-series data of the measurement values X _k (t _i ). Specifically, the data storage means 101 receives time-series data of the discretized measurement value X _k (t _i ) from the K sensors 20 via an A / D converter (not shown), Accumulate it.

FIG. 2 is a diagram illustrating an example of time-series data stored in the data storage unit 101. For each of the K sensors 20, the data storage unit 101 extracts data of the number L of data from the time series data of the accumulated measurement values X _k (t _i ) in time series order. Accumulated data Y _k (t _j ) (j = i,..., I + L−1) is generated. Note that the number of data L is a number that allows hypothesis testing within a range in which the physical quantity to be monitored can be regarded as a steady state. The data storage unit 101 outputs the generated storage data Y _k (t _j ) to the sensor state determination unit 102.

  FIG. 3 is a block diagram illustrating a configuration of the sensor state determination unit 102. As shown in FIG. 3, the sensor state determination unit 102 includes a data correction unit 301, a relative error calculation unit 302, a statistic calculation unit 303, a test unit 304 that performs a hypothesis test, and a determination unit 305.

  Before describing each component of the sensor state determination unit 102 in detail, an outline thereof will be described.

The data correction unit 301 corrects the stored data Y _k (t _j ) from the data storage unit 101 for each of the K sensors 20 to generate correction data S _k (t _j ). The relative error calculation means 302 forms a plurality of sensor pairs {m, n} (m, nε {1,..., K}, m <n) from the K sensors 20. Then, the relative error calculation unit 302 relates the correction data S _m (t _j ), S of the two sensors 20-m, 20-n forming the sensor pair {m, n} with respect to each sensor pair {m, n}. _n (t _j) subtracting the relative error d _{m of,} it is calculated as n (t _j).

The statistic calculation unit 303 calculates a statistic (for example, sample average μ _{m, n} ) of a distribution with a relative error dm _{, n} (t _j ) as a sample for each sensor pair {m, n} The statistics are output to the determination unit 304. As will be described later, this statistic is used for a hypothesis test in the test means 304. Hereinafter, the statistic calculated for each sensor pair {m, n} and used for the hypothesis test is referred to as “sensor pair-specific test amount”. Further, the statistic calculation unit 303 according to the present embodiment uses a distribution (hereinafter referred to as “hypothesis”) based on the relative error d _{m, n} (t _j ) of all the sensor pairs {m, n}. (Referred to as “allowable distribution D”), and the determined hypothetical allowable distribution D is output to the test means 304.

  The test unit 304 performs a hypothesis test for each sensor pair {m, n} based on the sensor pair-specific test amount and the hypothesis allowable distribution D from the statistic calculation unit 303. And the test | inspection means 304 records the sensor pair {m, n} which rejects the hypothesis of a hypothesis test. The determination unit 305 includes the sensor pair {m, n} recorded by the verification unit 304, that is, the sensor pair {m, n} that rejects the hypothesis of the hypothesis test. The sensors 20 in the abnormal state are narrowed down from the two sensors 20-m and 20-n.

  Next, each component of the sensor state determination unit 102 will be described in detail.

<Data correction unit 301>
The data correction unit 301 includes an outlier removal unit 301 a that removes an outlier included in the accumulated data Y _k (t _j ) from the data accumulation unit 101. For example, a conventionally known search method is used as the outlier search method in the outlier removal unit 301a. Data correction means 301 uses the outlier removal unit 301a, performs like accumulated data Y _k (t _j) the disturbance signal removal and interpolation included, the correction data S _k of the stored data Y _k (t _j) (T _j ) is generated. Since the correction data S _k (t _j ) is _{obtained by} correcting the accumulated data Y _k (t _j ), the correction data S _k (t _j ) is time-series data within a certain period, like the accumulated data Y _k (t _j ). The data correction unit 301 outputs the generated correction data S _k (t _j ) to the relative error calculation unit 302.

<Relative error calculation means 302>
The relative error calculation means 302 forms all sensor pairs {m, n} obtained by combining the K sensors 20. For example, when K = 4, the total number of combinations β of the sensors 20 is “6” according to the following equation (1). Specifically, the six sensor pairs {m, n} in this case are {1, 2}, {1, 3}, {1, 4}, {2, 3}, {2, 4}, { 3, 4}.

Then, the relative error calculation means 302 uses the following equation (2) to correct each sensor pair {m, n} in the two sensors 20-m and 20-n that form the sensor pair {m, n}. data S _m (t _j), calculates the difference S _n (t _j) relative error d _m, as n (t _j).

The relative error calculation unit 302 outputs the calculated relative error d _{m, n} (t _j ) to the statistic calculation unit 303.

<Statistics calculation means 303>
FIG. 4 is a block diagram showing a configuration of the statistic calculation unit 303. As shown in the figure, the statistic calculation unit 303 includes a test amount calculation unit 303a that calculates a sensor pair-specific test amount, and a distribution determination unit 303b that determines a hypothesis allowable distribution D. Next, each component of the statistic calculation unit 303 will be described in detail.

The test amount calculation means 303a calculates a sensor pair-specific test amount based on the relative error d _{m, n} (t _j ) of the sensor pair {m, n}. For example, when the type of the hypothesis allowable distribution D determined by the distribution determining unit 303b is a normal distribution, the test amount calculating unit 303a uses the sample average μ shown in the following equation (3) as the sensor pair-specific test amount. _{m, n} is calculated for each sensor pair {m, n}.

It should be noted that the test amount calculation unit 303a may change the type of statistic used as the sensor pair-specific test amount in accordance with the type of hypothesis allowable distribution D determined by the distribution determination unit 303b. For example, when the type of hypothesis allowable distribution D determined by the distribution determination unit 303b described later is not a normal distribution, the test amount calculation unit 303a uses a statistic other than the sample average μm _{, n} as the sensor pair-specific test amount, For example, the sample standard deviation σ _{m, n} shown in the following equation (4) may be calculated for each sensor pair {m, n}.

If the hypothesis allowable distribution D determined by the distribution determining means 303b is a normal distribution, the sample standard deviation σ _{m, n} need not be calculated if the number of data is sufficiently large. The quantity calculation unit 303a may calculate the sample standard deviation σ _{m, n} shown in the following equation (4) for each sensor pair {m, n} as necessary.

In the following description, it is assumed that the test amount calculation unit 303a calculates the sample average μm _{, n} shown in the equation (3) as the sensor-pair-specific test amount. The verification amount calculation unit 303 a outputs the calculated sensor-specific verification amount (sample average μ _{m, n} ) to the verification unit 304.

Next, the distribution determining unit 303b will be described. The distribution determining unit 303b determines the hypothesis allowable distribution D based on the relative error d _{m, n} (t _j ) of all the sensor pairs {m, n}. Specifically, the distribution determining unit 303b determines the type of the hypothesis allowable distribution D based on the distribution of the relative errors d _{m, n} (t _j ) of all the sensor pairs {m, n} from the relative error calculating unit 302. decide. At the same time, the distribution determining means 303b uses parameters (hereinafter referred to as “distribution definition”) for defining the hypothesis allowable distribution D based on the relative error d _{m, n} (t _j ) of all the sensor pairs {m, n}. Called "parameter"). For example, when the distribution determining unit 303b determines a normal distribution as the type of the hypothesis allowable distribution D, the pseudo population average μ _tot and the pseudo population shown in the following expressions (5) and (6) are used as the distribution defining parameters. Standard deviation σ _tot is calculated.

The distribution determining unit 303b may change the type of the statistic used as the distribution defining parameter according to the type of the hypothesis allowable distribution D determined by itself. For example, when the type of the determined hypothesis allowable distribution D is not a normal distribution, the distribution determination unit 303b may calculate a statistic other than the pseudo population mean μ _tot and the pseudo population standard deviation σ _tot as the distribution defining parameters. .

In the following description, it is assumed that the distribution determining unit 303b determines a normal distribution as the type of the hypothesis allowable distribution D and calculates the pseudo population mean μ _tot and the pseudo population standard deviation σ _tot as the distribution defining parameters. The distribution determining unit 303b outputs the determined hypothesis allowable distribution D (the type of hypothesis allowable distribution D and the distribution defining parameter) to the test unit 304.

<Testing means 304>
FIG. 5 is a block diagram showing a configuration of the verification unit 304. For each of the sensor pairs {m, n}, the test means 304 determines “relative error d _{m, n} (t _j) based on the sensor pair-specific test quantity from the statistic calculation means 303 and the hypothesis allowable distribution D. The hypothesis test is performed as to whether or not the hypothesis that the distribution with the sample) follows the hypothesis allowable distribution D is rejected. As shown in the figure, the verification unit 304 includes a rejection verification unit 304a, a reference value setting unit 304b, and a rejection determination recording unit 304c. Next, each component of the test | inspection means 304 is demonstrated in detail.

  Rejection test means 304a outputs a significance level α (0 <α <1) as a reference for the probability of rejecting the above-mentioned hypothesis to reference value setting means 304b. The significance level α is a value set in advance, and a value of 0.05 or 0.01 is usually selected.

Based on the significance level α from the rejection test means 304a and the hypothesis allowable distribution D from the statistic calculation means 303, the reference value setting means 304b is a sensor-to-sensor test quantity region (rejection area) that rejects the above hypothesis. Reference values c ₁ and c ₂ (c ₁ <c ₂ ) that define R) are calculated. As described above, the hypothesis allowable distribution D is determined to be a normal distribution. In this case, the rejection area R exists at the bottom of both ends of the distribution. In this case, the following equation (7) is established for the sensor-by-sensor verification amount (sample average μ _{m, n} ) in the rejection region R.

The limit value c appearing in the equation (7) is a value corresponding to the reference values c ₁ and c ₂ , and | μ _{m, n} | ≧ c is the sensor-to-sensor verification amount (sample average μ _{m, n} ) Is in the rejection zone R, and Pr (| μm _{, n} | ≧ c) indicates the probability that | μm _{, n} | ≧ c.

As described above, since the hypothesis allowable distribution D is determined to be a normal distribution, the hypothesis allowable distribution D can be converted into a standard normal distribution having an average of 0 and a variance of 1. Here, the test amount Z _{m, n} in the standard normal distribution is defined by the following equation (8) using the sensor pair-specific test amount (sample average μ _{m, n} ) in the normal distribution.

When this verification amount Z _{m, n} is in the rejection zone R, the following equation (9) obtained by modifying the above equation (7) is established.

When the portion to be compared with the test amount Z _{m, n} is set to Z _{α / 2} as in the following equation (10), the limit value c can be expressed as in the following equation (11).

Therefore, the region of the sensor pair-specific test amount (sample average μ _{m, n} ) that does not reject the hypothesis in the hypothesis allowable distribution D can be expressed as the following equation (12), and the reference values c ₁ and c ₂ are respectively expressed as It can obtain | require as following Formula (13), (14).

Here, Z _{α / 2} is expressed as in equation (10), but from another viewpoint, it is the value of the test amount when giving the probability on the right side (α / 2) of the standard normal distribution. is there. Therefore, the value of Z _{α / 2 is} a value that can be obtained when α is determined. Specifically, if the probability density function p (x) of the standard normal distribution and its cumulative distribution function F (x) are defined as in the following equations (15) and (16), Z _{α / 2} The value can be obtained from the inverse function of the cumulative distribution function F (x).

In the calculation of the reference values c ₁ and c ₂ described above, the significance level α and the hypothesis allowable distribution D are used, and the sensor pair-specific test amount (sample average μ _{m, n} ) is not actually used. Therefore, the reference value setting unit 304 b can calculate the reference values c ₁ and c ₂ based on the preset significance level α and the hypothesis allowable distribution D from the statistic calculation unit 303. The reference value setting unit 304b outputs the calculated reference values c ₁ and c ₂ to the rejection test unit 304a.

Rejection test means 304a performs hypothesis test based on the sensor pair-specific test quantity (sample average μ _{m, n} ) from statistic calculation means 303 and reference values c ₁ and c ₂ from reference value setting means 304b. This is performed for each sensor pair {m, n}.

Here, if the sensor 20-u (uε {1,..., K}) is in an abnormal state, the above hypothesis is rejected for the sensor pair {m, n} where m ≠ u and n ≠ u. The above hypothesis is rejected in the sensor pair {m, n} where m = u or n = u. If the sensors 20-u ₁ , 20-u ₂ (u ₁ , u ₂ ε {1,..., K}) are in an abnormal state, mε {u ₁ , u ₂ } is not satisfied, and n∈ {u _1, u _2} non sensor pair {m, n} is above hypothesis is not rejected in, m ∈ is {u _1, u _{2}, or, n∈ {u 1, u 2} } The above hypothesis is rejected for the sensor pair {m, n}.

FIG. 6 is a diagram illustrating the relationship between the distribution of the sensor-by-sensor verification amount when the hypothesis is true and the rejection areas R1 and R2. When μ _{m, n} <c ₁ , that is, when the sensor pair-specific verification amount of the sensor pair {m, n} is included in the rejection area R1, the rejection verification means 304a determines that the sensor pair {m, n} Reject the above hypothesis. Further, when μ _{m, n} > c ₂ , that is, when the sensor pair-specific verification amount of the sensor pair {m, n} is included in the rejection area R2, the rejection verification means 304a uses the sensor pair {m, n}. Rejects the above hypothesis.

  Rejection determination recording means 304c records the result of the hypothesis test in rejection test means 304a. Specifically, the sensor pair {m, n} whose hypothesis is rejected and whether the sensor pair-specific verification amount of the sensor pair {m, n} is included in the rejection area R1 or the rejection area R2 are recorded.

<Determination means 305>
3 is based on the sensor pair {m, n} recorded by the rejection determination recording unit 304c and the unrecorded sensor pair {m, n}. }, The sensors 20 in the abnormal state are narrowed down from the two sensors 20-m and 20-n included in the. Specifically, the determination unit 305 determines that the specific sensor pair is in an abnormal state when the sensor pair including the specific sensor is recorded and the sensor pair not including the specific sensor is not recorded. judge. For example, the sensor pairs {1, 2}, {1, 3}, {1, 4} are recorded by the rejection determination recording unit 304c, and the sensor pairs {2, 3}, {2, 4}, {3,4} Is not recorded, the determination unit 305 determines that the sensor 20-1 has an abnormal state.

  The determination unit 305 outputs the sensor 20-k determined to have an abnormal state to the determination result notification unit 103 illustrated in FIG. If the number of abnormal sensors 20-k is larger than [K / 2], it cannot be determined which sensor 20 is abnormal. In this case, the abnormal sensor 20 Is output to the determination result notifying means 103. [X] represents the largest integer equal to or smaller than x with respect to the real number x.

  The determination result notifying unit 103 notifies the information (for example, an operator) about the sensor 20 output from the determining unit 306 to the outside (for example, a worker).

According to the sensor state determination device 10 according to the present embodiment as described above, a statistical amount of the relative error d _{m, n} (t _j ) (sensor-pair-specific test amount) without using a standard model such as a regression model. ) To determine the state of the sensor. Therefore, since it is not necessary to be affected by the prediction error that occurs when using the model, it is possible to prevent the normal sensor 20 from being erroneously determined to be abnormal.

Next, in order to explain another effect of the sensor state determination apparatus 10 according to the present embodiment, a hypothesis test is performed on the statistic of the correction data S _k (t _j ) of each sensor 20-k. Compared with the case where the hypothesis test is performed on the statistic of the relative error d _{m, n} (t _j ) of the sensor pair {m, n} as in the present embodiment.

First, in the former case, when the statistics of the observed physical quantity changes due to a disturbance or the like, the distribution of the correction data S _k (t _j ) of each sensor 20-k is different from the normal distribution. There is a high possibility that the distribution using S _k (t _j ) as a sample does not follow the specific hypothesis allowable distribution D. Therefore, in the former case, if the observed statistics of the physical quantity are non-stationary, it may be determined that the sensor 20-k is actually abnormal even if it is normal.

On the other hand, in the latter (the sensor state determination device 10 according to the present embodiment), when the physical quantity changes due to a disturbance or the like, the correction data S _k (t _j ) of each of the K sensors 20 is the same level. Since it changes, the distribution of the relative error d _{m, n} (t _j ) of the correction data S _k (t _j ) hardly changes from the normal distribution. Therefore, in this case, there is a high possibility that the distribution using the relative error d _{m, n} (t _j ) as a sample follows the specific hypothesis allowable distribution D. Therefore, according to sensor state determination apparatus 10 according to the present embodiment, sensor 20-k is abnormal if sensor 20-k is actually normal even if the observed statistics of the physical quantity fluctuate. Can be prevented from being erroneously determined.

Moreover, according to the sensor state determination apparatus 10 according to the present embodiment, the hypothesis allowable distribution D is determined based on the relative error d _{m, n} (t _j ) of the plurality of sensor pairs {m, n}. Therefore, even if the population distribution of the relative error d _{m, n} (t _j ) of the sensor pair {m, n} changes over time for some reason, an appropriate hypothesis allowable distribution D is obtained at the time when the hypothesis test is performed. Can be used.

  Moreover, according to the sensor state determination apparatus 10 according to the present embodiment, since the abnormal value is removed from the time series data by the data correction unit 301, the reliability of the hypothesis test can be improved. Therefore, the reliability of the state determination of the K sensors 20 can also be improved.

  In addition, in the inspection of nuclear power plants, it is not necessary to perform time-consuming disassembly inspections and function checks on most equipment, so that it is possible to suppress the reduction in the operation rate of nuclear power plants and・ It is possible to prevent equipment failure due to assembly.

  In the above description, the physical quantity measured by the sensor 20-k has been described as being the vapor pressure in the nuclear power plant. However, the physical quantity measured by the sensor 20-k is not limited to the physical quantity related to the nuclear power plant, but may be a physical quantity related to another type of plant. In addition, the physical quantity measured by the sensor 20-k is not limited to the vapor pressure, and may be another type of physical quantity.

<Embodiment 2>
In the first embodiment, the hypothesis allowable distribution D is determined based on the relative error d _{m, n} (t _j ) of all the sensor pairs {m, n}. On the other hand, in the second embodiment, a sensor pair {m1, n1} (m1, n1ε) composed of the remaining sensors other than one specific sensor 20-v (vε {1,..., K}). The hypothesis allowable distribution D is determined based on the relative error d _{m1, n1} (t _j ) of {1,..., K}, m1 <n1, m1 ≠ v, n1 ≠ v).

Hereinafter, the sensor state determination apparatus 10 according to the present embodiment will be described. The sensor state determination apparatus 10 according to the present embodiment includes a statistic calculation unit 313 and a test unit 314 instead of the statistic calculation unit 303 and the test unit 304 shown in FIG. This is the same as the sensor state determination device 10 according to the first embodiment. Therefore, only the statistics calculation unit 313 and the test unit 314 will be described below. Note that the hypothesis allowable distribution D determined based on the relative error d _{m1, n1} (t _j ) of the sensor pair {m1, n1} may be hereinafter referred to as a hypothesis allowable distribution D _v .

<Statistics calculation means 313>
FIG. 7 is a block diagram illustrating a configuration of the statistic calculation unit 313. As shown in the figure, the statistic calculation unit 313 includes a test amount calculation unit 303a and a distribution determination unit 313b. Since the test amount calculation unit 303a is the same as that of the first embodiment, only the operation of the distribution determination unit 313b will be described below.

  FIG. 8 is a flowchart showing the operation of the distribution determining unit 313b. Hereinafter, the operation of the distribution determining unit 313b will be described with reference to this flowchart.

In step s1, the distribution determining unit 313b sets v = 1. If the step s1 is the first time carried out, one particular sensor 20-v which are excluded from the hypothesis acceptable distribution D _v becomes sensor 20-1. In step s2, the distribution determining unit 313b determines the relative error d of the sensor pair {m1, n1} not including the sensor 20-v from the relative error d _{m, n} (t _j ) of all the sensor pairs {m, n}. Select _{m1, n1} (t _j ). For example, when K = 4 and v = 1, there are three sensor pairs {m1, n1} {2, 3}, {2, 4}, {3,4], so in step s2, distribution determination means 313b comprises three relative error _{d 2,3 (t j), d} 2,4 (t j), selects d _{3, 4} and (t _j).

In step s3, the distribution determining unit 313b determines the type of the hypothesis allowable distribution D _v based on the distribution of the selected relative error d _{m1, n1} (t _j ). In step s4, the distribution determining unit 313b calculates a distribution defining parameter that defines the hypothesis allowable distribution D _v based on the relative error d _{m1, n1} (t _j ) of the sensor pair {m1, n1}. Hereinafter, in step s3, the distribution determining unit 313b will be described as determining the normal distribution assumption acceptable distribution D _v. In this case, in step s4, the distribution determining unit 313b calculates a pseudo _population mean μ _vtot and a pseudo _population standard deviation σ _vtot shown in the following equations (17) and (18) as distribution defining parameters.

Here, γ is the number of combinations of sensor pairs {m1, n1} composed of the remaining sensors other than the sensor 20-v. As described above, the distribution determining unit 313b determines the hypothesis allowable distribution D _v (the type of the hypothesis allowable distribution D _v and the distribution defining parameter) based on the selected relative error d _{m1, n1} (t _j ). For example, when K = 4 and v = 1, the relative errors d _2,3 (t _j ), d _2,4 ( ₃ ) of the three sensor pairs {2, 3}, {2, 4}, {3,4] Based on t _j ), d _3,4 (t _j ), the hypothesis allowable distribution D ₁ is determined.

  In step s5, the distribution determining unit 313b increments the value of v by 1 so that v = v + 1. In step s6, the distribution determining unit 313b determines whether v> K. If it is determined in step s6 that v> K, the operation is terminated, and if not, the process returns to step s2.

By repeating the loop process of steps S2 to S6, the hypothesis acceptable distribution D _v is determined for each of the K sensor 20. The distribution determining unit 313b outputs the determined K hypothetical allowable distributions D _v to the testing unit 314.

<Testing means 314>
FIG. 9 is a block diagram showing a configuration of the verification unit 314. For each of the sensor pairs {m, n}, the test unit 314 calculates “relative error d _{m, n} (t) based on the sensor pair specific test amount from the statistic calculation unit 313 and the hypothesis allowable distribution D _v. distribution of _j) the sampling performs hypothesis test whether the hypothesis that follows the hypothesis acceptable distribution D _v "is rejected. As shown in the figure, the verification unit 314 includes a rejection verification unit 314a, a reference value setting unit 314b, and a rejection determination recording unit 304c. Since rejection determination recording means 304c is the same as that of the first embodiment, only the rejection verification means 314a and the reference value setting means 314b will be described below.

Similar to the reference value setting unit 304b according to the first embodiment, the reference value setting unit 314b is configured based on the preset significance level α and the hypothesis allowable distribution D _v from the statistic calculation unit 313. c _v1 and c _v2 are calculated. Then, the rejection test unit 314a is based on the sensor pair-specific test amount from the statistic calculation unit 313 and the reference values c _v1 and c _v2 from the reference value setting unit 304b, similarly to the hypothesis test of the rejection test unit 304a. Thus, the hypothesis test is performed for each sensor pair {m, n}.

According to the sensor state determination device 10 according to the present embodiment as described above, a statistical amount of the relative error d _{m, n} (t _j ) (sensor-pair-specific test amount) without using a standard model such as a regression model. ) To determine the state of the sensor, it is possible to prevent the normal sensor 20 from being erroneously determined to be abnormal as in the first embodiment. Moreover, according to the sensor state determination apparatus 10 according to the present embodiment, the relative error d _{m1, n1 of the} plurality of sensor pairs {m1, n1} configured by the remaining sensors other than the one specific sensor 20-v. Since the hypothesis allowable distribution D _v is determined based on (t _j ), an appropriate hypothesis allowable distribution D can be used at the time of performing the hypothesis test.

  In the above description, it is assumed that there is one specific sensor 20-v. However, even if there are a plurality of sensors, it can be the same as described above.

<Embodiment 3>
In the first embodiment and the second embodiment, the allowable hypothesis distribution D is determined based on the relative error of a plurality of sensor pairs. On the other hand, in the third embodiment, the hypothesis allowable distribution D is set in advance.

  Hereinafter, the sensor state determination apparatus 10 according to the present embodiment will be described. The sensor state determination apparatus 10 according to the present embodiment includes a statistic calculation unit 323 and a test unit 324 instead of the statistic calculation unit 303 and the test unit 304 shown in FIG. This is the same as the sensor state determination device 10 according to the first embodiment. Therefore, only the statistic calculation unit 323 and the test unit 324 will be described below. In the present embodiment, it is assumed that a standard normal distribution having a population mean μ and a population standard deviation σ is set in advance as the hypothesis allowable distribution D.

<Statistics calculation means 323>
FIG. 10 is a block diagram illustrating a configuration of the statistic calculation unit 323. As shown in the figure, the statistic calculation unit 323 includes a parameter calculation unit 323a and a test amount calculation unit 323b.

The parameter calculation unit 323a determines a parameter (statistic) necessary for performing a temporary test based on the relative error d _{m, n} (t _j ) of the sensor pair {m, n}, and the sensor pair {m, n}. Calculate for each. When the hypothesis allowable distribution D is set in advance as a standard normal distribution as in the present embodiment, the parameter calculation unit 323a uses the sample average μ _{m, n} shown in Equation (3) as the parameter. And the sample standard deviation σ _{m, n} shown in Equation (4) is calculated as necessary.

The test amount calculation unit 323b calculates the sensor pair-specific test amount Z _{m, n} based on the parameter calculated by the parameter calculation unit 323a. As in the present embodiment, when the hypothesis allowable distribution D is preset as a standard normal distribution with a population mean μ and a population standard deviation σ, the test amount calculation means 323b calculates the following equation (19): The sensor pair-specific test amount Z _{m, n} is calculated based on the parameter (sample average μ _{m, n} ) calculated by the parameter calculation means 323a. The test amount calculation unit 323b outputs the calculated sensor pair-specific test amount Z _{m, n} to the test unit 324.

<Testing means 324>
FIG. 11 is a block diagram showing the configuration of the test means 324. As shown in FIG. This test means 324 samples “relative error d _{m, n} (t _j ) for each of the sensor pair {m, n} based on the sensor pair-specific test quantity Z _{m, n} from the statistic calculation means 323. The hypothesis test is performed as to whether or not the hypothesis that “the distribution follows the hypothesis allowable distribution D set in advance” is rejected. As shown in the figure, the verification unit 324 includes a rejection verification unit 324a, a reference value setting unit 324b, and a rejection determination recording unit 304c. Since rejection determination recording means 304c is the same as that of the first embodiment, only the rejection verification means 324a and the reference value setting means 324b will be described below.

The reference value setting means 324b calculates a reference value Z _{α / 2} that determines the region (rejection region R) of the sensor-to-sensor verification amount Z _{m, n} that rejects the above hypothesis based on a preset significance level α. To do. Specifically, the following equation (20), which is the same as equation (9), holds for the sensor-pair verification amount Z _{m, n} in the rejection area R.

A portion to be compared with the sensor pair-specific verification amount Z _{m, n} is set to a reference value Z _{α / 2} as shown in the following equation (21).

The reference value setting unit 324b calculates the value of the reference value Z _{α / 2} from the inverse function of the cumulative distribution function F (x) shown in Expression (16). When the hypothesis allowable distribution D is a standard normal distribution, the value of Z _{α / 2} may be obtained using a standard normal distribution table. As described above, in calculating the reference value Z _{α / 2} , only the significance level α is used, and the sensor pair-specific verification amount Z _{m, n} is not actually used. Therefore, the reference value setting unit 324b can calculate the reference value _{Zα / 2} based on the preset significance level α. The reference value setting means 324b outputs the calculated reference value Z _{α / 2} to the rejection test means 324a. Since the average is approximately 0 and the variance is 1 at the reference value _{Zα / 2} , the equation (21) can be rewritten as the following equation (22).

Rejection test means 324a performs hypothesis test on sensor pair {m, _n based on sensor pair-specific test quantity Z _{m, n} from statistic calculation means 323 and reference value Z _{α / 2} from reference value setting means 324b. n}.

FIG. 12 is a diagram showing the relationship between the distribution of the sensor-by-sensor verification amount when the hypothesis is true and the rejection areas R1 and R2. When Z _{m, n} <−Z _{α / 2} , that is, the sensor pair-specific verification amount Z _{m, n of} the sensor pair {m, n} is included in the rejection area R1, the rejection verification means 304a displays the sensor pair. In {m, n}, it is determined that the above hypothesis is rejected. Further, when Z _{m, n} > Z _{α / 2} , that is, when the sensor pair-specific verification amount Z _{m, n of} the sensor pair {m, n} is included in the rejection area R2, the rejection verification means 304a uses the sensor It is determined that the above hypothesis is rejected in the pair {m, n}.

According to the sensor state determination device 10 according to the present embodiment as described above, the population distribution (hypothesis allowable distribution D) of the relative error d _{m, n} (t _j ) of the sensor pair {m, n} is set in advance. Therefore, determination of the hypothesis allowable distribution D in the first and second embodiments can be omitted. Therefore, the processing of the sensor state determination unit 102 can be simplified, and the calculation of the reference value used in the hypothesis test can be facilitated.

  DESCRIPTION OF SYMBOLS 10 Sensor state determination apparatus, 20 Sensor, 101 Data storage means, 102 Sensor state determination means, 103 Determination result notification means, 301 Data correction means, 304, 314, 324 Verification means, 305 Determination means

Claims (6)

A sensor state determination device that determines the state of a plurality of sensors that measure the same physical quantity,
Data storage means for storing time-series data of the physical quantities measured by the plurality of sensors;
A statistic of a relative error, which is a difference between the time-series data within a certain period in two sensors forming a sensor pair, is calculated for each of the plurality of sensor pairs, and an abnormal state is calculated based on the statistic of each of the sensor pairs. Sensor state determination means for narrowing down the sensors;
A sensor state determination device comprising: determination result notification means for notifying the outside of a result of narrowing down by the sensor state determination means. The sensor state determination device according to claim 1,
The sensor state determination means includes
Testing means for performing a hypothesis test on the statistics of each of the sensor pairs;
A sensor state determination apparatus comprising: a determination unit that narrows down an abnormal sensor from two sensors included in the sensor pair based on the sensor pair when the hypothesis of the hypothesis test is rejected by the verification unit. The sensor state determination device according to claim 2,
The sensor state determination means includes
A sensor state determination device that determines a distribution allowed in the hypothesis test based on a relative error between the plurality of sensor pairs. The sensor state determination device according to claim 2,
The sensor state determination means includes
A sensor state determination device that determines an allowable distribution in the hypothesis test based on a relative error of a sensor pair composed of remaining sensors excluding at least one specific sensor. The sensor state determination device according to claim 2,
In the sensor state determination means, a distribution allowed in the hypothesis test is set in advance. The sensor state determination device according to any one of claims 1 to 5,
The sensor state determination means includes
Data correction means for removing abnormal values from the time-series data stored in the data storage means,
A sensor state determination device that generates a relative error of the plurality of sensor pairs based on the time-series data from which the abnormal value is removed.

Images