JP2645017B2 - Plant diagnostic method and apparatus - Google Patents

Description

Description: Object of the Invention (Industrial Application Field) The present invention relates to a method and an apparatus for diagnosing an abnormal event occurring in a large-scale power plant such as a nuclear power plant.

(Prior Art) Many proposals have been made on an abnormality diagnosis system for a large-scale power plant.

Next, a typical conventional plant abnormality diagnosis system will be described.

(1) A diagnostic system that captures observation time-series data at major observation points online, converts it into a frequency domain pattern using noise analysis, and compares it with a normal pattern to identify abnormalities. is there. In this system, if the pattern of the abnormality is known, the cause of the abnormality can be identified by comparison with the pattern.

(2) A model of plant equipment is prepared, and observation time-series data is input to the input of the model, the output is predicted, and an abnormality is detected by comparing the output with an actually measured value. There is a diagnostic system based on a so-called model comparison method. In this system, there are two types of models: a model in which a physical model is prepared in advance, and a system in which a physical model is automatically created from observation data using an autoregressive model or the like.

(3) There is a diagnostic system that summarizes the causal relationship of each device in the plant as a cause / result tree (so-called CCT), and does not identify the cause by following this CCT from the observed events and predicts the result. There is a system of knowledge optics application that collects CCT so-called production type rules and creates and diagnoses a chain of events corresponding to CCT according to events.

The conventional diagnostic system as described above has a direction / unsuitability depending on the type of abnormal event that occurs. In addition, the system usually assumes online real-time.

(Problems to be Solved by the Invention) One of the problems of such a conventional system is that it is based on real time. In other words, since high speed is required, the plant model and CCT used for diagnosis are fixed or have simplified the actual device configuration to some extent. Therefore,
In many cases, it is not possible to obtain detailed information necessary for investigating the cause during an actual trouble. Further, in solving such a practical problem, it is often difficult to apply a fixed diagnostic model due to a lack of observation information or a mistake in viewing.

The present invention has been made in view of the above circumstances, and has as its object to provide a plant diagnosis method and apparatus which can be used in accordance with an abnormal state of a plant and can also diagnose the reliability of observation values themselves required for plant diagnosis. Is to provide.

[Configuration of the invention]

(Means and Actions for Solving the Problems) In order to achieve the above object, a first invention of the present application is a typical plant diagnosis method for diagnosing a plant based on observation data obtained from the plant. A basic module that describes a combination of various transfer functions, a dynamic characteristic module including a corresponding effective program and a device module that describes a system to which the device belongs are generated from the basic module, and a system module is further formed from the device module. A hierarchical frame-type database was constructed by creating modules for the entire plant from the system modules, and each module required for prediction was automatically combined from the frame-type database according to the observation data of the plant. Determine the presence or absence of abnormality by comparing the calculated value with the observation data It is characterized by having made it.

The plant diagnostic apparatus according to the second aspect of the present invention provides a basic module frame describing a representative combination of transfer functions, a dynamic characteristic module frame including a corresponding effective program, and a device describing a system to which the device belongs. A hierarchical frame type database comprising a module frame and a system module frame, an observation time series data file for filing data observed from the plant, and an inference request or a request from a result display request device, the hierarchical frame type according to the request. An inference control device that compares a prediction value obtained from a database with an observation value obtained from the observation time-series data file, and infers an abnormal device or sensor by performing the comparison with any of the observation values; And a display device for displaying the inference result. It is intended.

As described above, according to the present invention, a dynamic characteristic model representing an input / output relationship is created for each component unit of a plant and stored in a computer. Next, the dynamic characteristic models of the respective modules are automatically combined in accordance with the target system and the observed signal at the time of the abnormality, and a dynamic characteristic model indicating a causal relationship between the observed signals is automatically created. Since this model is not a fixed model as in the past, even if the observed signal differs depending on the situation, it is automatically created as a model adapted to it.

Therefore, according to the present invention, diagnosis is performed in the following procedure using the model thus created. That is, the value of the observation signal of interest is predicted using the other observation signals and the above model. If there is a discrepancy between the predicted value and the actual observed value, one of the devices corresponding to the model has failed, or the other observation signal used for prediction is incorrect (the sensor has failed), or It can be estimated that there is a possibility that the observation value corresponding to the predicted value is incorrect (the sensor has failed). Conversely, if the observed value and the predicted value match, it is understood that all the observed signals and devices described above are normal. If such a check is performed for all combinations of observed signals, normal equipment and observation signals are extracted, and it can be inferred that the remaining equipment or observation signals may be abnormal.

First, a modular dynamic characteristic model used in the present invention and how to use it will be described.

As shown in FIG. 5 (a), the basic module includes elements such as an adder, an integrator, a proportional element, and a first-order lag element which are the most basic elements of the dynamic characteristic model. These elements can predict the temporal transition of the output signal, given the time-series data of the input signal and the initial value of the output signal.

For example, in the case of a first-order lag element, S is a Laplace operator,
If T is a time constant and x and y are input / output signals Can be expressed as

If we translate this into a time domain model, It becomes a recurrence formula. Here, Δt is the minimum step interval of calculation, t is time, the initial value _yo and the input data sequence x
Given _t (t ≧ o), the output signal y _t can be predicted.

In the computer, a function for performing calculation based on the above algorithm is prepared for each dynamic characteristic module such as a first-order lag, and is referred to from the basic module of the plant in the form shown in FIG. I do.

The input of a certain basic module is connected to the output of another basic module, and when these input / output relations are connected to each other, the component modules of the plant as shown in FIG. can do. When the equipment modules are combined, a system module as shown in FIG. 5 (c) is formed, and when the system modules are further combined, a hierarchical structure of the entire plant as shown in FIG. 5 (d) is formed. Has become.

If the plant dynamic characteristic database is arranged in such a hierarchical form, dynamic characteristic models can be created at different levels according to the purpose and situation of abnormality diagnosis (which system is to be diagnosed and what kind of signal is being observed). Can be built and used. For example, when only a representative process signal of each system is observed and it is desired to diagnose which system is bad, the diagnosis may be performed with respect to a block diagram of each system (FIG. 5 (d)). There is no need for a detailed diagnosis of the equipment. If the system is narrowed down, the diagnosis may be performed by focusing on only the devices in the system, and the diagnosis can be performed without looking at the devices in other systems. As described above, the hierarchical data format enables flexible diagnosis. FIG. 6 shows these data structures in a hierarchical format.

Next, the principle of abnormality diagnosis based on the above hierarchical data will be described. Here, description will be made based on FIG. 5 (c), but the same applies to FIGS. 5 (b) and (d).

In FIG. 5 (c), input and output x ₁ ~x ₄ of each device 1-3
Assuming that is observed, the diagnostic system proceeds with inference according to the following procedure.

(1) With respect to a certain observation signal (for example, x ₂ ), a signal necessary for predicting the observation signal is obtained by following the direction opposite to the arrow. (In this case, x ₁ and x ₄ ), that is, the input of a module having an output x ₂ from among the equipment modules (here, x ₁ ,
x ₄₎ look for. If this input is observed, it ends. If not observed, this input (x ₁
Or Yuku looking similarly input modules with x ₄₎ as output. By such a procedure, a dynamic characteristic model of input observation value → combination of equipment modules → output observation value can be automatically created.

(2) Predicted value by the above dynamic characteristic model (here, predicted value by model using equipment 1 as equipment module using observation values of x ₁ and x ₄ ) and actual observation value (here, x ₂ ) To see if they match. That is, it is determined that a match occurs when an inequality sign in any of the following equations is satisfied, and that a mismatch occurs when the inequality sign is not satisfied.

As a comparison method, for example, Match with or A method such as matching is considered. here,
(T) is a value calculated by the dynamic characteristic model, Is the value x (t) calculated by the dynamic characteristic model and the measured value x during the time (sec) t = 0 to t = T ₀ when the abnormality occurs.
(T) is the maximum absolute value of the difference, and ε is a judgment threshold value.

(3) In the above, when the values match, the output observation value (x ₂ ), the device in the model (device 1), and the input observation value (x ₁
Both the x ₄₎ of can be regarded as normal.

On the other hand, when they do not match, it is presumed that any of these may be abnormal.

Here, the abnormality of the equipment is an abnormality of the plant itself, and the abnormality of the input or output measured value is an interface unit (for example, an A / D conversion board) for taking time-series data into a sensor or a computer used for observation. Means abnormal. (It is very important in practice to distinguish between an abnormality in the observation system and an abnormality in the plant itself.) (4) (1) to (1) for all signals observed
The diagnosis of (3) is performed. Table 3 is created by listing whether there is a possibility of normality or abnormalities for all the observation signals and instruments of interest.

This is an example assuming a sensor failure of x _3, x ₂
It is assumed that the prediction and the observation of coincide with each other, and the prediction and observation of x ₃ and x ₄ do not coincide with each other. Then, a normal / abnormal judgment table as shown in Table 3 can be created.

First, (a) is the result of comparing each output value with and making a comprehensive judgment. Where single
Assuming failure only Okoshira, all × marks of two overlapping x ₃ sensor remains is determined that the abnormality can be regarded as normal. On the other hand, if this is not assumed,
The possibility of abnormalities remains. Therefore, the following steps are used to further narrow down the abnormal location.

(5) After narrowing down in steps (1) to (4),
A signal determined to have a possible sensor failure (here,
x ₃ ) is removed from the observation signal, and the procedures of (1) to (4) are performed again. Here, an example is shown of the case excluding the x _3. At this time, in order to predict the x ₄ dates back to the block diagram of FIG. 5 (c), it is found this is necessary x ₂ and the device 2. If the predictions based on this model match, a decision table as shown in Table 3 is created. The results may have abnormalities in sensor x ₃ as shown in the Table 3 it is determined together (b), the other can be proved that all is normal.

With the above-described procedure, it is possible to construct a dynamic characteristic model based on various combinations of observation signals and determine an abnormality based on the dynamic characteristic model.

(Example) An example of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention.
As shown in FIG.
The devices 3 and 4 each include an adder, a proportional / integral controller, an adjustment valve, and a valve opening sensor, and these devices constitute a feedback system. In addition, the diagnostic device 10 for diagnosing the above-described diagnostic object has a structure as shown in FIG.
It is composed of a hierarchical frame composed of a dynamic characteristic module related frame 11, a basic module frame 12, an equipment module frame 13, a system module frame 14, and the like, and this hierarchical frame is connected to the inference control device 16. On the other hand, observation data obtained from the plant is stored in an observation time-series data file 15, and this file 15 is also connected to the inference control device 16. When an inference request or a result display request is issued from the inference request device or the result display request device 18 to the inference control device 16, the inference control device 16 stores the inference result in the storage file 17 and the display unit 19 according to the procedure described later. Is configured to be displayed.

Next, a case where the diagnostic apparatus 10 is applied to the feedback system shown in FIG. 1 will be described below.

First, as an abnormality in the feedback system, consider the case where reduction of the signal x ₅ by the valve opening degree sensor failure has occurred. Since the request signal x ₁ at this time is constant, the input of the controller 2 increases and further the actual valve opening degree x ₄ through an increase in the output x ₃ of the controller 2 increases. This is shown in the graph of FIG. Diagnosis in such a case is performed in the following procedure.

It can not be a model to predict the verification → x ₁ of x ₁ bypass.

Verification of x ₂ → x ₁ and x _5, to predict the x ₂ using device 1 (summer). (See the model in FIG. 2.) Since both match, x ₂ , x ₁ , x ₅ , and device 1 can be regarded as normal. As a result, a judgment table of the item models in Table 1 is created.

with verification of x ₃ → x ₂ and the device 2 (proportional / integral control system), it is determined table of prediction and items in Table 1 model of Figure 3.

with verification of x ₃ → x ₃ and the device 3 (control valve) can judgment table prediction and items in Table 1 model of Figure 3.

prediction using validation x ₅ → x ₄ and the device 4 (valve opening degree sensor) does not coincide with the observation would be a model of Figure 3. Therefore, as shown in Table 1, it can be considered that any of x ₄ , x ₅ and the device 4 has an abnormality.

Comprehensively judging the above, it is found that there is an abnormality in the device 4 as shown in Table 1.

Next, consider the case where the observed value x ₄ was observed by mistake.
In this case, as shown in FIG. _4, only x4 decreases, and the other data remain constant. The diagnostic procedure in such a case is described below.

It can not be a model to predict the verification → x ₁ of x ₁ bypass.

verification of _{x 2 → x 1, x 5} , to predict the x ₂ from the device 1 (adder). Since the results are as shown in the model of FIG. 4 and are the same as the observed values, a judgment table such as the item model in Table 2 can be obtained.

verification of x ₃ → Similarly, x _2, device 2 (proportional / integral control system)
It predicts x ₃ from making judgment table entry model in Table 2.

Verification of x ₄ → x _3, because it predicts the x ₄ from the device 3 (control valve) does not match the measured values as shown in the model of FIG. 4, the determination table as shown in item model of the second table to obtain Can be

verification of x ₅ → x ₄ and the device 4 from the (valve opening sensor) to predict the x _5. Since it does not match the actual measurement as in the model in FIG. 4, a judgment table like the item model in Table 2 is obtained.

verifies the verification → again x ₅ in x ₅ but, x ₄ which may abnormal as input using x ₃ in further upstream without. Therefore, it is necessary to perform prediction using the device 3 (adjustment valve) and the device 4 (valve opening degree sensor). In this case, since the data coincides with the observation as shown in the model ′ in FIG. 4, a judgment table such as the item ′ in Table 2 is obtained.

When the above-mentioned items are comprehensively determined, as shown in Table 2,
It may have x ₄ only abnormal, it can be seen that the remainder is normal.

[Effects of the Invention] As described above, according to the present invention, the following effects can be obtained.

(1) Diagnosis can be made by separating the abnormality of the plant body and the abnormality of the sensor of the observation signal.

(2) Multiple faults can be dealt with by separating a group of devices that may be abnormal from a group of devices that have been confirmed to be normal.

(3) By using the dynamic characteristic model that simulates the normal operation of the plant, it is possible to accurately diagnose an abnormal event, such as a hunting phenomenon and a small drift event, which is difficult for an expert to distinguish.

(4) Hierarchical use of the plant dynamic characteristic model enables flexible diagnosis at a depth (system-level diagnosis or equipment-level diagnosis) according to the abnormal situation.

(5) Flexible diagnosis is possible by having a dynamic characteristic model and a module type.

[Brief description of the drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, FIG. 2 is a block diagram of a diagnostic apparatus of the present invention, and FIGS. 3 and 4 are graphs for explaining a diagnostic method according to the present invention. FIG. 5 is a diagram for explaining the hierarchical structure of the module according to the present invention, and FIG. 6 is a diagram for explaining the structure of the database according to the present invention. 1 ... Adder, 2 ... Proportional / Integral controller 3 ... Adjusting valve, 4 ... Valve opening sensor 11 ... Dynamic characteristic module function frame 12 ... Basic module frame 13 ... Device module frame 14 ... System module frame 15 ……………………………………………………………………………………………………………………………………………………… …………………………………………………………………………………………………………………………………………………………

Claims (4)

(57) [Claims] In a plant diagnosis method for diagnosing a plant based on observation data obtained from the plant, first, a basic module describing a combination of representative transfer functions is created, and a corresponding effective program is included from the basic module. Create a device module that describes the dynamic characteristics module and the system to which the device belongs, configure a system module from this device module, and further create a module for the entire plant from this system module to form a hierarchical frame-type database, Each module necessary for prediction according to the observation data of the plant is automatically combined from the frame type database. Plant diagnostic method to be used. 2. A dynamic characteristic model describing a basic module necessary for selecting and observing an arbitrary observation signal from a plurality of observation time series data obtained from a power plant and input observation data are automatically obtained from a frame type database. Search and
3. A dynamic characteristic model including a corresponding execution program, a predicted value is calculated, and the predicted value is compared with an observed value to determine the presence or absence of an abnormality. Plant diagnosis method. 3. An arbitrary observation signal is selected from a plurality of observation time series data obtained from a power plant, and a dynamic characteristic model and input observation data necessary for prediction are automatically searched from a frame type database. The procedure of making a predicted dynamic characteristic model, calculating the predicted value, and comparing the predicted value with the observed value to determine the presence or absence of an abnormality is repeatedly applied to any other observed signal. 2. The method according to claim 1, wherein an abnormality of the plant equipment and an abnormality of the sensor are determined separately. 4. A basic module frame describing a representative combination of transfer functions, a dynamic characteristic module frame including a corresponding effective program, and a device module frame and a system module frame describing a system to which the device belongs. A hierarchical frame database,
Observation time-series data file that files data observed from the plant, obtained from the observation time-series data file and the predicted value obtained from the hierarchical frame type database by an inference request or a request from the result display request device And an inference control device for inferring an abnormal device or sensor by comparing the observed values with any of the observed values, and a display device for displaying the inference result. Plant diagnostic equipment.