JPH04191932A - Diagnostic system - Google Patents

Abstract

PURPOSE:To attain effective failure diagnosis to an abnormal situation by advancing inference in the direction of cause and researching the failure cause from the result of observation data and so on with qualitative inference based on qualitative causal relation. CONSTITUTION:When a qualitative relational expression on a parameter does not exist for an object model 4, it is recorded that the qualitative deviation of the parameter is possible to be the cause of failure and the effect is notified to a hypothesis control means 6. Then, the candidate of the causal parameter considered until then is destroyed and it is investigated whether the other candi date of the causal parameter exists or not. Then, when the other candidate does not exist, the hypothesis control means 6 is notified of the relational expres sion used for inference there is possible to be errorneous and the inference processing process is backtracked on to one step before it. Besides, when the other candidate exists, the processing on the other candidate is advanced. Thus, various abnormalities are failure-diagnosed effectively.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial Application Field) The present invention relates to a diagnostic system for qualitatively diagnosing the cause of a failure in a control system of a power generation plant or the like based on its abnormal state.

(Prior Art) One of the important mechanisms in a patented system that is characterized by the control of plants such as power plants is a failure diagnosis function. Conventionally, this fault diagnosis has been carried out exclusively using pre-given diagnostic knowledge, for example, [if then] type rules such as "[jf then E-1 usual]" to diagnose the cause of the fault. ing.

However, when an abnormality occurs for which such explicitly indicated diagnostic knowledge is not available, that is, an unexpected abnormal situation occurs, there is a problem in that it is not possible to diagnose the fault that causes the abnormality.

Therefore, conventionally, in order to diagnose failures due to unexpected abnormal situations, we focused on more basic knowledge about the diagnosis target, and based on qualitative relationships (causal relationships) regarding, for example, the connection relationships of control equipment and the physical phenomena of the control target. It has been proposed to deduce the cause of the failure.

This method regards the occurrence of an abnormal situation in a plant as the occurrence of a significant deviation of plant parameters from normal values. For example, if a component of a plant malfunctions, its influence will be observed as deviations in a number of other plant parameters through process variables such as flow rate and pressure, that is, deviations from normal values. Therefore, if you have information (qualitative causal relationship formula) that allows you to qualitatively estimate the relationship using the deviations of multiple parameters observed in this way, you can detect it by tracing this qualitative causal relationship. The cause of the deviation can be investigated.

Therefore, it is possible to infer the cause of the failure by performing this search using qualitative inference using a qualitative cause-and-effect relationship expression and the qualitatively evaluated plant parameter deviation (qualitative deviation). becomes.

Therefore, in conventional control brand failure diagnosis that uses this type of qualitative reasoning, for example, several causes of failure, such as air leaks from conduits, are prepared in advance, and then these are determined using deep knowledge of the plant's qualitative equations. , other plants by propagating qualitative deviations from their causes to their effects.
We infer patterns of qualitative deviations of parameters. The cause of the failure is then diagnosed by comparing and evaluating the inferred qualitative deviation pattern with the actually observed qualitative deviation pattern.

However, on the contrary, based on the qualitative causal relationship mentioned above,
New problems are likely to occur, such as causing a combinatorial explosion of i, in which inference is made from the result to the cause. Moreover, as the configuration of the diagnosis target becomes more complex, the number of qualitative relational expressions also increases, which tends to cause the problem that the search tree becomes unnecessarily complicated and the number of search routes becomes enormous.

(Problem to be solved by the invention) As described above, the conventional diagnostic system using qualitative reasoning,
Diagnosis is achieved by assuming several causes of failure in advance and propagating those causes in the direction of the result of the causal relationship equation, so there is a problem that it is not possible to effectively deal with unexpected abnormal situations. Ta.

Furthermore, when an unexpected abnormal situation occurs, the validity of the qualitative relational expression used for inference may be violated due to the occurrence of a failure (knowledge uncertainty).

Furthermore, when considering a complex control plant, it is not always possible to obtain all the information necessary for diagnosis. For example, parameters such as pressure and temperature that indicate plant behavior (complexity in all of its plant components).

The present invention was made in consideration of these circumstances, and its purpose is to effectively identify the cause of unexpected abnormal situations in a control plant, even in the absence of explicit diagnostic knowledge. The purpose of the present invention is to provide a diagnostic system that can perform diagnosis.

[Structure of the Invention] (Means for Solving the Problems) A diagnostic system according to the present invention uses observed values in an abnormal state recorded in an observed value database and ideal normal states recorded in a normal value database. Compare values. A deviation detection means that calculates the qualitative deviation of the parameter to be diagnosed by evaluating it, and a qualitative deviation obtained from the target model, which is a database that records the connection relationship between the devices to be diagnosed and the qualitative causal relationship between the parameters. A qualitative inference means for qualitatively inferring the cause of the failure from the qualitative deviation of the parameters based on the relational expression, and a qualitative inference means for controlling the above inference by managing the hypotheses used in the inference process, and and a hypothesis management means for evaluating diagnostic results, which controls the qualitative inference in the qualitative inference means by comparing two observed values and narrowing down the inference search tree, and also controls the qualitative inference between the parameters used in the inference. Even when a qualitative causal relationship equation does not hold, the cause of the failure can be effectively diagnosed by recognizing the causal relationship as the cause of the failure.

(Function) In the present invention, the cause of failure is searched by using qualitative inference based on qualitative causal relationships to advance inference in the direction of the cause from the results of observation data, etc. Therefore, explicit diagnostic knowledge is not provided. This makes it possible to effectively perform failure diagnosis even for unexpected abnormal situations that may never occur. Furthermore, the validity of the qualitative causal relationship equation used in inference is evaluated by comparison with observed values, and incorrect qualitative causal relationship equations are associated with the cause of the failure.

Moreover, according to the qualitative inference means of the present invention, the validity of the qualitative causal relationship equation used in the inference is evaluated by comparing the inference result and the observed value, and a plurality of parameters are inferred, If there is nothing that can be observed among them, it is assumed that only one of them is the cause of the failure, and the inference is divided into multiple parts and each is executed independently (
It is possible to effectively avoid the combinatorial explosion of qualitative solutions caused by partiality of knowledge and qualitative ambiguity that occurs during the inference process, and also to solve the problem of the complexity of the diagnostic target. The increase in search trees due to the large number of sheath relations can also be effectively avoided by selecting search trees. Furthermore, the hypothesis management means can check for inconsistencies in the inference itself, prevent wasteful double inferences, and effectively advance the inference process.

(Example) Hereinafter, an example of the diagnostic system according to the present invention will be described with reference to the drawings.

FIG. 1 is a schematic configuration diagram of an embodiment system.

In this Figure 1, ■ is a normal value database that stores the values of various parameters when the diagnosis target such as a power plant is functioning normally, and This is to calculate the qualitative deviation between the parameter values stored in the value database 1 and the values of the parameters stored in the observed value database 2.

On the other hand, the target model 4 stores in advance the qualitative relationships among the parameters described above, the connection relationships between the plurality of devices constituting the diagnosis target, etc. as a database.

The qualitative inference means 5 qualitatively determines the cause of the failure from the qualitative deviation of the parameter calculated by the deviation detection means 3, based on the qualitative relational expression stored in the target model 4, from the result to the cause. This is an inference. The hypothesis management means 6 manages various inference hypotheses derived in the inference process in the qualitative inference means 5 and controls the inference processing, and also evaluates the diagnostic results obtained by the qualitative inference means 5 to determine a failure hypothesis. is output as the fault diagnosis result.

Due to the qualitative inference in this embodiment system, the values stored in database 1 and observed value database 2 are [A-
ao, D-do+E-eo]. Further, in the observed value database 2, the values of the parameters A, D, and H when an abnormality occurs in the diagnostic object are recorded as [A-a, D=d, E-e], respectively.

However, when the deviation detection means 3 is instructed to that effect regarding the parameter in which the abnormality has occurred, the deviation detection means 3 detects the qualitative deviations ΔA, ΔD, and ΔE for the values of the parameters A, D, and H. , for example ΔA fable a. Calculate as -a ΔD-d, -d ΔE-e, , -e, respectively. Here, the values of the above parameters are a>a□, d<do, ee.

Assuming that the relationship is as follows, the deviation detection means 3 detects the qualitative deviation ΔA,
ΔD, ΔE are 0, - The knowledge about the equipment/structure that constitutes the diagnosis target stored in 4 and the relationship between parameters, specifically, the value of parameter A is the value of parameters B, C - 1. .

[A-B+C], the value of parameter C is the value of parameter D multiplied by a constant negative value [C--constx D], the value of parameter D is constant to the value of parameter E It is the sum of the values [D
-const+E] The cause of the failure to be diagnosed is qualitatively inferred as follows.

The calculation of this qualitative deviation and the qualitative inference processing based on the calculated qualitative deviation and the knowledge stored in the target model 4 will be specifically explained with reference to the schematic diagram shown in FIG.

When an abnormality is found in the diagnostic object, first, parameter 8 of the diagnostic object indicating the abnormality is inputted to the deviation detection means 3 from the outside. Then, the deviation detection means 3 uses the observed value database 2.
and the normal value database l, calculate the qualitative deviation ΔA−[+, and use the deviation ΔA to infer the qualitative deviation of the parameter. Specifically, first, a qualitative relational expression regarding parameter A [A-B+
C] is obtained from the target model 4. From this relationship +, the reason for the deviation Δ-[+] of the parameter A is either the deviation ΔB or [+] of the value of parameter B, or the deviation AC of the value of parameter C is [+]. (Figure 3 ■■).

Therefore, the qualitative inference means 5 inquires of the deviation detection means 3 about the qualitative deviations regarding the parameters B and C.

In response to this, the deviation detection means 3 sets the parameter B.

A qualitative deviation regarding C is calculated with reference to the observed value database 2 and the normal value database 1.

At this time, the deviation detection means 3 detects that the parameters B and C have no observed values, and replies to that effect to the qualitative inference means 5.

However, in this case, the qualitative inference means 5 calculates the deviation of parameter B ΔB−[ using the inference strategy of [single cause assumption]
+] and the relational expression [C=-constX D 3 shown in the deviation edge model 4 of the parameter C, it is inferred that the deviation ΔD-[-] is the cause.

At this point, when the qualitative deviation for the parameter D is sent to the deviation detecting means 3, the deviation detecting means 3 refers to the observed value database 2 and the normal value database 1, and the qualitative deviation ΔD- is [- (Fig. 3 ■).

Therefore, here, it is judged that inference based on the assumption of the other deviation ΔB-[+] described above is unnecessary, and the deviation ΔB-[+]
] Execute pruning of the search tree starting from ]. That is, at this point, the inference based on the assumption of the deviation jB-[+] is stopped ((■) in FIG. 3). Then, as described above, the cause of the qualitative deviation ΔD-'[-], which is a hypothesis that matches the observed value, is further inferred. This type of inference control is called [pruning the search tree using observed values].

Therefore, for the parameter D, the following relational expression [D −E +const
], the cause of the qualitative deviation ΔD-[-] is the qualitative deviation ΔE-[+] of the parameter E. 3 refers to the observed value database 2 and the normal value database 1, and finds that the qualitative deviation ΔE is [ 0, and using the result as a qualitative reasoning means, we find that it is inconsistent with the deviation ΔE-[-] which is the cause of the failure.

Therefore, the qualitative inference means 5 uses the target model 4 as described above.
doubt the validity of the relational expression [D −E +C0n5t ] obtained from the above, infer the failure of the equipment related to this knowledge as [generation of failure hypothesis], and notify the hypothesis management means 6 of the failure hypothesis. Stop the inference process (Figure 3■
).

In this way, by proceeding with qualitative inference processing based on parameter deviations and using the knowledge shown in the target model 4, if an unexpected abnormal situation occurs for which no explicit diagnostic knowledge is given. However, the cause of the failure can be effectively inferred and the failure diagnosed.

Figure 4 shows the inference control procedure in the qualitative inference described above.
In other words, it shows the procedure for expanding and pruning a search tree.

This inference control starts by first performing inference using the qualitative relational expression obtained from the target model 4, and inferring the parameter of the diagnosis target that causes the abnormality (failure) and its qualitative deviation (step The deviation detecting means 3 determines whether or not the observed value exists (step C).

If the observed value for the causal parameter does not exist in the observed value database 2, then it is inquired whether a qualitative relational expression regarding that parameter exists in the target model 4 (step d). If a qualitative relational expression regarding the above parameters exists in the target model 4, the qualitative relational expression is used to further infer the parameter causing the problem and its qualitative deviation, and the process returns to step b described above. (Step e).

On the other hand, if the target model 4 does not have a qualitative relational expression regarding the above parameters, it is recorded that the qualitative deviation of the parameter may be the cause of the failure, and this is notified to the hypothesis management means 6. (Step f). Then, discard the causal parameter candidates that have been considered until now,
Check whether there are any other causal parameter candidates (step g). If there are no other candidates, processing is continued for candidates with possible performance for which the relational expression used for inference is wrong.

On the other hand, if the observed value for the causal parameter exists in the observed value database 2, the deviation detection means 3
It is evaluated whether or not the qualitative deviation of the above-mentioned causal parameters obtained by the method matches the inference result (step i). If there is no qualitative deviation that matches the inference result, as described above, the causal parameter candidates that have been considered so far are discarded and it is checked whether there are other causal parameter candidates (step g). , if there are no other candidates, the hypothesis management means 7 is notified that the relational expression used for the inference may be wrong, and the inference processing process is backed up to the previous step. Track (
Step h). If other candidates exist, the process returns to step b described above to proceed with the processing for the other candidates.

On the other hand, if there is a qualitative deviation that matches the inference result, then the target model 4 is inquired as to whether there is a qualitative relational expression regarding that parameter (step j). Then, the hypothesis management means 7 informs the target model 4f.
and stops the inference process (step k).

Furthermore, if a qualitative relational expression regarding the parameters exists in the target model 4, the search tree divided so far is pruned (step m).

Then, with respect to the parameters organized by this pruning, inference is performed using the qualitative relational expression obtained from the target model 4, and the causal parameters and their qualitative deviations are inferred, and from step b mentioned above, The process returns to (step n).

By controlling the qualitative inference process as described above, the validity of the qualitative causal relationship equation used in the inference can be evaluated by comparing the inference results with observed values, and the causal relationship equation can be If this does not hold true, it is possible to generate a hypothesis that the cause of the failure is in a device element related to the causal relationship equation (generation of a failure hypothesis). Also, during inference, multiple parameters are inferred for one deviation, and
If there is nothing observable among them, the combinatorial explosion of qualitative solutions caused by assuming only one of them as the cause of the failure can be effectively solved by the single-cause assumption described above. It becomes possible to effectively diagnose the cause of failure even in the case of unexpected abnormal situations that are not indicated as avoidable knowledge.

As a result, the hypothesis management means 7 can check for contradictions in the inference itself, prevent wasteful double inference, proceed with the inference process effectively, and handle unexpected abnormal situations that are not shown as explicit knowledge. , it becomes possible to effectively diagnose the cause of the failure.

Note that the present invention is not limited to the embodiments described above. For example, although not shown in the example, qualitative reasoning means 7
It is also possible for the hypothesis management means 7 to detect contradictions in the inference itself during the inference process in , or to monitor the execution of the search to prevent double searches for the same parameter. In this way, the inference process can be performed more efficiently. In addition, the present invention can be implemented with various modifications without departing from the gist thereof.

[Effects of the Invention] As explained above, according to the present invention, partiality and incompleteness of knowledge to be made explicit can be addressed by the [single cause hypothesis] and [observation] described above. Pruning the search tree by
By using inference control as a strategy, it is possible to effectively suppress the increase in the inference search tree. As a result, it is possible to effectively and reliably diagnose various abnormalities to be diagnosed, which brings about great practical effects.

[Brief explanation of the drawing]

Fig. 1 is a schematic configuration diagram of a diagnostic system according to an embodiment of the present invention, Fig. 2 is a diagram showing the concept of processing in the embodiment system, and Fig. 3 is an inference processing process of the qualitative inference means in the embodiment system. FIG. 4 is a diagram schematically showing a processing procedure for inference control in the qualitative inference means of the embodiment system. 1. Normal value database, 2.. Observed value database, 3.. Deviation detection means, 4.. Target model, 5..
Qualitative reasoning means, 6...Hypothesis management means. To the applicant Ken F (Y), Director of the Agency of Industrial Science and Technology, Sugiyu

Claims (1)

[Scope of Claims] A diagnostic system that infers the cause of failure of the diagnostic target using knowledge regarding the function and configuration of the diagnostic target, wherein when an abnormality occurs in the diagnostic target, when the diagnostic target is in a normal state, deviation detection means for calculating a qualitative deviation of the parameter using a normal value database that records parameter values and an observed value database that records observed values of the parameter when the diagnosis target becomes abnormal; Using the qualitative deviation of the parameters calculated by this deviation detection means and the target model that records the connection relationship between the devices to be diagnosed and the qualitative causal relationship between the parameters, the cause of the abnormality of the target to be diagnosed is determined. qualitative inference means for inferring qualitatively, and hypothesis management means for controlling the inference process in the qualitative inference means and evaluating the inference results, the hypothesis management means being the result of the inference. It is assumed that this is caused by a qualitative change in at least one of the parameters, and the intermediate result of the inference is compared with the observed value to narrow down the inference search tree and the qualitative change in the qualitative inference means is performed. A diagnostic system comprising a means for controlling inference and a means for recognizing a causal relationship as a cause of a failure even when a qualitative causal relationship equation between parameters used in the inference does not hold.