JPH0355854B2 - - Google Patents

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a device that supports equipment failure diagnosis by inspection and maintenance personnel in large-scale plants, and in particular improves the accuracy of diagnosis by improving the accuracy of failure signs. The present invention relates to an equipment failure diagnosis support device suitable for.

[Background of the invention]

One device has multiple ways of failure, that is, failure modes. For example, ``the valve is stuck in the open position and does not move'', or conversely, ``the valve is stuck in the closed position and does not move''. Large plants are composed of a large number of devices, and the failure modes of the devices are linked by complex causal relationships. Therefore, various difficulties arise in estimating the true cause of failure (faulty equipment and its failure mode) from observed abnormal phenomena (symptoms). As a countermeasure to this problem, many automatic diagnosis systems using computers have been proposed.

Their general principles are as follows. Assuming that a is a vector of i failure causes, b is a vector of j failure symptoms that appear as a result, and R is an i×j matrix that associates causes and results, the causal relationship of failures can be expressed by the following equation.

a _p R=b; ∨ i (a _i ∧r _ij )=b _j ; ∨: max, ∧:
min} ...(1) Here, a _i = 1 Failure 0 No failure r _ij = 1 Causal relationship 0 No causal relationship Equipment failure diagnosis is based on the causal relationship R created in advance and the observed symptoms b Based on this, in order to find the cause of failure a, we must perform the inverse calculation of equation (1).

This method had the following problems.

(1) At the initial stage of a failure, it is unclear whether it is a failure or not.

(2) There is also ambiguity in causal relationships.

(3) Due to (1) and (2), diagnosis may not be possible or there may be a misdiagnosis.

As a countermeasure for this, for example, "Abnormality diagnosis using fuzzy logic", "System and Control" magazine, Vol. 24,
No.11, pp.719-725.1980 was proposed. In this ambiguous diagnosis method, the concept of strength and weakness is included in a _i and r _ij , and a _i = [0, 1] r _ij = [0, 1]. In this method, ambiguous symptoms can also be input, and the diagnosis will naturally be as follows, reflecting the degree of certainty (accuracy): b _j =[0,1]. Furthermore, the diagnosis is reinforced by a back-of-the-envelope diagnosis using a ^c _p R = b ^c (2), which relates that there is no failure. In order to improve diagnostic results with a certain level of accuracy and diagnostic accuracy, a request for confirmation of necessary symptoms is output.

However, even when using the ambiguous diagnosis method, the following problems still occur.

(1) During the process of failure diagnosis by humans, judgments are made based not only on cause and effect relationships, but also on whether the causative equipment is likely to fail (high or low probability of failure).

(2) Failures cannot be simply divided into cause and effect. In some cases, there are many layers, and in some cases, the result is the simultaneous failure of multiple devices (multiple causes).

(3) Even though symptoms may occur due to failure of multiple devices (multiple causes), diagnosis is made on the assumption that the symptoms are caused by one cause (common cause).

As a countermeasure to these problems, we have developed a fault tree, a cause-effect tree, and a cause-effect tree.
There is a method for diagnosing failures by using a consequence tree (Consequence Tree) or a failure cause-and-effect relationship network, and correlating accuracy and failure probability with this. This method uses a failure cause-and-effect relationship network to multiply the accuracy of b and R by the failure probability of a with the accident probability converted using the Bayes decision rule based on the observed conditions of b. It is unique in that it is taken as the diagnostic accuracy (load).

However, this method still has the following deficiencies.

(1) Common causes and multiple causes are viewed at the same time, but
Common causes are more likely to occur.

(2) When a contradiction occurs when inferring a cause from a symptom and conversely inferring a symptom from the cause, either an error occurred in the observation of the symptom or a new causal relationship that is not in the causal relationship entered in advance has occurred. Although this is a possibility, no countermeasures have been considered.

The following literature introduces recent trends in fuzzy logic: ``Fuzzy theory that has begun to be put into practical use'', ``Nikkei Electronics'' magazine, 1984.12.3.No.
357, pp. 165-192.

[Purpose of the invention]

An object of the present invention is to provide an equipment failure diagnosis support device that can estimate the true cause of a failure based on a combination of common cause and multiple cause estimation results, and can also suggest the possibility of an error in symptom observation or the occurrence of a new causal relationship. That's true.

[Summary of the invention]

In the present invention, the cause-and-effect relationship of failure is defined as “if-
The method of calculating the weight of cause estimation is shown below.
b _g ^* , and its observation accuracy is <b _g ^* > (g=1~
G). Assuming that any k of b _g ^* occur due to one common cause, the weight <q _ai > _k for a _i to be the common cause is calculated using the following formula.

That is, the weight of cause estimation is set as the product of the probability of an accident when k symptoms b _g ^* are observed, the possibility of reaching a causal relationship, and the geometric mean of the accuracy of the symptoms. In the load evaluation formula (3), the weight ranges from 0 to 1 for both multiple causes and common causes, giving almost equal weights, making it impossible to distinguish between common causes and multiple causes.

By the way, in the process of fault diagnosis, humans consider not only the causal relationship of the fault, but also whether the cause is likely to occur. In this case, when comparing common causes and multiple causes, common causes are generally more likely to occur.

In addition, in order to improve diagnostic accuracy, we search backwards from the presumed cause to the result, and when a new symptom that is not among the observed symptoms is estimated, we treat it as an overlooked symptom and request observation. The method is a well-known method for abnormality diagnosis.

However, it is sufficient if the symptom requested for observation is confirmed, but if it is not confirmed, there is a possibility that the original symptom observation was incorrect. Therefore, it should be considered that an event corresponding to a new causal relationship that is not in the pre-input causal relationship has occurred only after confirming that there are no errors in the original observation.

The present invention provides a device that not only determines missed symptoms, but also determines the possibility of false symptoms and the possibility of new causal relationships occurring, and displays this to maintenance personnel or fault diagnosis experts. It is something to do.

Observing new symptoms may require operating equipment outside of normal inspection and maintenance procedures. In this case, it is important to indicate the location where the operation will be performed, the possibility of inspection during normal operation, and each step required for the operation.

[Embodiments of the invention]

An embodiment of the present invention will be described below.

FIG. 1 is a block diagram showing the overall configuration of an equipment failure diagnosis support device in one embodiment of the present invention.
In the figure, 1 is an input/output device with a display screen, 2 is an input/output control device, 3 is an arithmetic processing device, 4 is a processing program storage device, and 5 is a database storage device. The processing program storage device 4 further includes a cause side search program storage device 4a, a symptom side search program storage device 4b, and a screen display program storage device 4c. In addition, the database storage device 5
further comprises a causal relationship storage device 5a, a cause candidate storage device 5b, a symptom candidate storage device 5c, and an observation operation procedure storage device 5d.

Next, the failure diagnosis processing procedure of this device will be explained step by step with reference to FIG.

1 In process 6, the observed failure symptoms are input from the display screen input/output device 1 and stored in the symptom candidate storage device 5c.

2 In process 7, first, the cause-side search program 4a is called in the arithmetic processing unit 3, and the calculation shown in equation (3) is performed based on the causal relationship stored in the causal relationship storage device 5a. The procedure for this calculation proceeds as follows.

(a) The load of the cause is calculated assuming that all the symptoms are caused by one common cause, and if the load is close to 1 in process 8, it is determined to be a common cause and stored in the cause candidate storage device 5b.

(b) If the load is close to 0, calculate the loads of multiple causes sequentially using equation (3), and if it is determined in process 9 that the load is close to 0, proceed to process 10 as failure diagnosis is not possible.

(c) If the loading is close to 1, it is either an error in symptom observation or multiple causes. In process 11, it is determined whether or not it is the first observation result. If it is the first time, it is determined that it is an error symptom, the screen display program 4c is called by the arithmetic processing unit 3, displays the error symptom observation request 12, and the program is initialized.

(d) If it has been re-observed, it will be determined that there are multiple failures, and the cause estimation will end.

3 In process 13, the cause and symptom storage device 5b
Symptoms are estimated by the symptom-side search program 4a called by the arithmetic processing unit 3 based on the causes stored in the causes and the causal relationships in the data 5a.
The judgment procedure here proceeds as follows.

(a) In process 14, if all the estimated symptoms are observed, the cause of the failure is identified and the diagnosis ends.

(b) If there is a new symptom that is not included in the symptoms stored in 5c, or if there is a problem whose cause of failure cannot be estimated by processing 9, there is a possibility that the symptom has been overlooked. If it is determined by the process 10 that the object has come to this processing section for the first time, an observation request 15 is displayed indicating that a missed symptom exists, and the program is initialized.

(c) If you come to process 10 twice, process 16
If this is the first time it has come, it is determined that there is an error in the causal relationship, a review request 17 is displayed, and the program is initialized.

(d) If the process 16 is entered again, a message indicating that diagnosis is not possible is displayed and the diagnosis is terminated.

In addition, when displaying the error indication 12 or the missed indication 15, the position of the relevant equipment, the possibility of observation, and the procedure (if an operation is required) stored in the observation operation procedure storage device 5d are also displayed. It is also displayed on the input/output device 1 with a display screen by the screen display program 4c.

Next, one embodiment of the present invention shown in FIGS. 1 and 2 will be described in more detail using a specific example.

FIG. 3 is a system diagram showing a piping system (when control rods are inserted) of a control drive system of a boiling water nuclear power plant used to explain in detail an embodiment of the present invention. Driving water 1 supplied from driving water header 18
9 is a manual valve 20, a check valve 21, and an electromagnetic cutoff valve 2.
2. It flows into the lower part 25 of the control rod drive mechanism 24 via the manual valve 23 and pushes up the control rod 26. At this time, the water accumulated in the control rod drive mechanism upper part 27 passes through the manual valve 28, the electromagnetic cutoff valve 29, and the manual valve 30, and is discharged as discharge water 31. in this case,
Solenoid cutoff valves 22 and 29 are open, and solenoid cutoff valves 32 and 33 are closed. Conversely, when the control rod is withdrawn, the electromagnetic cutoff valves 32 and 33 are open, and the electromagnetic cutoff valves 22 and 29 are closed. The driving water flows in a completely reverse process to that during insertion. Air-operated valves 34 and 35 that are normally closed during scram (emergency reactor shutdown)
is opened, and the water in the accumulator 36 is rapidly injected into the control rod drive mechanism lower part 24. In the figure, 3
7 is scram supply water, 38 is scram discharge water, 3
9 is cooling water, and 40 is a drive water flowmeter.

Hereinafter, the failure diagnosis procedure of the present invention will be explained in more detail in correspondence with each step shown in FIG. In each sentence, the steps in parentheses are the processing steps shown in FIG.

Figure 4 is a part of the causal relationship network diagram regarding the control rod piping system, and consists only of OR relationships. Solid-line ellipses represent failure modes that can be observed during operation, and dashed-line ellipses represent failure modes that can be observed during periodic inspections.

It is now assumed that failure mode 41 (high drive water pressure) is observed as a symptom and the certainty of observation is 0.9 (steps 6 and 7). In other words, in equation (3), the number of symptoms is g = 1, and symptom b ₁ ^* is failure mode 41
Therefore, <b ₁ ^* >=0.9. Since the number of possible combinations is one, only k=1, and L=1. Failure mode a _i targets all failure modes except for symptoms on the causal network diagram. Among these, failure modes that can be traced back to the cause from the symptom b ₁ ^* , that is, in which π g∈kP _ig is not 0, are failure modes 42 and 49. Symptom 41
If we find the posterior probability under the condition that is observed, we can obtain all the weights from equation (3). These are the numbers shown to the right of failure modes 42 to 49 in FIG. 4, and this value represents the ratio of the probability of occurrence of each failure mode. In this example, if we look at the cause side, failure mode 46 is selected as the one with the largest load (step 8).

By the way, if failure mode 49, which is the third candidate, is also the true cause, then in addition to failure mode 41, failure modes 45, 50, and 51 should be observed as symptoms. Once we have reached a tentative conclusion that there is a common cause, we estimate the symptoms (step 13).
When compared with the input symptoms (step 14),
It is expected that there will be new symptoms that are not stored there. In this case, it is determined that this is the first symptom (step 10), and an observation request is displayed as a missed symptom (step 10).
15), the program is initialized.

A case in which failure mode 50 is observed as a result of maintenance/inspection personnel re-simplifying based on a missed observation request will be described using FIG. 5. Figure 5 shows
Failure mode 50 was observed, and the observation accuracy was 1.0.
The case is shown below. That is, <b ₂ ^* >=1.0. In this case, the number of symptoms will be g=2. The number of possible combinations is k=2 and k=1. First, calculations will be made for the case of k=2, that is, a common cause. The results are shown without the cutlet on the right shoulder of Figure 5. There are failure modes 45 and 49 that can be reached from two symptoms, and their load is 0.95.
Since this is close to 1 (step 13), the diagnosis ends with the assumption that the cause of the failure has been estimated (step 14).
Note that the inside of the box shows the load when g=2 and k=1 for reference (steps 13 and 14). In this case, failure modes 43 and 45 are evaluated to have occurred independently.

Next, an example of displaying the diagnosis results will be shown, taking as an example a case where another symptom of the control rod drive system is observed. Here, we will discuss the case where insertion and stop of the control rod drive mechanism is impossible. FIG. 6 is a display example of the cause estimation results. In this case, four types of cause candidates are estimated until the load is ~0.5 (steps 6, 7, 8),
Unable to narrow down the cause (Step 9). Therefore, in this case, since it is a new symptom (step 10),
As shown in FIG. 7, an observation request (for missed signs) is displayed (step 15).
FIG. 8 shows an example of display based on the re-observation results based on this. This is a case where the symptom observation accuracy of the electromagnetic cutoff valve 121 is 0.7, and the cause of the failure has been narrowed down to the leakage of the electromagnetic switching valve (step 6,
7, 8, 13). Also shown here are the results of estimating the failure mode of the component of the device that is the cause. Note that the valve numbers are the numbers shown in the brackets in Figure 3.

Now, some inspections to observe symptoms require operations. For example, in Fig. 3, valve 1
21. In checking whether there is a leak in the valve 123, the information in the observation operation procedure storage device 5d is
The screen display program 4c displays the operation as shown in the figure. Accordingly, systematic tests need to be carried out. In the example shown in FIG. 9, a leak can be determined based on the difference in system flow rate.
This case is an example of a case where symptoms of the estimated cause candidate can be observed, and the cause of the failure can be reliably identified.

Next, a simple example will be given of a request to observe an error sign other than a request to observe a missed sign, and a request to review a causal relationship. Here, in order to simplify the discussion, failure modes of each device will be abbreviated using symbols. Figure 10 shows cases in which the cause cannot be estimated.
The figure shows an example where it is impossible to estimate a common cause but it is possible to estimate multiple causes, and FIG. 12 shows an example where a result occurs due to simultaneous failures of devices (connected by AND gates).

In Figure 10, events 52 and 53 have observation accuracy.
If 0.5 is observed (steps 6, 7, and 8), the certainty of the causal relationship is also low, so the certainty of candidates 54 and 55 estimated as causes is as low as 0.05 (step 9). Therefore, since additional observation information is required here, the missed observation request shown in FIG. 13 is displayed (steps 10 and 15). If the observation result remains unchanged (from step 6 to step 10)
At this time, it is determined that there is a possibility of an error in the causal relationship (Step 16), and as shown in Figure 14, a request is made to a skilled maintenance worker (expert) who is sufficiently familiar with the causal relationship to review the causal relationship. Display (Step 17). Furthermore, if there is no error in the causal relationship, the result of inestimability shown in FIG. 15 is displayed (reenter step 16).

In Figure 11, the two observed symptoms 56,5
7 common causes are estimated (steps 6, 7, 8).
In this example, cause 62 is estimated, but the accuracy is as low as 0.05, as shown by the numerical value without a bracket in the figure. Next, estimate multiple causes (step 9).As shown by the numbers in brackets in the figure, causes 60, 61, and 62 are 0.5
indicates a value equivalent to In this case, if the failure probability of each device is 0.01, the probability of occurrence of a common cause is 0.01, but the probability of occurrence of multiple causes is as low as 0.0001.
In other words, there is a contradiction between the load and the probability of occurrence regarding cause estimation. Since this is the first time, a request to review the symptoms (error symptom observation request) is first displayed as shown in Figure 16 (steps 11 and 12). If the observed symptoms remain unchanged, it is determined that multiple failures have occurred and the diagnosis is terminated (steps 9, 11,
13, 14). Note that 58 and 59 are intermediate failure modes.

In FIG. 12, result 63 is connected to failure modes 65 and 66 via AND gate 64. If result 63 is observed with an observation accuracy of 1.0 (steps 6, 7, and 8), causes 65 and 66 are sure to occur (step 9). However, if the symptom side search is performed, failure mode 67 will definitely occur. Therefore, as shown in Figure 17,
Display AND symptom observation request (steps 13, 14,
Ten). This request is a special example of a missed symptom observation request. If failure mode 67 is not observed (step 10), a request for review of symptom 63 is displayed (steps 16, 17). If the observation result remains unchanged (step 10), a request to review the causal relationship is displayed (steps 16 and 17). If the causal relationship remains unchanged, a message indicating that the cause cannot be estimated is displayed and the diagnosis is terminated (step 16 re-entering).

In addition, in the fault diagnosis of the present invention, a method is adopted in which a judgment is made based on the observation results of the inspection and maintenance personnel, and the procedure of requesting observation to further improve the accuracy of diagnosis is repeated interactively until sufficient accuracy is obtained. There is.
If inspection and maintenance are automated in the future, an automatic diagnostic device can be constructed by applying the processing procedure shown in the present invention. The diagnosis results will be displayed on the display screen in the central control room.

〔Effect of the invention〕

According to the present invention, by combining search results for common causes and multiple causes, in addition to probable causes, missed symptoms,
Error symptoms and new causal relationships can be determined, more detailed failure symptom observations can be made, and failure diagnosis can be performed accurately.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the equipment failure diagnosis support system according to the present invention, FIG. 2 is a flowchart showing its processing procedure, and FIG. 3 is a control rod drive used as an example of the failure diagnosis target of the present invention. A system diagram showing the structure of the system, Figures 4 and 5 are causal relationship network diagrams of the system, Figures 6 to 9 are screen diagrams of the system diagnosis results, and Figures 10 to 12 are diagrams of the system diagnosis results. Causal relationship network diagram of other systems, Figures 13 to 1
FIG. 7 is a screen diagram showing the diagnosis results of these systems. 1... Input/output device with display screen, 2... Input/output control device, 3... Arithmetic processing unit, 4... Program storage device, 4a... Cause side search program storage device, 4b... Symptom side search program storage equipment, 4c
...Screen display program storage device, 5...Database storage device, 5a...Causal relationship storage device, 5
b... Cause candidate storage device, 5c... Symptom candidate storage device, 5d... Observation operation procedure storage device.

Claims (1)

[Claims] 1. An input device for inputting observed failure symptoms, etc., a storage device for storing causal relationships, cause candidates, and symptom candidates of the failure, and an input device for searching for the cause of the failure based on the failure symptoms. In an equipment failure diagnosis support system consisting of an arithmetic processing unit that searches for failure symptoms and makes various judgments based on the cause of the failure, and an output display device that displays failure diagnosis results, etc., the arithmetic processing unit is unable to identify the cause of the failure. Occasionally, it has the function of issuing requests for observation of erroneous signs, requests for observation of missed signs, and requests for reviewing causal relationships, depending on the situation.
An equipment failure diagnosis support device, wherein a storage device stores observation procedures to be executed when each of the requests is issued, and an output display device displays identification results and the requests. 2 In claim 1, the arithmetic processing device has a common cause and multiple cause search function as a failure cause search function, and first determines that the common cause is a common cause when the probability of the common cause is high; When the accuracy of the common cause and the multiple failure are both low, it is judged that the cause of the failure cannot be estimated, and when the accuracy of the common cause is low but the accuracy of the multiple failure is high, it is judged that there is a possibility of an error in the first observation result. An equipment failure diagnosis support device characterized by having a function of determining that there is a multiple failure when there is no change in the second observation result. 3. In claim 1 or 2, an arithmetic processing device performs a symptom search based on a symptom search based on a highly probable cause candidate obtained through a cause search based on observed symptoms. 1. An equipment failure diagnosis support device characterized by having a function of determining that there is an error in a causal relationship stored in a storage device and issuing a request to review the causal relationship when an unobserved symptom is present in the device.