JPH0991037A - Fault diagnostic device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To select a diagnostic method suitable for either of a fault tree diagnostic method and a CDM diagnostic method at each alarm and to seek a cause by using a fault tree format even in the case of the CDM diagnostic method. SOLUTION: Fault information from a plant 1 is sent to a diagnostic device 10 and inputted to a knowledge selecting means 11. The means 11 judges whether fault diagnosis using a fault tree is to be executed or fault diagnosis based upon a CDM is to be executed based upon the fault information and starts a fault tree diagnostic means 12 or a CDM diagnostic means 21. When the means 12 is started, fault diagnosis is executed by referring to fault tree diagnostic knowledge 13 and its diagnostic result is displayed on a display means 14. When the means 21 is started, fault diagnosis is executed by referring to CDM diagnostic knowledge 22 and its diagnostic result is outputted to a fault tree conversion means 23. The means 23 converts a CDM diagnostic process into a fault tree format and displays the converted result on the display means 14.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a failure diagnosing device for diagnosing failures of various plants and equipment such as thermal power, nuclear power, water and sewage, chemicals, and cold heat.

[0002]

2. Description of the Related Art Conventionally, for example, thermal power, nuclear power, water and sewerage,
The main diagnosis methods that have been put into practical use as failure diagnosis devices for diagnosing failures of various plants and equipment such as chemistry and cold heat include a failure tree diagnosis method and a matrix diagnosis method. The failure tree diagnosis method is to narrow down the cause of failure in order based on the value of "no refrigerant in the float chamber" based on the observation data such as the top phenomenon "low evaporator pressure" and the observation event "presence or absence of refrigerant in the float chamber". Is. A sample diagram of the flow is shown in FIG. FIG. 9 shows an example of diagnosing a refrigerator, which is similar to the fault tree used in the fault tree analysis, but the value of the observed event is always found in the tree branch from the top toward the cause, for example, “float room”. The presence / absence of refrigerant is narrowed down in different cases such as Yes (established) or No (not established). In the case of this fault tree diagnosis method,
The flowchart, which is the diagnostic knowledge, needs to be created by the designer based on the knowledge such as the failure tree analysis in advance, which is a troublesome work. However, since the created knowledge is logical, it has a characteristic that it is easy to understand as an explanation of the diagnosis result.

On the other hand, in the matrix diagnostic method, as shown in FIG. 10, the observed event is on the vertical axis and the cause candidate is on the horizontal axis, and both are, for example, "0-100" (0: no correlation. 100: completely correlated. 50: Neither can be said.) The cause candidate is obtained based on the CDM (Cause Diagnosis Matrix).

The characteristic of this diagnostic method is that it is relatively easy to formulate empirical knowledge. That is, empirical knowledge such as "when such a cause of failure is such an observed event was observed" is collected, and the degree of correlation is applied based on past experience.

With abundant empirical knowledge, diagnostic knowledge can be created relatively easily. However, the correlation coefficient is not necessarily logical because it is empirical, and it is inferior to the failure tree in terms of verification of created knowledge.

In summary, although it is difficult to create knowledge in the fault tree diagnosis, the knowledge verification and the explanation function are excellent. Although matrix diagnosis is relatively easy to create knowledge, it has a problem in terms of knowledge verification and explanation function.

FIG. 11 (a) shows a configuration example of a fault tree-based diagnostic device, and FIG. 11 (b) shows a configuration example of a CDM-based diagnostic device. In the fault tree-based diagnostic device shown in FIG. 11 (a), the diagnostic device 2 is activated by an alarm such as “low evaporator pressure” from the plant 1, and is input to the fault tree selection means 3. The alarm from the plant 1 has a table indicating which fault tree is used for diagnosis, and the fault tree diagnosis means 4 is activated based on the table. The failure tree diagnosis means 4 makes a diagnosis using the failure tree diagnosis knowledge 5 based on the data from the plant 1 and the observation result from the operator, and displays the result on the display means 6.

In the CDM-based diagnostic device shown in FIG. 11B, the diagnostic device 2'is activated by an alarm from the plant 1 and is input to the CDM selecting means 3 '. The CDM selecting means 3'uses the CDM based on the table included in the alarm.
The diagnostic means 4'is activated. This CDM diagnostic means 4'is
Based on the data from the plant 1 and the observation result from the operator, the CDM diagnosis knowledge 5'is used for diagnosis, and the result is displayed on the display means 6 '.

A display example by the display means 6 and 6'is shown in FIG.
It is shown in FIG. At the top of the screen, there is an observation event (symptom data) input section, and at the bottom, the cause candidates are displayed together with the certainty factor. If the value of the observation event is determined by the sensor, it is automatically input.

[0010]

As described above, the fault tree diagnostic device is characterized in that the created knowledge is logical and is easy to understand as the explanation of the diagnostic result.
There is a problem in that the flowchart, which is diagnostic knowledge, needs to be created by a designer in advance based on knowledge such as failure tree analysis, and the work is very troublesome.

Further, the CDM diagnostic apparatus has an advantage that the diagnostic knowledge can be created relatively easily if the empirical knowledge is abundant, but since the correlation coefficient is empirical, it is not always logical. However, there is a problem that it is inferior to the failure tree in terms of verification of created knowledge and explanation function.

The present invention has been made in order to solve the above-mentioned problems, and it is possible to select either a failure tree diagnosis method or a CDM diagnosis method suitable for each alarm, and verify and explain the knowledge of the CDM diagnosis method. Provides a fault diagnostic device that solves the inferior point in terms of function, reduces the diagnostic creation time, and can investigate the cause in the fault tree format even when the diagnostic result is the CDM diagnostic method and can easily confirm the diagnostic result. The purpose is to do.

[0013]

A failure diagnosis apparatus according to the present invention receives failure information from a failure diagnosis target, and judges whether the failure information is to be subjected to failure diagnosis by a failure tree or CDM. Knowledge selection means, failure tree diagnosis means, failure tree diagnosis knowledge, CDM diagnosis means, C
It is characterized by having DM diagnosis knowledge and a fault tree conversion means for converting a CDM diagnosis process into a fault tree format.

(Operation) The knowledge selecting means receives an alarm from the plant and determines whether the alarm is diagnostic knowledge based on a fault tree or a CD.
It is determined whether or not it is the diagnostic knowledge by M, the diagnosis is performed by the determined diagnostic method, and the diagnostic result is displayed on the display means. CDM
When the diagnosis is selected, the diagnosis result is converted into a failure tree using the failure tree conversion means and then displayed on the display means.

[0015]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a failure diagnosis device according to an embodiment of the present invention. The alarm (fault information) from the plant 1 is sent to the diagnostic device 10,
It is input to the knowledge selecting means 11. This knowledge selection means 11
Determines whether to perform a fault diagnosis using a fault tree or a CDM based on the fault information from the plant 1, and the fault tree diagnosing means 12 based on the determination result.
Alternatively, the CDM diagnostic means 21 is activated. When activated by the knowledge selecting means 11, the failure tree diagnosing means 12 refers to the failure tree diagnosing knowledge 13 to diagnose a failure, and displays the diagnosis result on the displaying means 14. CDM diagnostic means 21
When activated by the knowledge selecting means 11, the CDM knowledge diagnosing knowledge 22 is referred to diagnose a failure, and the diagnosis result is output to the failure tree converting means 23. The fault tree conversion means 23 converts the CDM diagnosis process into a fault tree format and displays it on the display means 14.

Next, the operation of the above embodiment will be described. The plant 1 issues an alarm when the sensor detection signal exceeds the upper limit value. This alarm is coded failure information and constitutes a data file as shown in FIG. 2, for example. That is, the alarm is "diagnosis type", "FT / CDM
No. ”,“ TAG No ”,“ judgment type ”,“ question sentence name ”, etc., and is the head data of the line“ diagnosis type ”
Indicates a fault tree diagnosis if it starts with "F", and indicates CDM diagnosis if it starts with "C". The alarm generated in the plant 1 is input to the diagnostic device 10.

As shown in the flow chart of FIG. 3, the diagnostic device 10 detects when the plant 1 issues an alarm (step A
1), the type of the alarm is determined by the knowledge selecting means 11 (step A2), and whether the alarm is the knowledge of the fault tree (FT) or C
Whether the data (knowledge) for DM is determined by the first letter "F" or "C" of the alarm (step A3). If the first character of the alarm is "F", the knowledge selecting means 11 activates the failure tree diagnosing means 12 (step A4), and if the first character is "C", activates the CDM diagnosing means 21 ( Step A5). When the failure tree diagnosis means 12 is activated, the failure tree diagnosis knowledge 13 is referred to diagnose the failure, and the diagnosis result is displayed on the display means 14 (step A7). When the CDM diagnosis means 21 is activated, the CDM knowledge diagnosis knowledge 22 is referred to diagnose a failure, the failure tree conversion means 23 converts the diagnosis process into a failure tree format, and the display means 14 is displayed. To be displayed (Steps A6, A
7).

In the case of the fault tree diagnosis means 12, the diagnosis process can be viewed in the form of a fault tree in the past. The search for the cause by the fault tree diagnosis method will be described below with reference to FIG. In FIG. 4, AND is AND logic, and OR is OR logic. Also, mi is Observat
ion and Xi are Fault, and fi is Fault rates.

[0019] (1) from plant alarms, established atop m _'F (m _F is abnormal), the established m _E (m _E is normal) was. (2) then 'evaluates _B, m' m _B is satisfied (m _B is abnormal)
If so, the cause is searched further down.

[0020] (3) If m _'B is satisfied (m _B is normal), the search for the underlying aborts. (4) If the formation of x _A , x _B, or m _{A in} the middle of the upper layer is unknown, the formation of x _A or x _{B in} the lower layer is also unknown.

[0021] (5) m _'B is unknown, m _D is the time of incorporation, enacted intermediate x _C, searches ahead from the direction of x _D. (6) When both m ′ _B and m ′ _D are unknown, the sum of the failure probabilities of all the failure events in the lower layer, that is, f _A + f _B and f _C + f _D are compared, and the larger one is prioritized. Then explore.

(7) When no cause can be searched for, only one observation event is inverted, and diagnosis is performed in the order in which the failure event with the highest failure probability is the cause. (8) Diagnosis is repeated at the operator's discretion at the timing when the establishment or non-establishment of at least one observation event is input.

(9) "Unknown observation" means unobserved or observed but the success or failure cannot be determined. (10) Calculate the probability of the cause.

(11) For the double failure, the failure upstream of the signal transmission is searched first. However, the second failure is searched for in the re-diagnosis by performing maintenance for removing the cause promptly.

The cause is searched by the fault tree diagnosis method as described above. Further, when the diagnosis is made by the CDM knowledge diagnosis knowledge 22, the failure tree converting means 23 converts it into a failure tree. This fault tree conversion means 2
The conversion process by 3 is performed according to the flowchart shown in FIG. 5, for example. Data separation by information entropy E is used as the algorithm of the fault tree conversion. Entropy E is a concept originally defined in the fields of thermodynamics and statistical physics, and represents the order and degree of disorder or disorder, but was introduced as a concept of unknown degree in information theory. is there. The information entropy E is defined as the expected value of the amount of information obtained by one trial.
When the occurrence probability that the cause is X is r (0 ≦ r ≦ 1.0), the information amount I (r) of x = −log2r (unit bit).
Entropy is f (r) as its expected value (weighted average)
= −r log2 r f is “f (0) = 0, f
(0.5) = 0.5, f (1) = 0 ”, as shown in FIG.

When r = 1 / e = 0.368, the maximum value is 0.53
Take Since entropy is a quantity indicating the degree of randomness, highly accurate prediction can be performed if the entropy of a certain cause candidate set is small. This means that if the entropy becomes 0, the cause candidates will be narrowed down to one.

In the fault tree conversion process, first, the information entropy before separation is calculated (step B1). When the result of the CDM diagnosis as shown in FIG. 7 is obtained by the failure tree diagnosing means 12, the entropy E before separation has a normal occurrence frequency / abnormality ratio of 3/8: 5/8. E = f (3/8) + f (5/8) = (− 3/8 log2 3/8) + (− 5/8 log2 5/8) = 0.9544 (bits).

Next, the entropy Ei after separation by the observation event Ai is calculated (step B2), and the maximum entropy Ei is calculated.
Find mini (step B3). And entropy E
It separates by the observation event which becomes mini (step B4), and it is checked whether there is an unused observation event (step B4).
5) If there is an unused observation event, return to step B1 and repeat the same processing. That is, the entropy before and after separation is compared for each observation event, and the observation event with the smallest entropy is selected. When the observed event has several attribute values, the weighted average value of each entropy is compared. By repeating this for all the observation events, it is converted into a failure tree (step B6). FIG. 8 shows an example of the above conversion result. In this way, by performing the conversion processing into the failure tree by the failure tree conversion means 23, it becomes possible to pursue the diagnosis result even in the case of the CDM diagnosis method.

[0029]

As described above in detail, according to the present invention, C
It is possible to solve the inferiority of the knowledge verification and the explanation function of the DM diagnosis method (matrix diagnosis method), and select one of the failure tree diagnosis method and the CDM diagnosis method for each problem, The diagnostic preparation time can be reduced, and the diagnostic result can be pursued even in the case of the CDM diagnostic method, so that the diagnostic result can be easily confirmed.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a failure diagnosis device according to an embodiment of the present invention.

FIG. 2 is a diagram showing an example of failure information output from the plant in the same embodiment.

FIG. 3 is a flowchart showing a processing operation in the embodiment.

FIG. 4 is a view for explaining an example of searching for a cause by the fault tree diagnosis method according to the same embodiment.

FIG. 5 is a flowchart showing a processing example of a fault tree conversion unit in the same embodiment.

FIG. 6 is a diagram showing information entropy in the same embodiment.

FIG. 7 is a view showing a display example by CDM diagnosis in the same embodiment.

FIG. 8 is a diagram showing a conversion example of a fault tree conversion unit in the same embodiment.

FIG. 9 is a flowchart showing an example of a diagnostic fault tree.

FIG. 10 is a diagram showing an example of CDM diagnosis.

11A is a diagram showing a configuration example of a conventional fault tree diagnostic device, and FIG. 11B is a diagram showing a configuration example of a conventional CDM diagnostic device.

FIG. 12 is a diagram showing a screen display example of a conventional failure diagnosis device.

[Explanation of symbols]

 1 Plant 10 Diagnostic Device 11 Knowledge Selection Means 12 Fault Tree Diagnosis Means 13 Fault Tree Diagnosis Knowledge 14 Display Means 21 CDM Diagnostic Means 22 CDM Knowledge Diagnostic Knowledge 23 Fault Tree Conversion Means

Claims (1)

[Claims] 1. A CD for receiving failure information from a failure diagnosis target and performing failure diagnosis on the failure information using a failure tree?
Knowledge selecting means for deciding whether or not to carry out failure diagnosis by M, failure tree diagnosing means, failure tree diagnosing means, CDM diagnosing means, CDM diagnosing knowledge, and a failure converting the CDM diagnosing process into a failure tree format. A fault diagnosis device having a tree conversion means.