JPS59112307A - Device for estimating cause of failure - Google Patents

Abstract

PURPOSE:To estimate the cause of a failure automatically and to change the combination of system constitutional factors easily by applying the structure and actions of the system as knowledge without using preceeding analysis. CONSTITUTION:A detector 13 in a failure cause estimating device enter observation data, a failure detecting part 14 detects a failure of the observation data, and if a failure is detected, a normal value estimating part 15 estimates a value to be indicated by the observation data. At the estimation, the structure and actions in a knowledge storing part 19 for the structure of an objective system and a knowledge storing part 21 for the actions which are prepared in a knowledge storing part 20 are used. Subsequently, a failure candidacy selecting part 16 selects failure detecting element by using the structure and knowledge and an I/O inconsistency detecting part 17 discriminates the inconsistency of the observed I/O elements out of the failure detecting elements. An automatic test generating part 18 generates the test of only one failure constitutional element by using the structure and actions and a result displaying part 22 displays the tested result.

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention relates to anomaly cause estimation), particularly an abnormality cause estimation suitable for a system having - or more observation points and consisting of - or more constituent elements. Regarding equipment.

[Prior art]

Conventional anomaly cause estimation involves describing the relationship between possible anomalies and the causes of the anomalies regarding the target system as a scenario, inputting it into a computer, comparing it with actual abnormal events, and estimating the cause of the anomaly. The condition part (+r) and the result part (the
We propose a method to identify the cause of an abnormality by inference, combining the necessary knowledge when necessary, by using the knowledge consisting of n) to rule out the causal relationship between cause and effect and organizing it as a knowledge base. has been done. However, in any case, these methods require experts to perform a cause-and-effect analysis in advance, and the ability to infer the cause of an abnormality strongly depends on the completeness of this preliminary analysis. . These analyzes require a great deal of effort, and if the combination of system components is changed, the analysis has to be restarted from the beginning.

[Purpose of the invention]

An object of the present invention is to provide an apparatus that eliminates the need for prior analysis, which was essential in the prior art, and estimates the cause of an abnormality by inference by providing knowledge of the structure and behavior of the system.

[Summary of the invention]

The inference mechanism that is the core of the present invention will be explained below. Now, consider a simple example in which knowledge is described by a combination of a conditional part (if) and a conclusion part (then).

Example 1 Knowledge i f p q (1
) (If P holds, then q holds) Observation data 5 (2) (For simplicity of description, write the negation of q as 艮) Example 2 Knowledge if ((a xl△(b x)) (c x
) (3) Observation data (bq>
(4) (AND is written as △ for ease of description) Example 3 Knowledge 1f ((ax)/\(b x)) (Cx)
(5) Observation data (CQ), (b ql
(6) In the above three examples, the conclusion obtained from knowledge and observation data is Example 1 p (7) Example 2 1f(a q)(e (1) (8) Example a
(aq) (9) It is. There is a ``derivation principle'' often used in the field of artificial intelligence that provides a means to mechanically perform such inferences. This converts all knowledge into the formula (1
The principle is to convert the two expressions into expressions that are combined using OR (described for simplicity) as in 0), and to combine the two expressions by removing mutually contradictory items into one expression.

ao a2...a,
Q (For example, if the first knowledge is expressed by the formula αυ and the second knowledge is expressed by the formula circle, the conclusion obtained is expressed by the formula (
It will look like 1st stage.

PIVP2...P+-1 几p++ t...R, (
11) QI VQ2 V・VQ+-tydozoQ++x・
・・Q, αTsu Puri P2 V・VP I-IVP
+ ++...VP, Qt VQ2...QJ-1Q+
+ t...Q, 03 On the other hand, all knowledge can be converted into an expression combined with p and OR by repeatedly applying the following 11 rules.

Specifically applying the derivation principle to each of the three examples above, ↓ If (a q) (Cq) Example 3 (ax) (b X) (c x) (c ql□−□
−□■■−−−−−Chi−□−■□□−−１−１１□□□
Shizukua-■■--1■■■11□--■1■■■■1--
□−□□□□□,.

(a q) Then, the conclusions of equations (7), (8), and (9) are obtained. Example 2
Inferences like Example 1 and 3 cannot be obtained using forward or backward reasoning methods, which are extensions of so-called syllogisms.

The feature of the present invention is that the inference mechanism based on the above-mentioned derivation principle is applied to the estimation of the cause of the abnormality.

[Embodiments of the invention]

Hereinafter, the method of estimating the cause of an abnormality according to the present invention will be explained in detail using examples. Consider the electrical circuit shown in Figure 1 as a system to be diagnosed. This embodiment is a system with three input points and two output points, and the number of components is five. Component 1.2.
3 is a multiplier, and component 4.5 is an adder. Now, if the value 1.1.3 is input to the three inputs A, B, and C, the values of the outputs G and H will be 4.6 if the circuit is normal. Suppose that in this circuit, the value of one of the outputs, G, becomes 2. That is, suppose that the value that should be 4 becomes 2, and an abnormality is detected here.

A human would make the following inference.

1) It is unlikely that everything is abnormal and only the output G value became 2 by chance. Therefore, the abnormality is component 1~
I think it occurred in one of the 5 cases.

21 In order for G to be abnormal, component 1.2.4
One of these must be abnormal.

3) Since the candidates for the abnormal component are 1.2.4, 1) the components 3 and 5 are normal.

4) Since H is normal and components 3 and 5 are normal, component 2 is normal.

5) Therefore, the number of components with a possibility of abnormality is 1.4.
It is narrowed down to two.

6) It is not possible to determine which of component 1 or 4 is abnormal from observation data alone. So we need to do a test.

7) If component 1 is abnormal, component 4 is normal. Since the second human power E of component 4 is 3 and the output G is 2, the first human power should be -1.

8) Therefore, if the second human power E of component 4 is set to 4, the output G should be 3.

9) In order to set the second human power E of component 4 to 4 and not change the relationship between the input and output of component 1, which is assumed to be abnormal, the value of input C should be set to 4.

10) That is, set the value of input C to 4 and test whether the value of output G becomes 3. If the value of G is 3
If not, the assumption 7) is incorrect and the abnormal component is identified as 4.

The above is the process of human reasoning. Here, assume that the value of the output G is 5 as a result of the 10p test.

This means that input A=1. B=1. Correct G for C=4
is the value of However, the above reasoning leads to the conclusion that the component 4 is abnormal.

The conventional method is to organize the knowledge obtained by humans through reasoning and use it for diagnosis, but the purpose of the present invention is to provide a device that automatically performs such reasoning itself. be.

Hereinafter, in the example of FIG. 1, a method for implementing the above inference using the derivation principle will be described.

Describe the necessary knowledge by dividing it into system structure and system behavior.

System structure (A and B are connected by ConnA Bl, and the values of A and B are promised to be equal. Also, component 1.43 is Ml, M, ,M, and components 4 and 5 are Name it AI+A2.

Conn (Out Ml I"tAt)
Q5Conn (Out Ml In2A
1) αeConn (Out Ml
In1A2) (LηConn (
Out Ml In2A2) The α barrel formula α9 states that the output of Ml is equal to the first human power of A1. The formula αQ-bet is similar.

System behavior if (Mr△I't (MIl=B△In2 (M
l )=AAD=AXB 1(Out (Ml 1=D
) α■(9) 'f(A1△I'': (At)=D△Ink (A+
) = EAG = D + E) (Out (AI) - G)
(2t)' f (I to At
n: (At)=E△Qut (h胛G, AJ)=G-
E1(Int'(AI)=D)
elD)if
(Int (X)=”AI n, (X)=β△Out
”(x) = r△In ns (X) = α△In1(X)
=β)(Out'(x)-γ)
2) In 2H, component M is normal, the value of the first human force is B1, the value of the second human force is A, and the product of A and B is D, then MI
It is written that the output of is D.

Formula ■, e]) is a constituent person! It describes the behavior of
In particular, Shikikan shows that if the second manpower and output are determined, the first manpower can be determined. Component M2Mm, A
The behavior of I can be similarly described.

) is a necessary condition for 7) of the human reasoning process to hold true, and it is assumed that the relationship between input and output does not change during reasoning.

Using the following 2α4 rule, convert 20 to 4 into an expression combining them with OR.

(10) MrIn+' (Mt 1=BVInz' (Mt)
=AVD=A x13Out (Mt)=D
nA1■I
n+ (A1) -DIn2 (AI)=EVG=D+
E■Qut (At)-G
HA! ■IQI (AI) =FVOut'
(At)=GVD-G-E■ Resistance (At)=D
(c) Once the knowledge is organized in the form described above, the above-mentioned derivation principle can be applied. Below, of the human reasoning processes 1) to 10), 2) and 6
) to 10) will be explained in detail.

First, we start from the fact that Out''(, At) is not 4 as an observation of an anomaly.

In this way, you can choose A1 or M based only on the knowledge about structure and behavior.
It is concluded that either l or Ml is abnormal.

When it comes to test generation, we start with knowledge that describes the behavior of the AI, assuming it is correct.

(12) AtV deaf〉KanoDV bileary stop VG-t=;l):+,EV
OLIt' (At)=GGSi1+EQut (A
ll” G (13) In the last expression of formula (c), C is chosen to be 4, and formula α4
Using the rule, If(A1(th(Ins'(Mt) = 1△In
z (Mt) = 1△I” z (Ms) = 1@
△Ir1z (Mx)-41
(Out(At)=3)))
@ is obtained.

This means that if AI is normal, A = 1 in Figure 1.
, B=1, C! It shows that when e: 4, G=3, which exactly matches the conclusion of the human reasoning process 10).

The case where the abnormality cause estimation method of the present invention is applied to the diagnosis of an electric circuit, which is one of the embodiments, has been described above.

As can be seen from the above specific explanation, the target system is not limited to electric circuits. Any system may be used as long as it has - or more observation points and consists of - or more components.

For example, if part of the water supply system of a boiling water reactor is considered as the target system, the system will be as shown in FIG.

The figure shows a system consisting of seven components, whose inputs are a reactor water level signal, feed water flow signal J, and main steam flow signal K, and whose output is the outlet flow rate of a turbine-driven feed water pump. Here, 6 is a manual control system, 7 is a 1-element control system, 8 is a 3-element control system, 9 is a 1-element/3-element selection switch, and 10 is an automatic/
The manual selection switch 11 is a control valve, and the reference numeral 12 is a turbine-driven water pump.

Suppose that the outlet flow rate becomes abnormal during operation in the three-element control mode. At this time, in the abnormality cause estimation method of the present invention, 3
Element control system 8. It is inferred that either the switch 9, the 10° control valve 11, or the turbine-driven pump 12 is abnormal,
Next, if you switch to the 1-factor control system and the outlet flow rate is abnormal, check that the 3-element control system 8 is not abnormal, then switch to the manual control system and if the outlet flow rate is abnormal, the 1-element and 3-element control system 8 are abnormal. It is inferred that the element selection switch 9 is not abnormal.

(15) The above has mainly explained the method of estimating the cause of an abnormality, and the apparatus for estimating the cause of an abnormality of the present invention will be explained below with reference to FIG.

That is, observation data is taken in from the detector 13 of the target system, and the abnormality detection unit 14 detects whether the observation data is abnormal. If an abnormality is detected, the normal value estimation unit 15 estimates the value that the observed data should originally indicate.

For this estimation, knowledge 19 regarding the structure of the target system and knowledge 21 regarding the behavior stored in the knowledge storage unit 20 are used. Next, based on the fact that the abnormality detection data is not a normal value, the abnormal component is selected by the abnormality candidate selection section 16 using the knowledge 19.21 of the knowledge storage section 20. Among the constituent elements whose inputs and outputs are observed, the behavior knowledge 21 of the knowledge storage unit 20 is used to determine whether there is a contradiction between the input and output, and the abnormal constituent elements are narrowed down. If the abnormal component cannot be narrowed down to one, the test automatic generation section 18 automatically generates a test using the knowledge@19, 21 of the knowledge storage section 20. These results are displayed in the result display section 22.
(16) is displayed.

If each block of the above-mentioned abnormality cause estimating device corresponds to the diagnosis of the above-mentioned electric circuit, which is one of the embodiments, the result will be as follows.

Signals A, B, C, G, and H are detected by the detector 13,
The abnormality detection unit 14 determines that the signal G is abnormal. In general plant systems, alarms and the like occur here. The content of the knowledge storage unit 20 is that the structure-related knowledge 19 is the formula 0.
Knowledge 21 regarding behavior is added to S-(Is, which corresponds to equation ).The input/output contradiction detection unit 17 is not necessary in this electric circuit example, but for example, if signal R, E, and F are also observed, looking at D, E, and G, it is possible to identify the constituent elements. This function corresponds to the case where 4 is determined to be abnormal.

Finally, the automatic test generation unit 18 performs the inference as shown in the equation.

〔Effect of the invention〕

As described above, according to the present invention, by simply providing knowledge regarding the (17) structure and behavior of the target system, if an abnormality is observed, the inference that a human would make is automatically performed, so that it is possible to prevent failures in advance. No analysis is required, and changes in the combination of system components can be easily accommodated. Furthermore, since the present invention allows all logically derivable conclusions to be drawn, there is no possibility of mistakes that humans might make. moreover,
Since no prior analysis is required, a large amount of effort required for analysis is no longer required.

[Brief explanation of drawings]

Figure 1 is a configuration diagram of an electric circuit taken up as one of the target systems, Figure 2 is a configuration diagram of a part of the water supply system of a boiling water reactor taken up as one of the target systems, and Figure 3 is a diagram of this book. FIG. 1 is a configuration diagram of an embodiment of the invention. 13... Detector, 14... Abnormality detection section, 15...
Normal value estimation section, 16... Abnormality candidate selection section, 17...
Input/output contradiction detection unit, 18...Test automatic generation unit, 19
...Structure-related knowledge storage section, 20.. Knowledge storage section, 21.. Behavior-related knowledge storage section, 22.. Result display section. Agent Patent Attorney Akio Takahashi (18)

Claims (1)

[Claims] 1. In a system that has two or more observation points and consists of two or more components, means for detecting observable data of the system, and based on the data detected by the detection means A means for detecting abnormal data, a means for estimating the normal value that the data detected as abnormal should originally take,
An abnormality cause estimation device comprising: means for estimating candidates for constituent elements considered to be abnormal; and means for generating a test method for narrowing down the candidates.