JPS63156271A - Trouble diagnosis system - Google Patents

Abstract

PURPOSE:To finally detect troubled component to the condition of an unexperienced trouble by obtaining an observation required from the result of the formation of the hypothesis of the trouble, displaying to a user and reducing the number of the troubles component from the result. CONSTITUTION:When the trouble is generated, for instance, the user applies the condition of the trouble to a trouble hypothesis forming part 22. At this time, at the same time as the input of the condition of the trouble, which part is normal or not is inputted. This is the observed result and the trouble hypothesis forming part 22 uses the operation rule of a knowledge base part 21 from the observed result to obtain the component relating to said trouble. The observation relating to the component concerning the obtained trouble is obtained from the operation rule 24 of the base part 21 by a discrimination observation forming part 23 and displayed. According to this display, for instance, the user executes a required observation and applies to the trouble hypothesis forming part 22. According to the repetition thereof, the troubled component can be detected.

Description

[Detailed Description of the Invention] [Summary] In a fault diagnosis system, only typical faults such as faults that have been observed in advance can be diagnosed. The present invention infers the location of a fault by generating a fault hypothesis and generating discriminative observations based on knowledge regarding the operation of the system, making it possible to diagnose the fault even if an unexpected fault occurs.

[Industrial application field]

The present invention relates to a fault diagnosis method for detecting a faulty component at the time of a fault.

[Traditional technique]

Fault diagnosis is generally performed by a specialist in the faulty device. However, such a system has the problem that an expert must always be assigned, and if one is not assigned, one must wait until an expert for the device comes. Nowadays, with the development of computers, it has become possible to diagnose failures using computers. In this type of computer-based fault diagnosis, a knowledge base of experts is stored in the computer in advance, and when a system failure occurs, this knowledge base is used to infer the cause of the failure from the observed symptoms. be.

The following things are desired for this computer-based fault diagnosis.

■ Being able to describe knowledge in the same form that experts have in their heads.

■Applicable to atypical failure symptoms that have never existed before.

■Dialogue with the user during inference should be natural.

On the other hand, in the past, as shown in FIG. 4, a knowledge base 10 is provided, and rules for guiding faulty parts from observed symptoms are stored as empirical rules 11. When a failure occurs, the inference engine 14 outputs or asks the user the observation 12, and based on the input observation results, the failure part 13 is detected from the empirical rule 11 and displayed, for example.

[Problems that the invention sought to solve] Despite the above-mentioned requirements from ■ to ■, failure parts are detected using empirical rule 11. I know this, but I am unable to describe this knowledge exactly as it is.

B. It cannot be applied to cases of failure symptoms that have never existed before (for example, even if experts know it, it is impossible to describe all the rules in the form of empirical rules).

C. Since the system asks the user questions as necessary, the user cannot freely input the symptoms he/she observed, and the order of questions depends on the order in which the rules are written, so it is not necessarily in a natural order. I had a problem.

In view of the above-mentioned conventional drawbacks, the present invention aims to provide a fault diagnosis method that detects faulty parts even in the case of failure symptoms that have never existed before, and further allows the user to freely input the observed symptoms. purpose.

[Means for solving problems]

FIG. 1 is a functional block diagram of the present invention.

The knowledge base unit 21 performs observations depending on parts using operation rules 24.
In other words, the expected failure observation results corresponding to the parts are stored as (knowledge). The failure hypothesis generation unit 22 adds the input observation results, that is, failure states, and detects failure-possible parts from the operation rules stored in the knowledge base unit 21 based on the observation results.

The discriminative observation generation unit 23 obtains observations that depend on the faulty parts added from the failure hypothesis generation unit 22 from the knowledge base, and outputs them to the user or the like.

[For production]

When a failure occurs, for example, the user adds the failure state to the failure hypothesis generation unit 22. At this time, input which part is normal at the same time as inputting the failure state. This is the observation result, and the failure hypothesis generation unit 22 uses the observation result to find the parts related to the failure using the operating rules of the knowledge base unit 21.

Then, the discriminative observation generation section 23 obtains observations related to the parts related to the obtained failure from the operation rules 24 of the knowledge base section 21 and displays them. Based on this display, the user, for example, makes a required observation and adds it to the failure hypothesis generation section 22. By repeating this process, failed parts can be detected.

〔Example〕

Hereinafter, the present invention will be explained in detail using the drawings.

FIG. 2 is a block diagram of an embodiment of the present invention. The input unit 31 is an input device through which a user 30 using the control system or the like inputs a failure state. When a failure occurs in the control system, the user manually inputs the failure state, that is, the observation result, from the input unit 31. For example, if observations X and Z are obtained as a result, and observation X is defective (X=defective) and observation Z is operating (Z=good), this is input to the input section 31.

The observation results input from the input unit 31 are sent to the failure hypothesis generation unit 32.
join. The failure hypothesis generation unit 32 generates a knowledge base 35 of parts that are expected to be faulty based on the input observation results.
This circuit is obtained from the operating principle 36 of . In addition, A and B-P of the operating principle 36 in the figure represent that observation P depends on the operation of parts A and B, p and c-x represent that observation P depends on part C, and A-YSB→ Z indicates that observations Y and Z depend on parts A and B, respectively.

When X=defective and Z=good, as described above, since Z=good, it is detected that part B is normal based on the operating principle of BZ. Furthermore, since X=defective, it is inferred that observation P or component C is defective. On the other hand, observation P
depends on parts A and B, but part B is detected to be normal than B-Z, Z = good, so when part A is defective, it is inferred that observation X is defective. .

Based on the above logic, it is detected that component A or component C is defective based on the input X=defective and 2=good. The failure hypothesis generation unit 32 outputs this result to the discriminative observation generation unit 33. The discriminative observation generation unit 33 obtains observations from the knowledge base unit 35 to detect good, ie, working, parts in order from the parts for which defects are predicted to occur from the failure hypothesis generation unit 32.
It is output to the display section 34.

In the case of parts A, , C mentioned above, observation Y depends on part A more than A - Y, so if the result of observation Y is obtained,
It is determined that either part A or part C is defective. The discriminative observation generation unit 33 is a circuit that obtains this observation Y, and outputs the obtained result to the display unit 34 for display.

As a result, a message to perform observation Y appears on the display section, so the observation corresponding to the display is performed and added to the input section 31 again. As a result, for example, when observation Y is good, it is determined that component C is defective.

Through the above operations, defective parts can be detected.

Compared to the conventional method, which sequentially performs observations for defects that can be predicted in advance, the present invention stores an operating principle and uses that operating principle to request the necessary observation, so that the observation is directly related to the defective part. There are many things, and unnecessary observations are not required.

Further, the relationship between each observation and the component may be obtained from, for example, the designer who developed it.

Further, the operation will be explained in detail using the operation flowchart of the embodiment of the present invention shown in FIG.

First, when a failure is observed, the user is asked to input any number of normal observations obtained at that time at the same time as the failure observation (Sl). Next, the initial value of the possible failure hypothesis set S is set as a set of all parts (S2).

That is, all the parts on which all observations depend were set as a failure hypothesis set S. Then, for all failure observations The product set is obtained as a new failure hypothesis set S (S3). This process S3 is a process to obtain all the parts related to the failure observation X, and then a process is performed to remove the normal parts obtained from the normal observation Y from the set of possible failure hypotheses S obtained in the above process S3. That is, for all normal observations Y, a set T of normal parts related to the normal observations is derived by tracing the operation rule backwards from the normal observations, and the elements of the normal parts set T are removed from the failure hypothesis set S. (S4). The aforementioned processes 82 to S4 are performed by the failure hypothesis generation section. Incidentally, the operating principle 36 of the knowledge base section is naturally used in the processes S2 to S4.

As described above, the set of possible failure hypotheses S is determined by the process S4. If one component can be extracted by simple failure or observation of the failure, the failed component will be set.

Therefore, after the process $4, a determination S5 is performed to determine whether the number of elements, ie, parts, of the failure possibility hypothesis set S has been determined to be one. If there is one (YES), this process ends. When there is not one (No
), the discriminative observation generation unit 33 operates and performs the following processing.

First, processing S6 is performed to arbitrarily select elements (part A and part B) from the set of possible failure hypotheses. Then, by tracing the operating principle from the part A forward (in the direction of the arrow in Fig. 2), a process S7 for obtaining a set SA of observations related to the part.
I do. The same process is also performed for part B (S8). Through these two processes S7 and S8, observations are obtained correspondingly, but next, although they are included in the set SA,
Obtain observations P that are not included in the set SB, and perform a process of asking the user a question (S9). The set SA contains the set S
Since part B is not included, if the answer to the above question is failure, part B is not dependent on the failure, and in that case (normal), part B is dependent on the failure. Therefore, next, a process (SIO) is performed in which part B is removed from the set of possible failure hypotheses S, and if it is normal, part A is removed. Since one component is determined to be normal in one execution of the above-mentioned processes 86 to S10, the next step is to determine the elements (components) of the set of possible failure hypotheses.
The process S5 of determining whether the number has become one is performed again. By repeating the above-described operations, one failed component can be identified.

Through the above operations, it is possible to find a failed component corresponding to the failure observation.

The operation of the present invention has been described above using examples.

In the present invention, one component is determined for one or more failure observations, but the present invention is not limited to this, and for example, multiple failed components can be determined in the same manner.

〔Effect of the invention〕

As described above, the present invention uses a knowledge-based operating principle to generate failure hypotheses, obtains necessary observations from the results obtained and displays them to the user, and then uses the results again to identify failed parts. By reducing the number of failed parts, the faulty parts are finally determined.According to the present invention, faulty parts can be detected even for failure symptoms that have never existed before, and the user can also freely input the observed symptoms. It is possible to obtain a fault diagnosis method that can perform the following steps.

[Brief explanation of the drawing]

FIG. 1 is a functional block diagram of the present invention, FIG. 2 is a configuration diagram of an embodiment of the invention, FIG. 3 is an operation flowchart of an embodiment of the invention, and FIG. 4 is a system configuration diagram of a conventional system. 21... Knowledge base section, 22... Failure hypothesis generation section, 23... Discrimination observation generation section, 24... Operation rule. Figure 1: Structure of the embodiment of the present invention Figure 2: Functions of the main performance

Claims (1)

[Claims] 1) A knowledge base unit (21) that stores observations that depend on parts as operation rules (24); and an operation rule (24) that is stored in the knowledge base unit (21) from input observation results. a failure hypothesis generation unit (22) that detects failure-prone parts from the failure hypothesis generation unit (22); and a discriminative observation generation unit (23) that obtains observations depending on the failure-probability parts added by the failure hypothesis generation unit (22) from the knowledge base unit (21). A fault diagnosis method characterized by having the following. 2) A patent claim characterized in that the discriminative observation generation unit (23) extracts at least two of the failure-prone components and obtains an observation that depends on one component and does not depend on the other component. The failure diagnosis method described in item 1.