JP3485202B2 - Failure diagnosis device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [0001] The present invention relates to a so-called expert
The system is used to lead to malfunctions in vehicles, aircraft, etc.
The present invention relates to a failure diagnosis device for searching for the cause of a failure. [0002] 2. Description of the Related Art In recent years, in various fields such as medicine, architecture, and chemistry,
Extracts that use computer to help solve problems
Part system is adopted. This expert
Stem compiles expert knowledge in a particular field.
Computer and use it to address complex issues with experts.
A system that can be solved at an equivalent level
It is. Conventionally, this expert system has been used in vehicles.
For example, Japanese Patent Application Laid-Open No. Sho 62
As disclosed in Japanese Patent Publication No.
The root cause of the failure that is causing the phenomenon
(Fault location) has knowledge data expressed as a set of rules
Reasoning using knowledge data that stores past cases
Is known. In this case, in order to obtain a highly accurate inference result,
Is required to ask questions such as the status of occurrence of failure phenomena and inspection instructions
If this inquiry is not carried out smoothly,
For example, Japanese Patent Application Laid-Open No. 3-185251
In the official gazette, questions that cannot be answered by the user
Put on the pending stack and check the success or failure of the event by the question
A technique is disclosed in which the next question is put on hold. [0005] However, the above problem is not solved.
As a result of consultation, correct answers may not always be obtained.
In addition, if the interview is for the user,
When the user gives an incorrect answer due to an illusion or the like
There is. [0006] Therefore, in such a case, it is inferred that the result of the medical inquiry is.
Theory is inconsistent, and the inference
The diagnosis stops due to a limit loop, inability to make inferences, etc.
I will. The present invention has been made in view of such circumstances.
If an incorrect interview result is entered by the user
Fault diagnosis equipment that can continue the diagnosis
The purpose is to provide a device. [0008] [MEANS FOR SOLVING THE PROBLEMS] To achieve the above object
The failure diagnosis device according to the present invention provides knowledge data necessary for failure diagnosis.
A knowledge base unit for storing data, and utilizing the knowledge data.
To determine the cause of the failure corresponding to the failure phenomenon entered by the operator.
Inference mechanism for inferring and generating a conclusion hypothesis
In the diagnostic device, the inference mechanism is at least
In order to verify the conclusion hypothesis, the
And conduct an inquiry about the problem
If the answer contradicts the above conclusion hypothesis,Answer the questionnaire
Modified to be consistent with the hypothesisTo do re-inference
Features. [0009] [Operation] In the failure diagnosis apparatus according to the present invention, an operator enters.
The stored failures are stored in the knowledge base
Infer the cause of failure using knowledge data
Generated and conclusions
When verifying the hypothesis, the questionnaire answer entered by the operator is concluded.
When contradicting the hypothesisThe answer to the questionnaire is inconsistent with the conclusion hypothesis
Fix it so that there is noTo continue the inference. [0010] Embodiments of the present invention will be described below with reference to the drawings.
I do. In this embodiment, the fuel system of the aircraft is
Explanation while describing the condition phenomenon as an example of a failure diagnosis target
I do. As shown in FIG. 8, a fault diagnosis according to this embodiment is performed.
The cutting device A includes an apparatus main body 1 and an input pen 2.
In addition, information is input to the device main body 1 using the input pen 2 described above.
A touch screen 1a for inputting is provided. [0013] The failure diagnosis apparatus A is used in an airport maintenance department or the like.
That will be deployed,The maintenance person who is the operatorThe above input pen
2 using the touch screen 1a
The built-in computer may cause a malfunction.
Or to search for the fault location,
The necessary information is displayed on the touch screen 1a and arranged.
Support staff. As shown in FIG. 1, built in the apparatus main body 1.
Computers have a function to execute fault diagnosis.
Window processing unit 12, system control unit 13, inference mechanism unit
14, technical information collection unit 15, knowledge base unit 16,
The recording data section 17 is configured. In addition, the touch
The touch screen 1a includes an input unit 11a and a display unit 11b.
It is configured. The window processing unit 12 includes pre-processing and
Character recognition unit 12a and operation input unit 12b
And a display control unit 12c as post-processing is provided.
You. In the character recognition section 12a, the touch screen
Window for pen input displayed on the screen 1a (see FIG. 11).
Phenomena, etc. entered by the maintenance staff with the input pen 2
Converts handwritten characters to character codes
Is output to the system control unit 13. Operation input section
At 12b, the graphic displayed on the touch screen 1a is displayed.
Fix or menu selected with input pen 2
At the time, the commands corresponding to the
13 is output. In addition, the display control unit 12c
Based on the signal output from the system control unit 13,
The display unit 11b displays characters, graphics, and the like. The system control unit 13 includes a man machine
Interface control unit 13a and operation mode control unit
13b and a system management unit 13c are provided.
In the man-machine interface control unit 13a,
Execution processing according to commands from window processing unit 12
I do. Alternatively, the character data is sent to the display control unit 12c.
And graphics data. action mode
The control unit 13b follows the use mode selected by the maintenance staff.
Suspension / resumption of diagnostic processing, printing and ordering of maintenance records
Equipment including peripheral equipment such as data transmission of the management system
Controls the overall operating mode. In the system management unit 13c
Is the whole system, such as system operation status and data management
Manage. The inference mechanism 14 has a character string detection function.
The search unit 14a, the rule-based inference unit 14b, and the case base
Inference unit 14c and inference control unit 14d are provided.
You. In the character string search section 14a, a mechanic inputs
The character string that represents the trouble phenomenon that was
(“.” “,” “Ga” “ha” etc.)
Using the decomposed character string, a knowledge base
Stored in the respective knowledge data storage units 16a to 16c.
Is the same or similar to the above character string
Character string, word unit, or character
Searched in units and stored in each of the knowledge data storage units 16a to 16c
It is totaled for each stored knowledge data. Rule-based inference unit 14b, case-based
The inference unit 14c uses the knowledge data to determine the cause of the failure.
Search for faults by inference and generate hypothesesYou. The hypothesis in the rule-based inference unit 14b is as follows:
Fault tree analysis (fault) provided in the knowledge base unit 16
Tree Analysis; hereinafter abbreviated as "FTA")
Knowledge data stored in the knowledge data storage unit 16b;
And failure mode effect analysis (matrix with confidence);
"FMECA") type knowledge data storage unit 16c
Using the knowledge data stored in the
It is generated by searching for a fault location by inference. The hypothesis in the case-based inference unit 14c
Are the trouble phenomena searched by the character string search unit 14a,
Phenomena having the same or similar phenomena are described in the above knowledge base.
Stored in the diagnosis case type knowledge data storage unit 16a of the
It is generated by searching for knowledge data. Then, the inference control unit 14d sets each of the above
Determine consistency of hypotheses generated by inference units 14b and 14c
And if the hypotheses are consistent, this hypothesis is considered the conclusion hypothesis.
OutputI do. Operators to verify the conclusion hypothesis
Of the maintenance technician,
If necessary, narrow down the content of the interview to
And the contents of the interview are displayed on the touch panel 1a of the device body 1.
Show. Here, a questionnaire was taken
It is a questionnaire, and the answer to the questionnaire of the maintenance person who is the operator is provisional
In case of contradiction, provisional priority is given. The technical information collecting unit 15 includes a data collecting unit 15
a, and a case registration unit 15b. Data collection
In the section 15a, the maintenance personnel input the error
Conditions, inspection points, measured values, etc. were entered during maintenance work.
Collect data. In the case registration unit 15b,
Diagnosis of the results of the inquiry (both successful and unsuccessful cases)
As the diagnostic case type knowledge data storage unit 16a described later,
Register in knowledge data. The knowledge base unit 16 stores diagnosis case type knowledge.
Diagnostic case type knowledge data storage unit 16a for storing data;
FTA type knowledge data storage for storing FTA type knowledge data
Unit 16b and FMEC storing FMECA type knowledge data
A-type knowledge data storage unit 16c and digitized manual data
And an electronic manual data storage unit 16d for storing
Has been established. The above-mentioned diagnosis case type knowledge data is based on past failure sources.
A database that summarizes the results of exploring factors as examples,
As shown in FIG. 12, a failure phenomenon (A
Or the location of the malfunction), the cause of the malfunction and
The actions, defective parts, and the type of knowledge source are stored.
Have been. In addition, as a knowledge source of this diagnosis case data,
Defect record slips, maintenance record slips, interviews with maintenance personnel
-And so on. As shown in FIG.
Records the “defects and inspection points” discovered by the crew.
And a column for recording the `` actions '' taken by the maintenance personnel.
Have been killed. The maintenance record sheet includes the “Disabled
Phenomena "and the department that performed the work
“Defective parts”, “failure status” and the like are recorded. An interview was conducted with the maintenance staff.
Know-how on maintenance procedures and troubleshooting that are not documented
These are collected and reflected in diagnosis cases. Recorded in the FTA type knowledge data storage section 16b
The remembered FTA type knowledge data consists of design materials and skilled
Analyze the maintenance staff's experience, etc.
Analyzed logically for each component system and expressed as a set of rules
As shown in FIG. 15, rules for each component
(Rules 1 to 6 in the figure)
They are linked in a chain via hypotheses. For example, le
Rule 2 assumes that "booster / pump abnormality" is an intermediate hypothesis.
The conclusion hypothesis (fuel leakage etc.) is shown, and Rule 4
Is the conclusion hypothesis when “fuel leakage” is the intermediate hypothesis (tightening)
Failure, pump internal failure, etc.). On the other hand, FMECA type knowledge data is
It is displayed in a box-shaped table format, and shows the experience and
Analyze the reliability information of the
In this way, the failure phenomenon and the cause of failure are
In the form of a table as a so-called FMECA table,
In the column of failure phenomena, failure of parts or system
The mode, the effect of the fault, and the method of detecting the fault are described.
I have. In addition, certainty is clearly indicated inside each grid.
ing. This confidence is based on the MTBF of the part or system.
Average failure interval) or the same failure phenomenon
The number of failure causes or the number of actual failures
It is set in consideration of
The depth of the causal relationship between the phenomenon and the cause of the failure is shown. Also, the digitized manual data is
Data for medical interviews such as the occurrence status, inspection items of component parts, etc.
Text that shows the procedure for replacing or assembling components,
Graphics data, for example, as shown in FIG.
Thus, the touch screen 1a of the device main body 1
The booster pump graphic and the booster pump
Data such as the inspection procedure of the
And display it in the window. The operation record data section 17 contains
Storage data storage unit 17a and work progress temporary storage unit 17b
The maintenance record data storage unit 17a
Then, the measures for the failure phenomena, inspection results, and trial run
The maintenance result such as the situation is stored for each case. In addition,
In the work progress temporary storage unit 17b, for example, the
Suspended to order parts, then resumed diagnostics
In such cases, continue from the interrupted maintenance work.
In order to be able to
The maintenance progress or maintenance status is sequentially stored. Next, a failure diagnosis procedure will be described with reference to FIGS.
This will be described according to the flowchart of FIG. In this embodiment,
A description will be given by exemplifying a malfunction of the fuel system of the aircraft. This failure diagnosis is performed by the inference processing routine shown in FIGS.
Is performed according to the routine. The power switch of the failure diagnosis device A is turned on.
9 on the touch screen 1a of the apparatus main body 1.
The use mode selection screen is displayed asBe an operator
The maintenance staff enters `` Start diagnosis '' from the displayed menu.
When selected (pointed) with the force pen 2, it is shown in FIG.
Window for inputting the failure phenomenon
Pen input for inputting the phenomenon of
The window is displayed and the inference process is started. In step S1, the handed-over defect record
Aviation obtained by referring to forms such as vote 21 (see FIG. 10)
Necessary information such as the operational state of the aircraft system, crew (user)
Trouble phenomena (symptoms) reported from etc. or described later
Answer to the inquiry displayed in step S14,
The maintenance staff uses the input pen 2 to touch the touch screen 1a.
To enter. It should be noted that the treatment of the defect record slip 21 shown in FIG.
After maintenance is completed by the maintenance staff,
Maintenance record data such as driving conditions can be written (Fig.
28). On the other hand, in step S2, a later-described step is performed.
In response to an inspection instruction based on the medical inquiry displayed in step S14.
Symptoms such as measured values obtained by star connection
You. Then, an error on the touch screen 1a is detected.
Complete the necessary items in the condition phenomena input window, etc.
And touch the `` End input '' button displayed on the screen.
When the user points at step 2, the entered information (year, month, day,
(Mission, contents, etc.) as input data in step S3.
Stored in moly. Next, in step S4, the following steps will be described.
In S14The conclusions made to the operator
And conclusion of the questionnaire about the problemContradiction between
It is determined whether or not. this isConclusionTo verify the hypothesis
Interview resultsConclusionIf the hypothesis contradicts, the inference stops and
This is a measure to prevent
If there is a contradiction, the process jumps to step S6.
In step S5, the response to the interview is corrected (details will be described later).
To step S6). At the time of the first execution of the routine
Have not yet obtained a hypothesis and need to test the hypothesis
Jump from step S4 to step S6
I do.Also, FTA inference is applied as a hypothesis.
Therefore, in the present embodiment, the conclusion
The content of the interview related to FTA inference
However, the content of the interview is not limited to this. In step S6, the data input in step S1 is entered.
The trouble phenomena stored in step S3 and
Data stored in the data storage units 16a to 16c
Is calculated. The calculation of the similarity is performed by the character string search unit 14a.
This is performed, for example, in the following procedure. (1) First, each of the knowledge data storage units 16a
From which the similarity is calculated from each of the knowledge data
Extract each character string (sentence). Note that each of these strings
Are stored in the respective knowledge data storage units 16a to 16c.
ID (self) for associating with the knowledge data
Certificate) number. (2) Next, the sentence to which this ID number is assigned
String, string, word, and character units
Each knowledge data is stored in the working memory for each knowledge data.
To pay. It is stored in each working memory.
The “decomposition character string” has the same ID number as that before decomposition.
Has been given. Here, a "string" is a character string
"Word" is a blank character, registered in advance
Separating characters (“.”, “,”, “Ga”, “wa”
)), For example, the string "Engine does not start"
Is disassembled into "engine" and "not started". A character is defined as a character string that is divided into a predetermined number of characters.
Refers to disassembly, for example, the string "
"No", "Eng", "N-begin", "No movement", or
Means "E", "Engine", "Start", "No", or
Three characters such as "En", "Jinga", "Start", "Zu" etc.
Decompose each time. (3) Next, store in the working memory
"Ignored word" registered in advance from the "decomposed character string"
Phrase ("but", "I am", etc.) ". (4) After that, the registered "synonyms"
("Failure" is "failure" etc.)) "
Replace "Unsolved string". (5) On the other hand, a trouble phenomenon input by the maintenance staff
Strings of (symptoms) can be changed to strings as in (1) to (4) above.
For each word, word, and character. (6) Then, the character of the broken down phenomenon
"Decomposed" stored in the working memory using the
Character string ". For example, when the engineer enters the
"Doesn't work" in the knowledge data storage units 16a to 16c.
Similar to character strings extracted from stored knowledge data
The similarity is as follows. [0052] (7) Next, a string, word, or
Is the ID number of the data that matches exactly in each character unit
Aggregated for each issue, and the aggregated result is used as the similarity
Output to the corresponding inference units 14b and 14c. In addition, total
String, word, or character
Using a weighting coefficient. As a result, the input fault phenomenon “fuel piece
Decrease "," Consumption only from right without decreasing left tank "
Knowledge data stored in the data storage units 16a to 16c and
Are as follows. Stored in the diagnosis case type knowledge data storage unit 16a
Of similarity calculation in the knowledge data
Diagnosis cases are as shown in FIG.
The similarity has the value shown in FIG. Among them, Case -1048
"Cruising at 3000 ft" "No left tank fuel consumption"
Is the highest similarity, that is, the entered defect phenomenon
This is the case with the most similar symptoms. The FTA type knowledge data storage section 16b stores
In the stored knowledge data, the similarity calculation target
The tree shown in FIG.
A tree composed of a chain of rules with the "event" as the top event
And the similarity of each hypothesis is as follows. 1) Rule 5 “Internal failure of pump itself” →
3, 2) Fuel leak, an intermediate hypothesis between Rule 2 and Rule 4 →
40, 3) Rule 3 "fuel leak" → 40, 4) An intermediate hypothesis between Rule 1 and Rule 2
Abnormal "→ 3, 5) “Transf is an intermediate hypothesis between Rule 1 and Rule 3.
Pump failure "→ 3 6) Rule 1 i) “Fuel shut-off valve error” → 35, ii) "Fuel sub tank abnormality" → 36, iii) "Incorrect indication of fuel indicating system" → 38, It is. On the other hand, the FMECA type knowledge data storage 16
c in the knowledge data stored in c.
Elephant phenomena are as shown in FIG.
Is the highest value of “fuel loss” as shown in FIG.
Is shown. Then, in step S6, each knowledge data
Knowledge data stored in the storage units 16a to 16c;
Once the degree of similarity to the failure phenomenon is calculated, each knowledge
Knowledge data stored in the data storage units 16a to 16c
Based on data, infer the cause of failure by inference and generate a hypothesis
Is performed in parallel in the next steps S7, S8 and S9.
Is performed. That is, in step S7, the diagnosis
Knowledge data stored in the example knowledge data storage unit 16a
Search for the cause of failure by case-based reasoning
Generate a theory. In step S8, FTA-type knowledge
Using the knowledge base stored in the data storage unit 16b
Investigate the cause of failure by rule-based inference and generate a hypothesis
In step S9, the FMECA type knowledge data
Using the knowledge base stored in the data storage unit 16c
Investigate failure causes and generate hypotheses by rule-based reasoning
I do. First, the diagnosis case inference in step S7 is performed.
Generation of a hypothesis will be described. This diagnosis case inference is
This is executed by the example-based inference unit 14c.
Is a hypothesis generation routine based on the diagnosis case inference shown in FIG.
It is performed according to. In step S31, the above inference processing routine
Diagnosis case type knowledge data storage unit calculated in step S6
The similarity for each case of the knowledge data stored in 16a
The hypothesis is narrowed down based on the hypothesis. Refine this hypothesis
Depends on whether the similarity is equal to or greater than a preset threshold.
to decide. For example, if the “threshold value” is 10, then FIG.
2. As shown in Fig. 13, "Case-1048" is a diagnosis case
It is a hypothesis generated by inference. 12 and FIG.
In the knowledge data shown in Fig. 3, cases exceeding the "threshold"
Only one but similar higher than this "threshold"
If there are multiple cases with a degree, they are all diagnosed
This is a hypothesis based on case inference. Thereafter, in step S 32,
The reliability of the hypothesis narrowed down in S31 is calculated and
First generate the interview questions needed to test this hypothesis.
You. The above reliability is determined by other inferences (FTA inference or FM
Eliminate conflicts with hypotheses obtained by ECA inference
The inference processing routine steps
Is calculated by multiplying the similarity calculated in step S6 by a coefficient.
You. In addition, the above interview is based on the passenger who first detected the malfunction.
(Users) and equipment for disassembling equipment
There is something related to staff, for example, a narrowed hypothesis
If it is "Case-1048", the booster pump
Ascertain inspection points and abnormal conditions to identify
Are designated as interview candidates. In this embodiment, as shown in FIG.
The coefficient for calculating the reliability is set to 1. Then, in the step S31, the narrowing down is performed.
Inference control unit
14d, and returns to the inference processing routine. Next, a provisional calculation based on the FTA inference in step S8 is performed.
Generation of the theory will be described. This FTA inference is based on the rules
-This is executed by the base inference unit 14b, and specifically
Is a hypothesis generation routine based on FTA inference shown in FIG.
This is done accordingly. First, in step S41, the above inference processing
Of the FTA type knowledge data calculated in step S6 of the routine
“Fuel” in the knowledge data stored in the storage unit 16b
Consists of a chain of rules with "system failure" as the top event
Based on the similarity of each hypothesis in the tree
Select hypotheses with similarities above the "threshold" and narrow down the hypotheses
Perform For example, if the “threshold value” is 15,
In the FTA type knowledge data of FIG. a) Intermediate hypothesis between rule 2 and rule 4 "fuel leakage" b) Rule 3 “Fuel leakage” c) of rule 1 i) `` Fuel shut-off valve error '' ii) "Fuel sub tank error" iii) `` Incorrect indication of fuel indication system '' Is narrowed down. Then, in step S42, the above steps
Search for a route linking each hypothesis narrowed in S41
I do. In FIG. 15, the top event, “Failure in fuel system,
"" Is the "fault phenomenon" this time.
The roots extending to the left of the figure have a causal relationship to this phenomenon.
, That is, a “hypothesis”. Above
Link the hypothesis selected in step S41 with the above-mentioned problem
As shown in FIG.
It becomes street. Then, in step S43, the above steps
The completeness of the route searched in S42 is calculated. This complete
The degree is the connection (causality) between the failure phenomenon and the cause of the failure.
The higher the value, the higher the value. Of the route searched in FIG.
The completeness is as follows. 1) Route 1 → 50 2) Route 2 → 50 3) Route 3 → 50 4) Route 4 → 30 5) Route 5 → 30 Next, in step S44, based on the completeness,
And whether this completeness is greater than or equal to the preset "threshold"
The hypothesis is narrowed down. For example, the "threshold"
In the case of 20, all of the above routes 1 to 5 are targeted
Become. Next, in step S45, the above steps
Calculate the reliability of the hypotheses narrowed down in S44
At the same time, an inquiry candidate for hypothesis verification is generated. the above
The reliability is set by multiplying the above perfection by a coefficient.
Then, for example, the above “threshold value” is set to 20, and the coefficient is set to 1.
In this case, as shown in FIG.
The theory applies to all of the routes 1 to 5. In addition, interview
The generation of the candidate is performed, for example, as shown in FIG.
Rule for "booster / pump malfunction", which is an intermediate hypothesis
Conclusion Hypothesis of Rule 4 links to [fuel leak]
If so, the location related to the booster / pump abnormality
It is done by specifying. Then, each of these hypotheses is described by “FTA inference
Is output to the inference control unit 14d as the "hypothesis".
Return to the processing routine. Next, step S of the inference processing routine is performed.
The generation of hypotheses by FMECA inference performed at 9
explain. This FMECA inference is the same as the above FTA inference.
As described above, the rule-based inference unit 14b executes
Specifically, generation of a hypothesis by FMECA inference shown in FIG.
It is performed according to a routine. First, in step S51, the FMECA type
Each of the knowledge data stored in the knowledge data storage unit 16c.
From the phenomena, narrow down the phenomena to be diagnosed this time.
U. The narrowing down of this phenomenon is a phenomenon corresponding to the malfunction phenomenon.
Is calculated in step S6 of the above inference processing routine.
Based on the degree, the degree of similarity is determined by a predetermined threshold
This is done by selecting phenomena equal to or greater than the value. For example, "Shiki
When the “value” is 10, the so-called FM shown in FIG.
The phenomena displayed in the ECA table indicate that "fuel decrease"
Restricted to "Fuel depletion" and "Unable to correct fuel tank remaining amount"
(See FIG. 19). Next, in step S52, the above steps
The degree of certainty corresponding to the phenomenon narrowed down in S51 is calculated using the FM
Determined from the ECA table. For example, as shown in FIG.
The FMECA table shows that the confidence level corresponding to “fuel reduction” is 1
6,21,63, confidence level corresponding to "fuel depletion" is 5
0, 31, 46, "Unable to correct fuel tank remaining amount"
Is 40,60. Then, in step S53, the above certainty factors
Select a certainty factor that is equal to or greater than the preset "threshold value" from
Then, narrow down hypotheses. For example, the above "threshold"
Is set to 20, as shown in FIG.
In step S54, this confidence
Select the parts and causes corresponding to the degrees. Next, in step S55, the above steps
Calculate the reliability of the parts or causes selected in S54.
As well as generate interview candidates,
Assuming that the theory is "hypothesis based on FMECA inference", for example,
Output data as shown in 20 to the inference control unit 14d
Then, the process returns to the inference processing routine. In the present embodiment, the above-described reliability is determined by the similarity
Is obtained by multiplying the sum of the values obtained by multiplying
This is shown in the table below. [0080] Then, the above steps S7, S8, S9
When the hypothesis is generated, the inference processing routine shown in FIG.
In step S10, it is determined whether the generated hypotheses conflict.
If there is a conflict, resolve according to the preset criteria
I do. The conflict resolution of this hypothesis is performed by the inference control unit 1 shown in FIG.
4d, specifically, the hypothesis competition of FIG.
This is performed according to the matching cancellation routine. First, in step S61, each of the above steps
The hypotheses generated in S7, S8, and S9 are collected.
The hypotheses generated in steps S7, S8, and S9 above are examples
For example, as shown in the table of FIG.
The ranking is given according to the calculated reliability. Then, in step S62, each of the above steps is performed.
Check the consistency of the hypotheses generated in steps S7, S8, and S9.
From the target, attributes, etc. For example, in FIG.
Rank 1 of hypothesis by case inference and provisional by FMECA inference
According to the theory, the defective part is "Booster pump"
Consistent in some respects, but the order of hypotheses based on diagnostic case inference
Rank 1 and rank 1 of the hypothesis based on FTA inference
Are common in terms of "fuel system failure"
However, defective parts are competing. Similarly, FTA inference
And the hypothesis based on FMECA inference
Are competing for defective parts. Then, when the operation proceeds to step S63, the step
The consistency of the hypothesis calculated in step S62 is
To determine whether there is consistency.
Perform settings. In the present embodiment, the “consistency criterion” is
All the defects with the highest reliability of the hypothesis in inference match
To comply with this definition.
Judge. As a result, for example, as shown in FIG.
Rank 1 hypotheses generated by each inference are not all the same
At step S64, it is determined that there is no
Proceeding to step S65 from step S64, the
Increments the count value C and proceeds to step S66.
No. In step S66, the above inference frequency measurement
The count value C is compared with the set value N, and when C ≦ N,
It is determined that "inference has not been completed", and the flow advances to step S67. Stay
In step S67, the hypothesis collected in step S61
Create data for re-inference according to certain criteria. This
The criterion is, for example, "The reliability of each hypothesis is the highest
If you define as "select one," in the summary table of FIG.
Is re-estimated based on the hypothesis of rank 1 by FMECA inference.
Rational data is created (see FIG. 22), and this data is
Output with the information of "Inference not completed" and skip the routine.
I can. On the other hand, in step S66, it is determined that C> N.
Rejection, that is, re-inference
If the hypotheses do not match, the process proceeds to step S68,
Resolve conflicts according to preset priority criteria
You. The priority in this embodiment is the most logically
Best hypothesis based on FTA type knowledge data constructed
No hypothesis is generated with this FTA type knowledge data
In case, adopt hypothesis based on FMECA type knowledge data
In addition, a hypothesis is found in both of these knowledge data.
If it is not generated, the temporary
Adopt the theory. Then, according to these priorities,
As a result of exploring the disappeared hypothesis, the information of "
Both are output and the routine exits. In each of the knowledge data,
If the cause of the failure cannot be investigated, step S68
Output information of "hypothesis not satisfied" and "end of inference"
And exit the routine. On the other hand, in step S64, each of the above steps is performed.
When the hypotheses generated in steps S7, S8, and S9 match
Makes this hypothesis a search result and branches to step S69.
After clearing the inference count measurement count value C,
The above search results are output together with the information of "end of inference",
Exit the routine. Then, step S1 of the inference processing routine
Returning to step 1, whether "inference is completed" is determined in step S10.
Judgment based on the information made
Goes to step S12, and if "inference not completed"
In this case, the process returns to step S6. Then, in the above step S11, "inference not completed"
If the process returns to step S6 after the
Failure phenomena stored in memory in S3 and predetermined measurement values
And the like, and the above-mentioned step S10 (step S6
Characters combined with "Re-inference data" created in 7)
Column, and stored in each of the knowledge data storage units 16a to 16c.
The similarity with the existing knowledge data is calculated again. In the data for re-inference shown in FIG.
Fuel loss, left tank not reduced, only from right
Consumption, booster pump, own malfunction
The respective character strings and the respective knowledge data storage units 16
defect phenomena of knowledge data stored in a to 16c,
Comparing defective parts, failure causes, etc., and performing logical operations, etc.
Calculate the similarity. Then, according to the similarity of each knowledge data,
In steps S7, S8, and S9, the re-inference
Investigate the cause of the obstacle. As a result, diagnosis case inference in step S7
Then, even at the time of re-inference, based on the diagnosis case type knowledge data,
The generated hypothesis is not changed.
The inference result is output. On the other hand, the FTA estimation executed in step S8
In theory, as shown in FIG.
From the character string, “Failure in fuel system” and “Booster Po
Is abnormal, and the hypothesis that is the cause of the failure is
The causal relationship with the phenomenon is the strongest in route 5 shown in FIG.
Therefore, as shown in FIG.
Is a high value of 80, and rank 1 of the hypothesis by FTA inference
become. On the other hand, the FMECA inference in step S9
Is a sentence corresponding to "fuel depletion" as shown in FIG.
There is a string, the hypothesis has not changed, and a similar result is output.
It is. As a result, the process is executed in step S10.
In step S64 of the hypothesis conflict resolution routine of FIG.
As shown in the table of FIG.
Since the defective parts are matched, the predefined
Judgment criteria "are satisfied, and in step S64, there is consistency.
Is determined, the flow branches to step S69, and the inference
After clearing the count value C for number measurement,
Output with "diagnosis end" information and exit the routine
You. Thereafter, in step S11, "inference is completed"
When the determination is made, the process proceeds to step S12,InterviewIs there
Judge,InterviewIf there is, branch to step S13.
hand,For the operatorNarrow down the content of the interview. In this interview
ContentThere are malfunction phenomena and inspection items.of
The refinement is, for example, generated according to the table of FIG.
Trouble occurrence, booster pump, fuel shut off
From the questionnaire candidates related to f
What is related to the hypothesis “booster / pump abnormality”
To test this conclusion hypothesis,
The information is used to generate a theory. Then, in step S14, the above steps
The contents of the medical inquiry narrowed down in S13 are
Display on the screen 1a. This inquiry display is performed in step S
When the maintenance person enters an answer using the input pen 2 in step 1,
In step S3, this data is stored, and in step S4
Conclusion of failure phenomena to verify temporaryInquiry check
After the check, in steps S7, S8 and S9, each knowledge data
Inference is performed again based on the data. [0101]For example, To test this conclusion hypothesis,
"The booster / pump stopped operating or fuel leaked.
"Taka"Maintenance technician who is operating the phenomenonConfirm to
if you did this,Maintenance technicians should respond based on information from the user.
There is. But,The user's memory is ambiguous or complicated
"No shutdown and no fuel leakage"
Then, "booster / pump abnormality" was narrowed down as the conclusion hypothesis.
Contradiction will arise with respect to the incorporated FTA inference. Therefore, from step S 4 to step 5
Go ahead and prioritize the most logical FTA inference as correct,
Inquiry of malfunction phenomenonAnswer"Booster pump shutdown
Stop or fuel leak occurs '' and proceed to re-inference
You. In this re-inference, there was no change to the previous hypothesis,
From step 12 to step S13, step S1
At 4, the next inquiry display is performed. Then, the next interview isInspection itemsConsultation about
In the case of, the contents of the interview are shown in FIG.
As shown in FIG.
At times, the contents of work such as related inspection procedures and necessary information are displayed.
Show. Then, return to steps S1 and S2,
Wait for the input of the inspection results. After that, the maintenance staff operates the touch screen 1
Input the inspection result from the input pen 2 or the tester into a
In step S3, this data is stored, and
In S6, the input character string and the previous "input data"
Data based on the character string of "
Is calculated again, and steps S7, S8, S9 and subsequent steps are performed.
Below, based on each knowledge data, infer again,
Investigate the cause. Then, the inference is completed, and the process proceeds to step S11.
Proceeds to step S12 to verify the inference.InterviewBut
If not, the process directly proceeds to step S15, and
Determine if the cause was explored. If the cause of the failure can be found, step S
At 16, for example, as shown in FIG.
The reason is displayed on the touch screen 1a, and the maintenance person
Confirm this content and click the "OK" button displayed on the screen
Is pointed with the input pen 2, and then step S 17
And necessary procedures such as replacement procedures for parts corresponding to the hypothesis
Is displayed. Next, in step S18, the
The result of the action, that is, whether the problem has been resolved
Wait for input. And when this treatment result is input,
Proceeding to step S19, the inquiry result is confirmed. On the other hand, in the above step S15,
If it is determined that the cause of the failure could not be determined,
Steps S16 to S18 are jumped to step S1.
Go to 9. Then, in step S19, the touch
A search result confirmation screen is displayed on the screen 1a. this
The confirmation screen of the inquiry result is, for example, as shown in FIG.
This is done by displaying the history of the inquiry. In addition, according to the search results
Not satisfactory after replacing parts
Next, the mechanic inputs that fact in step S18.
Then, in step S19, on the touch screen 1a
Next, a confirmation screen is displayed to the effect that this inquiry has failed.
In the dialog box, for example, Is displayed to the maintenance staff for confirmation. Then, the maintenance staff understands the registration of the case.
In step S20, first, the FMECA type knowledge
Of the knowledge data stored in the data storage unit 16c
Of the phenomena selected by the search
The confidence (see Fig. 18) according to the results of this failure investigation.
Update. In other words, if this inquiry was successful,
Gives a relatively high degree of confidence and, in the case of failure,
Reduce confidence. Then, in step S21, the search for this time
The result is used as a diagnosis case, and the diagnosis case type knowledge data storage unit 16 is used.
In the knowledge data stored in a of FIG.
Add as "Example-5721". In addition, the result of this research
In the case of a failure, the details are listed in the "Cause and action" column.
"Explanation failed"
It is described. Next, when the operation proceeds to step S22, the current
The details of the inquiry are shared with supplementary information such as the date of diagnosis,
In the maintenance record data storage unit 17a,
End the disconnection. A memory is stored in the maintenance record data storage section 17a.
Data can be retrieved via an external printer.
For example, as shown in FIG.
It can be described and tabulated. As described above, the user's uncertain memory and illusion
Wrong byFailure phenomenonEven if the interview result is entered,
Smooth diagnostic work can be performed without stopping inference
It is possible to further improve the flexibility of the system. The present invention is not limited to the above embodiment.
For example, the knowledge base unit 16 stores the diagnosis case type knowledge data.
Data storage unit 16a and FTA type knowledge data storage unit 16b,
Alternatively, the diagnosis case type knowledge data storage unit 16a and the FMEC
A type knowledge data storage unit 16c or FTA type knowledge
Data storage unit 16b and FMECA type knowledge data storage unit 1
6c.
It may be constituted by knowledge data. [0115] The target equipment for failure diagnosis is limited to aircraft.
Not limited to vehicles such as automobiles, railways, or ships
Is also good. [0116] As described above, according to the present invention,
Estimation using knowledge data stored in the knowledge base
Investigate the cause of failure by theory, generate a hypothesis, and ask
When testing your hypothesis,Failure phenomena from the operatorInterview answer
Is inconsistent with the hypothesis, the hypothesis takes precedence.
Incorrect interview result is input due to uncertain memory or illusion
Inference can continue without stopping
Diagnostic work can be performed smoothly and system
Excellent effects such as improved flexibility of the system
Is obtained.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a functional block diagram of a failure diagnosis device. FIG. 2 is a flowchart showing an inference processing routine. FIG. 3 is a flowchart showing an inference processing routine (continued). FIG. 4 is a hypothesis based on diagnosis case inference. FIG. 5 is a flowchart showing a hypothesis generation routine based on FTA type knowledge base inference. FIG. 6 is a flowchart showing a hypothesis generation routine based on FMECA type knowledge base inference. FIG. 7 is a hypothesis conflict resolution routine. FIG. 8 is an external view of the failure diagnosis device. FIG. 9 is an explanatory diagram showing a use mode selection screen. FIG. 10 is an explanatory diagram showing a defect record slip. FIG. 11 is an explanatory diagram showing a defect phenomenon input screen. FIG. 12 is an explanatory diagram showing diagnosis case type knowledge data. FIG. 13 shows a similarity calculation result in diagnosis case type knowledge base inference. FIG. 14 is an explanatory diagram showing the results of hypothesis generation by diagnosis case type knowledge base reasoning. FIG. 15 is an explanatory diagram showing FTA type knowledge data. FIG. 16 is an explanatory diagram of route search results in FTA type knowledge base reasoning. FIG. 17 is an explanatory diagram showing the result of hypothesis generation by FTA-type knowledge base inference. FIG. 18 is an explanatory diagram of FMECA-type knowledge data. FIG. 19 is an explanatory diagram of similarity calculation results by FMECA-type knowledge base inference. Explanatory diagram showing the result of hypothesis generation by FMECA type knowledge base inference [Fig. 21] Explanatory diagram showing aggregation of hypotheses obtained by each inference [Fig. 22] Explanatory diagram of data for re-inference [Fig. 23] FIG. 24 is an explanatory diagram showing a display screen at the time of a medical consultation. FIG. 25 is an explanatory diagram showing a display screen of inference results and arguments. FIG. 26 is a table for confirming inquiry results. FIG. 27 is an explanatory diagram of diagnosis case type knowledge data in which a new case is registered. FIG. 28 is an explanatory diagram showing a failure record form describing the contents of a treatment. 14 inference mechanism units 14b and 14c inference unit 16 knowledge base unit

────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Yasuo Kagei 1-7-2 Nishi-Shinjuku, Shinjuku-ku, Tokyo Inside Shimizu Tomi Heavy Industries Co., Ltd. (72) Inventor Masaaki Koyama 1-7-1, Nishishinjuku, Shinjuku-ku, Tokyo No. Shimizu Kogyo Co., Ltd. (72) Inventor Kunihiro Abe No. 1-7-2 Nishi Shinjuku, Shinjuku-ku, Tokyo Inside Shimizu Kogyo Co., Ltd. (58) Field surveyed (Int. Cl. ⁷ , DB name) G06N 5 / 00

Claims (1)

(57) [Claims] [Claim 1] A knowledge base unit for storing knowledge data necessary for failure diagnosis, and utilizing the knowledge data, a failure cause corresponding to a failure phenomenon input by an operator. In the fault diagnosis device having an inference mechanism that infers and generates a conclusion hypothesis, the inference mechanism unit at least asks an operator for a defect phenomenon related to the conclusion hypothesis in order to verify the conclusion hypothesis, If the answer to the inquiry is inconsistent with the above conclusion hypothesis ,
A fault diagnosis device , wherein correction is performed so as not to contradict the conclusion hypothesis and re-inference is performed.