JPH07271598A - Fault diagnostic device - Google Patents

Abstract

PURPOSE:To provide an unified searched result at all times even when a hypothesis is searched by inference based on the plural kinds of knowledge data. CONSTITUTION:In the respective inference parts 14b and 14c of an inference mechanism part 14, the cause of a fault is searched by the inference by utilizing the respective corresponding knowledge data from an inputted faulty phenomenon and the hypothesis is generated. When the hypotheses generated in the respective inference parts 14b and 14c are consistent, the hypothesis is outputted as a conclusion hypothesis. When the respective hypotheses compete, data for re-inference are prepared based on the hypothesis selected from the respective hypotheses corresponding to a fixed reference and the re-inference is performed. Then, when the hypotheses compete even when the re-inference is performed for a prescribed number of times, the conclusion hypothesis is derived from the respective knowledge data corresponding to a priority order set beforehand. The priority order is defined as FTA type knowledge data > FMECA type knowledge data > diagnostic example type knowledge data in the decreasing order of reliability. As the result, the unified conclusion hypothesis is obtained at all times.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a failure diagnosis apparatus for investigating a cause of a failure associated with a defective phenomenon of a vehicle, an aircraft, etc. by inference using a plurality of kinds of knowledge data, and in particular, it is generated based on each knowledge data. If the
The present invention relates to a failure diagnosis device that derives a unified conclusion hypothesis according to a certain standard.

[0002]

2. Description of the Related Art In recent years, an expert system has been adopted in various fields such as medicine, architecture, and chemistry to utilize a computer as a clue for problem solving. This expert system is a system in which the knowledge of a specialist in a certain specific field is input to a computer and it is possible to solve a complicated problem at a level equivalent to that of a specialist.

Conventionally, this expert system has been adopted for vehicle failure diagnosis, for example, as disclosed in JP-A-62-62.
As disclosed in Japanese Patent No. 6846, a defect phenomenon is input, and a fundamental failure cause (fault location) causing the phenomenon is represented by a set of rules as knowledge data or past cases. It is known to make inferences using stored knowledge data.

[0004]

By the way, in such a conventional fault diagnosing device, different types of knowledge data such as rule-type knowledge data or case-type knowledge data are stored in the knowledge base. In this case, since the inference result is obtained for each piece of knowledge data, the probability of success in fault investigation is high.

However, when different inquiry results are derived from the respective knowledge data, the maintenance personnel must separately perform the inspection work according to the respective instructions in order to confirm each inquiry result. Workability is poor.

The present invention has been made in view of the above circumstances, and provides a failure diagnosis apparatus capable of deriving a unified search result even when a failure cause is searched for by using a plurality of knowledge bases. It is an object.

[0007]

In order to achieve the above object, a failure diagnosis apparatus according to the present invention comprises a knowledge base unit for storing knowledge data necessary for failure diagnosis, and a reason for failure by reasoning using this knowledge data. In a failure diagnosis device having an inference mechanism section to be investigated, a diagnostic case type knowledge data storage section for storing past diagnosis cases as knowledge data in the knowledge base section, and a theoretical causal relationship between a failure phenomenon and a failure cause. Failure tree analysis-type knowledge data storage that stores knowledge data that is analyzed as a set of hypotheses and stored in a tree, and failure mode impact analysis that links failure phenomena and failure causes with certainty and organizes and stores them in a matrix A type knowledge data storage unit, and the defect phenomenon that is the same as or similar to the defect phenomenon input to the inference mechanism unit is stored in the diagnosis case type knowledge data storage unit. The knowledge data stored in the fault tree analysis type knowledge data storage unit for the case-based reasoning unit that searches for the fault cause by searching the stored knowledge data and generates the hypothesis, and the fault cause linked to the input failure phenomenon. While using inference to search and generate a hypothesis, a defect phenomenon that is the same as or similar to the input defect phenomenon is searched from the knowledge data stored in the failure mode effect analysis type knowledge data storage unit and searched. In the case where the rule-based reasoning unit that searches for the cause of failure and generates a hypothesis according to the certainty that connects the failure phenomenon and the cause of failure and the hypothesis generated by each of the above-mentioned reasoning units is judged and each hypothesis matches Outputs this hypothesis as an inference result. On the other hand, if the hypotheses generated by the above inference units conflict, at least one of the competing hypotheses is selected and this selected If each of the inference units executes re-inference based on the above theory and the above-mentioned trouble phenomenon and the re-inference is executed a plurality of times and the conflict of hypotheses is not resolved, the failure tree analysis type knowledge data storage unit is stored. The hypothesis generated based on the stored knowledge data is output as a search result. On the other hand, when the hypothesis is not generated by the knowledge data stored in the fault tree analysis type knowledge data storage unit, the above failure mode is output. If the hypothesis generated based on the knowledge data stored in the impact analysis type knowledge data storage unit is output as a search result, and if the hypothesis is not generated in the knowledge data in the failure mode influence analysis type knowledge data storage unit, And an inference control unit that outputs a hypothesis generated based on the knowledge data stored in the diagnostic case type knowledge data storage unit as a conclusion hypothesis. It

[0008]

[Operation] In the failure diagnosis apparatus according to the present invention, when a maintenance worker inputs information such as a failure phenomenon, the case-based reasoning unit inputs the information from the knowledge data stored in the diagnosis case type knowledge data storage unit. The past diagnosis cases corresponding to the trouble phenomenon are selected, the cause of the failure is searched, and the hypothesis is generated. In addition, the rule-based reasoning section selects a phenomenon and a cause of failure corresponding to the input failure phenomenon from the knowledge data stored in the failure tree analysis type knowledge data storage section, and then selects a set of this phenomenon or cause. A route connecting the rules to be constructed and the above-mentioned trouble phenomenon, that is, a chain of rules is searched, a cause of failure is searched and a hypothesis is generated according to the searched route, and stored in the failure mode influence analysis type knowledge data storage unit. A phenomenon corresponding to the above-mentioned trouble phenomenon or the like is selected from existing knowledge data, and the cause of failure is sought and a hypothesis is generated according to the certainty factor connecting this phenomenon to the cause and the failure location.

Then, the inference control unit determines whether or not the hypotheses generated by the case-based reasoning unit and the rule-based reasoning unit conflict with each other. If there is a conflict, at least one of the competing hypotheses is determined. Based on the input failure phenomenon, the inference units are made to perform inference again to investigate the cause of the failure. If the hypotheses match, this hypothesis is output as the inference result. On the other hand, if the hypotheses do not match even after the re-inference is executed multiple times, the hypothesis generated based on the knowledge data stored in the fault tree analysis type knowledge data storage unit is preferentially selected as the conclusion hypothesis.

On the other hand, when the hypothesis is not generated by the inference based on the knowledge data stored in the failure tree analysis type knowledge data storage unit, the knowledge stored in the failure mode influence analysis type knowledge data storage unit is used. The hypothesis generated by inference using data is selected as the conclusion hypothesis.

Further, the knowledge data stored in the diagnostic case type knowledge data storage unit is provided only when no hypothesis is generated by the inference based on the knowledge data stored in the failure mode influence analysis type knowledge data storage unit. The hypothesis generated by inference using is adopted as the conclusion hypothesis.

[0012]

Embodiments of the present invention will be described below with reference to the drawings.

In the present embodiment, a description will be given while describing a malfunction phenomenon of a fuel system of an aircraft as an example of a failure diagnosis target.

As shown in FIG. 8, the failure diagnosing device A according to the present embodiment comprises a device body 1 and an input pen 2, and information is input to the device body 1 by using the input pen 2. The touch screen 1a is provided.

The failure diagnosis device A is installed in an airport maintenance department or the like, and when a trouble phenomenon is input to the touch screen 1a using the input pen 2, the built-in computer causes or troubles the trouble phenomenon. Investigate the location and display necessary information such as the search result and the inspection procedure on the touch screen 1a to support the maintenance staff.

As shown in FIG. 1, the computer incorporated in the apparatus main body 1 has a window processing section 12, a system control section 13, an inference mechanism section 14, a technical information collecting section 15, and a knowledge as a function for executing a failure diagnosis. A base section 16 and an operation record data section 17 are configured. The touch screen 1a has an input unit 11a and a display unit 11b.

The window processing section 12 is provided with a character recognition section 12a and an operation input section 12b as preprocessing, and a display control section 12c as postprocessing.

In the character recognition unit 12a, handwritten characters such as a trouble phenomenon input by the maintenance staff with the input pen 2 in the pen input window (see FIG. 10) displayed on the touch screen 1a are converted into character codes, The commands are output to the system control unit 13. When the graphics or menu displayed on the touch screen 1a is selected with the input pen 2 in the operation input unit 12b, commands corresponding to the graphics are output to the system control unit 13. Further, the display control unit 12c causes the display unit 11b to display characters, graphics and the like based on the signal output from the system control unit 13.

The system control unit 13 is provided with a man-machine interface control unit 13a, an operation mode control unit 13b, and a system management unit 13c.
The man-machine interface control section 13a performs execution processing according to the commands from the window processing section 12. Alternatively, it outputs character data, graphics data, etc. to the display control unit 12c. The operation mode control unit 13b controls the operation mode of the entire device including peripheral devices such as interruption / restart of diagnostic processing, printing of maintenance records, and data transfer of the order management system according to the use mode selected by the maintenance personnel. To do. The system management unit 13c manages the entire system such as the operating state of the system and data management.

Further, the inference mechanism section 14 is provided with a character string retrieval section 14a, a rule-based inference section 14b, a case-based inference section 14c, and an inference control section 14d.

In the character string retrieving section 14a, the character string representing the trouble phenomenon inputted by the maintenance staff is decomposed by using pre-registered separating characters ("."",""Ga""wa" etc.), Using this decomposed character string, a character string that is the same as or similar to the above character string is extracted from the knowledge data stored in each of the knowledge data storage units 16a to 16c of the knowledge base unit 16 described later as a character string unit, The search is performed in word units or character units, and the knowledge data stored in each of the knowledge data storage units 16a to 16c is totaled.

In the rule-based inference unit 14b, the knowledge data stored in the failure tree analysis (fault tree analysis; hereinafter abbreviated as "FTA") type knowledge data storage unit 16b provided in the knowledge base unit 16, And failure mode impact analysis (matrix with confidence factor; hereafter "FME
(Hereinafter abbreviated as “CA”) knowledge data stored in the type knowledge data storage unit 16c is used to infer the cause of failure or the location of the failure by inference to generate a hypothesis (details will be described later in a flowchart).

In the case-based reasoning unit 14c, a phenomenon having the same or similar phenomenon as the defect phenomenon retrieved by the character string retrieval unit 14a is stored in the diagnostic case type knowledge data storage unit 16a of the knowledge base unit 16. It searches from existing knowledge data and generates a hypothesis of the cause or location of the failure.

Then, the inference control section 14d judges the consistency of the hypotheses generated by the inference sections 14b and 14c, and if the hypotheses are consistent, this hypothesis is output as a conclusion hypothesis.

The technical information collector 15 is a data collector 15
a, the case registration unit 15b. The data collection unit 15a collects data input during maintenance work, such as defective items, inspection points, and measurement values, which the maintenance personnel input to the device during failure investigation. The case registration unit 15b registers the results (both success cases and failure cases) searched for in the failure diagnosis as diagnosis cases in the knowledge data of the diagnosis case type knowledge data storage unit 16a described later.

The knowledge base unit 16 includes a diagnostic case type knowledge data storage unit 16a for storing diagnostic case type knowledge data, an FTA type knowledge data storage unit 16b for storing FTA type knowledge data, and an FMEC for storing FMECA type knowledge data.
It is composed of an A-type knowledge data storage unit 16c and an electronic manual data storage unit 16d that stores electronic manual data.

The above-mentioned diagnostic case type knowledge data is a data base that summarizes the search results of past failure causes as examples,
As shown in FIG. 11, a failure phenomenon (or a location where the failure has occurred), a failure cause of the failure and its remedy, a defective component, a type of knowledge source, and the like are stored for each diagnostic case. The knowledge source of the diagnosis case data includes a defect record slip, a maintenance record slip, and an interview with a maintenance worker.

As shown in FIG. 9, the defect record sheet is provided with a column for recording the "defects and inspection points" found by the occupant and a column for recording the "procedures" performed by the maintenance personnel. There is. The maintenance record sheet records "defective phenomena" that occurred at the time of inspection, as well as "defective parts" and "fault status" entered in the department that performed the work.

The interview is a collection of unwritten documented maintenance procedures, know-how for troubleshooting, etc. conducted for maintenance personnel and reflected in the diagnosis cases.

The FTA-type knowledge data stored in the FTA-type knowledge data storage unit 16b is designed to analyze design materials and experience of experienced maintenance personnel, and to logically identify failure phenomena and failure causes for each system of component parts. 14 and is expressed by a set of rules. As shown in FIG. 14, each component is classified into rules (rules 1 to 6 in the figure), and each rule is linked in a chain through an intermediate hypothesis. Has been. For example, rule 2 shows the conclusion hypothesis (fuel leakage, etc.) when "Booster pump abnormality" is the intermediate hypothesis, and rule 4 is the conclusion hypothesis (fueling leak) is the intermediate hypothesis (tightening) (Defect, internal defect of pump, etc.) is indicated.

On the other hand, the FMECA type knowledge data is expressed in a matrix form, and the experience of skilled maintenance personnel and the reliability information of parts or systems are analyzed, and as shown in FIG. And the cause of failure are shown in a tabular form as a so-called FMECA table in which each part or system is classified, and the failure mode column describes the failure mode of the part or system, the effect of the failure, and the method of detecting the failure. Has been done. Moreover, the certainty factor is specified in the area surrounded by each grid. This certainty factor is set in consideration of MTBF (mean time between failures) of parts or systems, the number of possible causes of failure for the same failure phenomenon, or the number of actual occurrences of failures. The degree indicates the depth of the causal relationship between the failure phenomenon and the cause of failure.

The computerized manual data is text and graphics data representing inspection, replacement, or assembly procedure of components and the like. For example, as shown in FIG. 23, the touch screen of the apparatus body 1 described above. 1a,
The data of the booster / pump graphic and the inspection procedure of the booster / pump are read from the electronic manual data and displayed in the window.

Further, the operation record data section 17 is composed of a maintenance record data storage section 17a and a work progress temporary storage section 17b. , And the maintenance results such as the status of trial operation are stored for each case. Further, in the work progress temporary storage unit 17b, for example, in the case of interruption for arranging replacement parts during maintenance, and then restarting the failure diagnosis, it is possible to continue from the interrupted maintenance work. In order to achieve the above, the progress of the maintenance performed by the maintenance staff or the maintenance status is sequentially stored.

Next, the failure diagnosis procedure will be described with reference to the flowcharts of FIGS. In addition, in the present embodiment, description will be given while exemplifying a defect of the fuel system of the aircraft.

This failure diagnosis is executed according to the inference processing routine of FIG.

First, when a maintenance worker selects "start diagnosis" from the menu displayed on the touch screen 1a of the apparatus body 1 of the failure diagnosis apparatus A, as shown in FIG. A window and a pen input window for inputting this trouble phenomenon in text representation are displayed, and the inference process is started.

In step S1, the maintenance worker refers to the form such as the defect record form 21 (see FIG. 9) handed over, and the necessary condition such as the movable state of the system of the aircraft or the defect transmitted from the crew member. The phenomenon (symptom) is input to the touch screen 1a using the input pen 2.

After the maintenance staff completes the inspection, maintenance record data such as the treatment and the trial operation status can be written in the action column of the defect recording form 21 shown in FIG. 9 (see FIG. 27). .

On the other hand, in step S2, symptoms such as measured values obtained by the tester connection are directly input in response to the inquiry displayed in step S12 described later.

When the necessary items have been entered in the fields such as the trouble phenomenon input window on the touch screen 1a and the "end of input" window is selected by the input pen 2, the entered information (date, month, day, (Duties, contents, etc.) are output as input data and stored in the memory in step S3.

Then, in step S4, the degree of similarity between the defect phenomenon input in step S1 and stored in step S3 and the knowledge data stored in each of the knowledge data storage units 16a to 16c is calculated.

The calculation of the similarity is performed by the character string search unit 14a.
Is performed by the following procedure.

(1) First, each of the above knowledge data storage units 16a
Character strings (sentences) whose similarity is to be calculated are taken out from the respective knowledge data of 16c. It should be noted that each character string is provided with an ID (self-certification) number for associating with the knowledge data stored in each of the knowledge data storage units 16a to 16c.

(2) Next, the character string to which the ID number is assigned is decomposed into each unit of string, word and character, and stored in the working memory for each knowledge data. The "disassembled character string" stored in each working memory has the same ID number as that before disassembly.

Here, the “string” means the character string itself, and the “word” means the blank character and the pre-registered separating characters (“.”, “,”, “Ga”, “ ”
Etc.), for example, the string "engine does not start"
Is disassembled into "engine" and "without starting".

The character means that a character string is decomposed into a predetermined number of characters. For example, the character string "engine does not start", "engine""starts","does not move", or " It is disassembled every three characters such as "d", "jinjin", "was started", "do not", or "en", "jinga", "started", "zu", etc.

(3) Next, the "ignored word (" but "," iwa ", etc.) registered in advance is deleted from the" decomposed character string "stored in the working memory.

(4) After that, the "decomposed character string" is replaced by using a "synonymous phrase (" abnormal "" is regarded as "fault" etc.) registered in advance ".

(5) On the other hand, the character string of the trouble phenomenon (symptom) inputted by the maintenance staff is decomposed into each unit of string, word and character, as in the above (1) to (4).

(6) Then, the "decomposed character string" stored in the working memory is searched using the decomposed character string of the trouble phenomenon.

For example, the degree of similarity between the character string extracted from the knowledge data stored in each of the knowledge data storage units 16a to 16c and "The engine does not start" input by the maintenance worker is as follows. Become.

[0052] (7) Next, data that completely matches in each unit of string, word, or character is totaled for each ID number, and the totaled result is output to the corresponding inference units 14b and 14c as the similarity. When totaling, the weighting coefficient is used for each string, word, or character.

As a result, each of the knowledge data storage units 1 indicates that the trouble phenomenon “fuel piece is reduced”, “left tank is not reduced and consumption is from right side”
The degree of similarity with the knowledge data stored in the knowledge data 6a to 16c is as follows.

Among the knowledge data stored in the diagnostic case type knowledge data storage unit 16a, the diagnostic cases for which the similarity is calculated are as shown in FIG. 11, and the similarity of each diagnostic case is The value becomes 12. In this, case-1048
“Cruising at 3000 ft” “Left tank does not consume fuel”
Is the highest similarity, that is, the case where the input failure phenomenon and symptom are the closest.

Further, in the knowledge data stored in the FTA type knowledge data storage unit 16b, the tree for which the similarity is calculated is the "fuel system malfunction" as shown in FIG.
Is a tree made up of a chain of rules with the top event as, and the similarity of each hypothesis is as follows.

1) Rule 5 "Internal defect in the pump itself" →
3, 2) "Fuel leakage", which is an intermediate hypothesis between rules 2 and 4 →
40, 3) "Fuel leak" of Rule 3 → 40, 4) "Booster pump abnormality" which is an intermediate hypothesis between Rule 1 and Rule 2 → 3, 5) "Transfer, which is an intermediate hypothesis between Rule 1 and Rule 3" 6) In Rule 1 i) "Fuel shut-off valve error" → 35, ii) "Fuel sub-tank error" → 36, iii) "Fuel indication system failure" → 38, is there.

On the other hand, the FMECA type knowledge data storage unit 16
In the knowledge data stored in c, the phenomenon for which the similarity is calculated is as shown in FIG. 17, and in this, the similarity is “fuel piece reduction” as shown in FIG. It shows a high value.

In step S4, each knowledge data storage unit 1
When the similarity between the knowledge data stored in 6a to 16c and the failure phenomenon is calculated, the following steps S5 and S are performed.
6 and S7 are executed in parallel (concurrent processing), and in each of steps S5, S6 and S7, the cause of failure is inferred and hypothesized based on the knowledge data stored in the knowledge data storage units 16a to 16c. To generate.

That is, in step S5, the knowledge data stored in the diagnostic case type knowledge data storage section 16a is used to search the cause of failure by case-based reasoning and generate a hypothesis. In step S6, the failure cause is searched for by rule-based reasoning using the knowledge base stored in the FTA type knowledge data storage unit 16b, and a hypothesis is generated. Further, in step S7, the cause of failure is searched for by rule-based reasoning using the knowledge base stored in the FMECA type knowledge data storage unit 16c, and a hypothesis is generated.

First, generation of a hypothesis by the diagnosis case inference in step S5 will be described. This diagnostic case inference is executed by the case-based inference unit 14c, and is specifically executed according to the hypothesis generation routine by the diagnostic case inference shown in FIG.

In step S21, hypotheses are narrowed down based on the similarity of each case of the knowledge data stored in the diagnostic case type knowledge data storage unit 16a calculated in step S4 of the inference processing routine. The narrowing down of this hypothesis is judged by whether the similarity is equal to or more than a preset "threshold value". For example, if the "threshold value" is 10,
As shown in FIG. 1 and FIG. 12, “case-1048” is a hypothesis generated by diagnostic case inference. In the knowledge data shown in FIGS. 11 and 12, there is only one case that exceeds the “threshold value”, but when there are a plurality of cases having a similarity higher than the “threshold value”, , They are all hypotheses based on diagnostic case inference.

Then, in step S22, the reliability of the hypotheses narrowed down in step S21 is calculated. This reliability serves as a reference when resolving a conflict with a hypothesis obtained by other inference (FTA inference or FMECA inference), and the similarity calculated in step S4 of the above inference processing routine is multiplied by a coefficient. And calculated. In addition,
In this embodiment, as shown in FIG. 13, the coefficient for calculating the reliability is set to 1.

Then, the cases narrowed down in step S21 are output to the inference control section 14d as "hypotheses by diagnostic case inference", and step S8 of the inference processing routine is executed.
Return to.

Next, generation of a hypothesis by FTA inference in step S6 will be described. This FTA inference is executed by the rule-based inference unit 14b, and is specifically executed according to the hypothesis generation routine by FTA inference shown in FIG.

First, in step S31, a rule having "fuel system malfunction" as a top event in the knowledge data stored in the FTA type knowledge data storage unit 16b calculated in step S4 of the inference processing routine is used. Hypotheses are narrowed down by selecting hypotheses whose similarity is equal to or more than a preset "threshold" based on the similarity of each hypothesis of the tree formed by the chain.

For example, when the "threshold value" is 15,
In the FTA-type knowledge data of FIG. 14, the above-mentioned a) “fuel leakage”, which is an intermediate hypothesis between rules 2 and 4, b) “fuel leakage” of rule c) i) of rule 1, “fuel shut off. "Valve abnormality" ii) "Fuel sub-tank abnormality" iii) "Faulty indication of fuel indicating system" is narrowed down.

Then, in step S32, a route connecting the hypotheses narrowed down in step S31 is searched for. In FIG. 14, the “phenomenon in the fuel system”, which is the top phenomenon, is the “defect phenomenon” this time, and the root extending in the left direction from the defect phenomenon has a causal relationship with this phenomenon, that is, It is a "hypothesis". As shown in FIG. 15, the routes connecting the hypothesis selected in step S31 and the malfunction phenomenon are 5 routes 1 to 5 as shown in FIG.
Get on the street.

Then, in step S33, the completeness of the route searched in step S32 is calculated. The higher the degree of perfection, the stronger the connection (causal relationship) between the failure phenomenon and the cause of failure, the higher the value. The completeness of the route searched in FIG. 15 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 S34, based on the completeness, this completeness is determined in advance. The hypothesis is narrowed down by whether it is equal to or more than the set "threshold". For example, when the "threshold value" is set to 20, all of the routes 1 to 5 are targeted.

Then, in step S35, the reliability of the hypotheses narrowed down in step S34 is calculated,
The respective hypotheses are output to the inference control unit 14d as "hypotheses by FTA inference" and the process returns to step S8 of the inference processing routine.

The reliability is set by multiplying the completeness by a coefficient. For example, the "threshold" is 20,
When the coefficient is 1, as shown in FIG.
The “hypothesis based on A inference” applies to all routes 1 to 5.

Next, step S of the above inference processing routine.
Generation of a hypothesis by FMECA inference performed in 7 will be described. This FMECA inference is executed by the rule-based inference unit 14b similarly to the FTA inference, and specifically, it is performed according to a hypothesis generation routine by FMECA inference shown in FIG.

First, in step S41, the phenomena to be diagnosed this time are narrowed down from each phenomenon of the knowledge data stored in the FMECA type knowledge data storage unit 16c. This phenomenon is narrowed down by selecting a phenomenon whose similarity is equal to or more than a preset "threshold" based on the similarity calculated in step S4 of the above inference processing routine for the phenomenon corresponding to the defective phenomenon. To do. For example, when the “threshold” is set to 10, the so-called FM shown in FIG.
The phenomena displayed in the ECA table are narrowed down to "fuel decrease", "fuel piece decrease", and "fuel tank remaining amount cannot be corrected" (see FIG. 18).

Then, in step S42, the certainty factor corresponding to the phenomenon narrowed down in step S41 is set to the FM.
Obtained from the ECA table. For example, in the so-called FMECA table shown in FIG. 17, the certainty factor corresponding to “fuel reduction” is 1
6,21,63, the certainty factor corresponding to "fuel depletion" is 5
0, 31, 46, and the certainty factors corresponding to "fuel tank remaining amount cannot be corrected" are 40, 60.

Then, in step S43, a certainty factor equal to or higher than a preset "threshold value" is selected from the above certainty factors to narrow down the hypotheses. For example, the above "threshold"
When is set to 20, as shown in FIG. 17, the certainty factor surrounded by a thick frame is targeted, and in step S44, the component and the cause corresponding to this certainty factor are selected.

Next, in step S45, the reliability of the component or cause selected in step S44 is calculated, and the reliability and the hypothesis are set as "hypothesis by FMECA inference", for example, data as shown in FIG. Output to the inference control unit 14d, and return to step S8 of the inference processing routine.

In the present embodiment, the reliability is calculated by multiplying the sum of the values obtained by multiplying the similarity and the certainty factor by the coefficient, which is shown in the table below.

[0078] Then, the process returns to step S8 of the inference processing routine shown in FIG. 2, and it is determined whether the hypotheses generated in steps S5, S6, and S7 conflict, and if they conflict, the hypotheses are resolved according to preset criteria. This hypothesis conflict resolution is executed by the inference control unit 14d shown in FIG. 1, and is specifically performed according to the hypothesis conflict resolution routine of FIG.

First, in step S51, the hypotheses generated in steps S5, S6 and S7 are summarized. The hypotheses generated in steps S5, S6, and S7 are aggregated for each inference, for example, as shown in the table in FIG. 20, and ranked according to the calculated reliability.

Then, in step S52, the consistency of the hypotheses generated in steps S5, S6, and S7 is obtained from the defect target, the attribute, and the like. For example, in FIG. 20, the rank 1 of the hypothesis based on the diagnostic case inference and the rank 1 of the hypothesis based on the FMECA inference are consistent in that the defective component is “booster pump”, but the rank of the hypothesis based on the diagnostic case inference is 1 and the hypothetical rank 1 based on FTA inference are common in that the defective phenomenon is "a malfunction of the fuel system", but the defective parts compete with each other. Similarly, in order 1 of the above-mentioned hypothesis based on FTA inference and the hypothesis based on FMECA inference, defective parts are in conflict.

Then, in step S53, the consistency of the hypotheses calculated in step S52 is collated with a preset "standard of judgment of consistency" to judge whether or not there is consistency. In this embodiment, the "matching criterion" is defined as "all defects having the highest reliability of the hypothesis in each inference match", and the matching is judged according to this definition.

As a result, for example, as shown in FIG.
If all the hypotheses of rank 1 generated by each inference do not match, it is determined that there is no consistency in step S54, the process proceeds from step S54 to step S55, the inference frequency measurement count value C is incremented, and the process proceeds to step S56. move on.

In step S56, the inference count measuring count value C is compared with the set value N. If C≤N,
It is determined that "diagnosis is not completed", and the process proceeds to step S57. In step S57, data for re-inference is created according to a certain standard from the hypotheses collected in step S51. If this criterion is defined as, for example, "selecting the one with the highest reliability of each hypothesis", the re-inference data is created based on the FMECA inference rank 1 hypothesis in the summary table of FIG. (See FIG. 21), this data is output together with the information of “diagnosis not completed”, and step S9
Return to.

On the other hand, when it is determined that C> N in the above step S56, that is, when the hypotheses do not match even after the re-inference is repeated the set number of times, the process proceeds to step S58.
The conflict is resolved according to a preset priority criterion.

That is, the hypothesis based on the FTA type knowledge data is given the highest priority, and when the hypothesis is not generated by the inference using this FTA type knowledge data, the hypothesis generated based on the FMECA type knowledge data is adopted, and If the hypothesis is not generated by any inference of both knowledge data, the hypothesis generated by the inference using the diagnostic case type knowledge data is adopted.

The FTA-type knowledge data is most reliable because it is expressed in a rule format by logically analyzing defective phenomena, defective parts, and causes of failures. Also,
The FMECA-type knowledge data links a failure phenomenon to a cause or a defective component with a certainty factor, and this certainty factor statistically indicates the depth of a causal relationship between the defective phenomenon and the cause of the failure. Although the reliability is a little inferior to the type knowledge data, each time failure diagnosis is accumulated, it is updated in step S18 of an inference processing routine, which will be described later, so that the reliability is gradually increased. Further, the diagnostic case type knowledge data is a knowledge base constructed on the basis of past diagnostic cases, and even if it is a hypothesis that is not generated by inference based on the above knowledge data, if it is registered as a past case, A hypothesis is generated based on this case, and the missing part of both knowledge data can be interpolated.

Then, the hypothesis in which the conflict is resolved according to this priority is output as a search result together with the information of "diagnosis completed", and the routine is exited. If the cause of the failure cannot be investigated in any of the knowledge data, the information of "hypothesis failure" and "diagnosis end" is output in step S58, and the process returns to step S9.

On the other hand, when the hypotheses generated in the steps S5, S6, and S7 are matched in step S54, the hypothesis is used as a search result, and the process branches to step S59 to calculate the inference frequency count value C. After clearing
Output the above inquiry result together with the information of "Diagnosis completed",
Return to step S9.

Then, step S9 of the inference processing routine.
Returning to step S10, it is determined whether "inference is completed" based on the information created in step S8. If "inference is completed", the process proceeds to step S10. If "inference is not completed", the process is completed. Return to S4.

When it is judged in step S9 that "inference is not completed" and the process returns to step S4, the trouble phenomenon stored in the memory in step S3, the input data such as a predetermined measurement value, and the step S8 ( Step S5
The similarity between the character string combining the “re-inference data” created in 7) and the knowledge data stored in each of the knowledge data storage units 16a to 16c is calculated again.

In the data for re-inference of FIG. 21, "fuel depleted", "left tank not reduced", "only consumed from the right", "booster pump", and "improper operation of own" are given as character strings. , The character strings and the knowledge data storage unit 16 described above.
defect phenomenon of knowledge data stored in a to 16c,
Comparing defective parts, causes of failure, etc., the degree of similarity is calculated by logical operation or the like.

Then, in accordance with the degree of similarity of each piece of knowledge data, the cause of the failure is sought by re-inference in steps S5, S6 and S7.

As a result, in the diagnostic case inference in step S5, the hypothesis generated based on the diagnostic case type knowledge data is not changed even at the time of re-inference, so the same inference result as the previous time is output.

On the other hand, in the FTA inference executed in step S6, as shown in FIG. 14, "fuel system malfunction" and "booster / pump abnormality" are "phenomenon" from the character string obtained by the similarity calculation. The causal relationship between the hypothesis that is the cause of the failure and the phenomenon is strongest in the route 5 shown in FIG. 15. Therefore, as shown in FIG. 22, the reliability of the route 5 is as high as 80, and the FTA is high. Inference hypothesis ranking 1
become.

Further, in the FMECA inference in step S7, as shown in FIG. 17, there is a character string that matches "fuel depletion", the hypothesis remains unchanged, and the same result as the previous inference is output. It

As a result, in step S54 of the hypothesis conflict resolution routine of FIG. 7, which is executed in step S8, as shown in the table of FIG. The predefined "matching criterion" is satisfied, it is determined in step S54 that there is consistency, and the process branches to step S59 to clear the inference frequency measurement count value C, and then the search result. Is output together with the information of "diagnosis completed", and the process returns to step S9.

If it is determined in step S9 that "inference is completed", the process proceeds to step S10, it is determined whether there is another inspection item, and if there is an inspection item, step S11 is performed.
Branch to and narrow down the contents of the inquiry to the maintenance staff or the user. Through this inquiry, the conclusion hypothesis of this time is verified and used as information when a new hypothesis is generated. That is, for example, in the FTA type knowledge base, as shown in FIG.
As shown in 5, there is a rule that the "fuel leak" that is linked to the intermediate booster / pump abnormality of rule 2 by the rule is:
The conclusion hypothesis of Rule 4 is chained. Therefore, by designating the place where the abnormality of the booster / pump is identified by the inquiry and inspecting this place, it is possible to further narrow down the faulty place.

Then, in step S12, the contents of the inquiry that have been narrowed down in step S11 are displayed on the touch screen 1a of the apparatus main body 1, and at the same time, the work contents such as related inspection procedures and necessary information are displayed, for example, in FIG. Display as shown in. Then, the process returns to steps S1 and S2, and waits for the maintenance staff to input the inspection result.

Then, the maintenance person touches the touch screen 1
When the inspection result is input to the a from the input pen 2 or the tester, this data is stored in step S3, and in step S4, the input character string, the character string of "input data" up to the previous time, and the measurement result are stored. Based on this, the degree of similarity with each piece of knowledge data is recalculated, and in steps S5, S6, and S7 and below, inference is performed again based on each piece of knowledge data to investigate the cause of the failure.

Then, after the inference is completed, the process proceeds from step S9 to step S10. If there is no inspection item for verifying the inference, the process proceeds to step S13 as it is, and it is determined whether the cause of the failure can be searched.

When the cause of the failure can be sought, step S
At 14, for example, as shown in FIG. 24, the result of reasoning and the reasoning are displayed on the touch screen 1a, and the mechanic
Confirm this content and enter the "OK" window. Pen 2
If selected by, the necessary procedure, such as the replacement procedure of parts corresponding to the hypothesis, is displayed in step S15. Next, in step S16, the maintenance staff waits for the input of the result of the treatment, that is, the result of whether or not the malfunction is resolved. Then, when this treatment result is input, the process proceeds to step S17, and the inquiry result is confirmed.

On the other hand, when it is determined in step S13 that the cause of the failure cannot be found by inference, steps S14 to S16 are skipped and step S1 is executed.
Proceed to 7.

Then, in step S17, a search result confirmation screen is displayed on the touch screen 1a. The confirmation screen of the inquiry result is, for example, as shown in FIG.
This is done by displaying the history of the inquiry. If a satisfactory result is not obtained even if the parts are replaced in accordance with the inquiry result, the maintenance worker inputs the effect in step S16, and in step S17, the inquiry result is displayed on the touch screen 1a. Displays a confirmation screen indicating that is failed.
Further, in the dialog box displayed on the left side of FIG. 25, for example, Then, the confirmation contents for obtaining the consent from the maintenance staff are displayed.

When the maintenance staff has accepted the registration of the case, in step S18, first, the phenomenon and the cause or the component selected in this exploration of the knowledge data stored in the FMECA type knowledge data storage section 16c are selected. The certainty factor (see FIG. 17) connecting with is updated according to the result of the failure investigation this time. That is, if the inquiry is successful, the certainty factor is made relatively high, and if the inquiry is unsuccessful, the certainty factor is made relatively low.

Then, in step S19, the diagnostic case type knowledge data storage unit 16 is set with the result of the current inquiry as a diagnostic case.
For example, as shown in FIG. 26, the knowledge data stored in a is added (case-5721). If the result of this inquiry is unsuccessful, the content is "cause and action".
The column of "Special notes" is described as "Failure to search".

Next, in step S20, the history of this investigation is stored in the maintenance record data storage unit 17a together with supplementary information such as the date of diagnosis and the name of the person who made the diagnosis, and the failure diagnosis is terminated.

The data stored in the maintenance record data storage section 17a can be taken out via an external printing machine or the like. For example, as shown in FIG.
You can add up and add up.

In the hypothesis conflict resolution routine shown in FIG. 7, the number of times of re-inference is counted in steps S55 and S56, and if the set number of times is exceeded, the knowledge data is prioritized to derive the conclusion hypothesis. However, the
As shown in step S61 of Step 8, when the data for re-inference (see FIG. 21) cannot be created, step S5
It is also possible to branch to 8 and derive the conclusion hypothesis in accordance with the above-mentioned priority order.

When the data for re-inference cannot be created, the criterion for creating the data for re-inference is, for example, "the reliability of each hypothesis is equal to or more than a predetermined threshold value,
In addition, when the highest one is selected for each ”, the reliability hypothesis satisfying this threshold value no longer exists.

As described above, in this embodiment, the failure diagnosis is
Diagnostic case type knowledge data, FTA type knowledge data and FMEC
With the knowledge data different from the A-type knowledge data, the cause of the failure is investigated, and then the consistency of the hypothesis generated by each knowledge data is judged. If there is no consistency, this reasoning result and maintenance engineer's Combining with the input data,
Since it is designed to investigate the failure, accurate inference results can be obtained. If the hypotheses do not match even after the re-inference is repeated a predetermined number of times, the knowledge data are always prioritized and used as the conclusion hypothesis. It is possible and easy to use.

Further, since a plurality of inferences are combined to generate a hypothesis, a high degree of perfection is not required for each piece of knowledge data, so that a knowledge base can be constructed at low cost. In addition, the cost required for maintenance can be suppressed.

The present invention is not limited to the above embodiment, and the target of failure diagnosis is not limited to an aircraft, but may be an automobile, a vehicle such as a railroad, or a ship.

[0113]

As described above, according to the present invention,
Even if multiple pieces of knowledge data are adopted, consistent search results can always be derived, so not only can inspection work be performed efficiently, but the missing parts of each piece of knowledge data can be interpolated by other knowledge data. Therefore, it is possible to obtain high reliability in the diagnosis result.

[Brief description of 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 flowchart showing a hypothesis generation routine by diagnostic case inference.

FIG. 5 is a flowchart showing a hypothesis generation routine by FTA type knowledge base inference.

FIG. 6 is a flowchart showing a hypothesis generation routine by FMECA type knowledge base inference.

FIG. 7 is a flowchart showing a hypothesis conflict resolution routine.

FIG. 8 is an external view of the failure diagnosis device

FIG. 9 is an explanatory diagram showing a defect recording form.

FIG. 10 is an explanatory diagram showing an input screen of a defect phenomenon.

FIG. 11 is an explanatory diagram showing diagnosis case type knowledge data.

FIG. 12 is an explanatory diagram showing the calculation result of the similarity in the diagnosis case type knowledge base inference.

FIG. 13 is an explanatory diagram showing a hypothesis based on a diagnosis case type knowledge base inference by reliability.

FIG. 14 is an explanatory diagram showing FTA type knowledge data.

FIG. 15 is an explanatory diagram of a route search result in FTA type knowledge base inference.

FIG. 16 is an explanatory diagram showing a result of hypothesis generation by FTA type knowledge base inference.

FIG. 17 is an explanatory diagram of FMECA type knowledge data.

FIG. 18 is an explanatory diagram of a calculation result of similarity by FMECA type knowledge base inference.

FIG. 19 is an explanatory diagram showing a result of hypothesis generation by FMECA type knowledge base inference.

FIG. 20 is an explanatory diagram showing aggregation of hypotheses obtained by each inference.

FIG. 21 is an explanatory diagram of re-inference data.

FIG. 22 is an explanatory diagram showing aggregation of hypotheses obtained by re-inference.

FIG. 23 is an explanatory diagram showing a display screen during an interview.

FIG. 24 is an explanatory diagram showing a display screen of inference results and reasoning.

FIG. 25 is an explanatory diagram showing a display screen for confirming the inquiry result.

FIG. 26 is an explanatory diagram of diagnosis case type knowledge data in which a new case is registered.

FIG. 27 is an explanatory diagram showing a defect record sheet in which details of measures are described.

FIG. 28 is a flowchart showing a hypothesis conflict resolution routine according to another aspect.

[Explanation of symbols]

14 inference mechanism section 14b rule base inference section 14c case base inference section 14d inference control section 16 knowledge base section 16a diagnostic case type knowledge data storage section 16b failure tree analysis (FTA) type knowledge data storage section 16c failure mode impact analysis (FMECA) ) Type knowledge data storage

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI Technical display location G01M 19/00 Z (72) Inventor Yasuo Kagei 1-7-2 Nishishinjuku, Shinjuku-ku, Tokyo Fuji In heavy industry Co., Ltd. (72) Inventor Masaaki Furuyama 1-2-7 Nishi-Shinjuku, Shinjuku-ku, Tokyo Fuji Heavy Industry Co., Ltd. (72) In-house Kunihiro Abe 1-2-7 Nishi-Shinjuku, Shinjuku-ku, Tokyo Fuji Heavy Industries Ltd. In the company

Claims (1)

[Claims] 1. A failure diagnosis apparatus having a knowledge base section (16) for storing knowledge data necessary for failure diagnosis, and an inference mechanism section (14) for investigating a cause of failure by inference using this knowledge data. , A knowledge base unit (16), a diagnostic case type knowledge data storage unit (16a) that stores past diagnostic cases as knowledge data, and a set of hypotheses by theoretically analyzing causal relationships between failure phenomena and failure causes A failure tree analysis type knowledge data storage unit (16b) that stores knowledge data represented as a tree, and a failure mode effect analysis type knowledge data that stores failure phenomena and failure causes by associating them with certainty A storage unit (16c) is provided, and a failure phenomenon that is the same as or similar to the input failure phenomenon is input to the inference mechanism unit (14). The case-based reasoning unit (14c) that searches the knowledge data stored in a) to search for the cause of failure and generates a hypothesis, and the cause of failure related to the input failure phenomenon to the knowledge data storage unit for failure tree analysis The knowledge data stored in (16b) is used to search by inference to generate a hypothesis, and a failure phenomenon that is the same as or similar to the input failure phenomenon is stored in the failure mode effect analysis type knowledge data storage unit (1
A rule-based reasoning unit (14) that searches the knowledge data stored in 6c), searches for the failure cause and generates a hypothesis in accordance with the certainty factor connecting the searched failure phenomenon and failure cause.
b) and the inference units (14b, 14c), the consistency between the hypotheses generated is judged, and if the respective hypotheses are consistent, this hypothesis is output as an inference result, while the inference units (14b, 14c, 14b, 14c) are output. 14
When the hypotheses generated in c) conflict, at least one of the competing hypotheses is selected, and each of the inference units (14b, 14b) based on the selected hypothesis and the malfunction phenomenon is selected.
When the re-inference is executed in c) and the re-inference is executed a plurality of times and the conflict of the hypotheses is not resolved, based on the knowledge data stored in the fault tree analysis type knowledge data storage unit (16b). The hypothesis generated by the above is output as a search result, while the fault tree analysis type knowledge data storage unit (16
When the hypothesis is not generated by the knowledge data stored in b), the hypothesis generated based on the knowledge data stored in the failure mode influence analysis type knowledge data storage unit (16c) is used as the search result. When the hypothesis is not generated by the knowledge data of the failure mode influence analysis type knowledge data storage unit (16c), the knowledge data stored in the diagnosis case type knowledge data storage unit (16a) is output. An inference control unit (14d) that outputs a hypothesis generated based on this as a conclusion hypothesis.