JPH07114482A - Diagnostic device for fault - Google Patents

Abstract

PURPOSE:To provide a diagnostic device for fault capable of developing a fault diagnostic expert system in parallel with the system development of a diagnosis object system without depending on the experimental knowledge of a professional maintenance engineer as it is, and dealing with even the occurrence of any type of test pattern. CONSTITUTION:Information quantity to be acquired as a result of a test is calculated from test definition knowledge as a test gain by a processing part 11 every kind of test, and furthermore, a test with highest effectiveness is found considering test cost by a processing part 12, and it is executed by a maintenance engineer, and a degree of conviction in the test definition knowledge is updated by the result, and such operation is repeated until all the tests are performed, and the narrowed down result of a faulty device can be obtained from the final execution result by repeating the processing of updating degree of conviction till the end of the last test.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is designed to test a plurality of suspected parts, each of which has a possibility of a failure, among a plurality of parts constituting a failure diagnosis target, by performing a test when presented. The present invention relates to a failure diagnosis device for removing a suspected normal part from a part (this is called a suspected range) and narrowing down the number of suspected parts belonging to the suspected range.

Here, the failure diagnosis target is, for example, an exchange, and in that case, each part constituting the failure diagnosis target is a package forming the exchange, a line connecting the packages to each other, or the like. There is a part.

[0003]

2. Description of the Related Art Conventionally, when a suspected portion is squeezed (that is, a failure portion is separated) by performing an appropriate test on the suspected area, a professional who has been involved in a failure diagnosis work. The maintenance staff used the knowledge of the isolation procedure by the test.

That is, among the knowledge possessed by the professional maintenance personnel, there are those related to the test and the test range, those related to the time required for the test, those related to the cost required for the test, and the results of the execution of the test (failure). Regarding the presence or absence of)
Based on what is related to updating the certainty factor and what is a fusion of these multiple elements, and describing them as empirical knowledge, select and execute an appropriate test,
A failure diagnosis method was used to isolate the failure point.

[0005]

The conventional failure diagnosis system has the following problems. That is, when the knowledge about the isolation procedure by the test is organized by receiving the instruction from the professional maintenance personnel involved in the failure diagnosis work, some of the knowledge is related to the test and the test range and the time required for the test. And about the cost of the test,
Update of confidence level based on test execution results,
It takes a lot of time to organize the knowledge, and it is also difficult to correct the knowledge based on the knowledge after that because multiple elements such as the above are fused.

[0006] Further, since the knowledge regarding the isolation procedure by the test is acquired depending on the empirical knowledge of the professional maintenance person, when the knowledge is obtained, the maintenance person experiences the failure diagnosis work for a certain period. Was needed. Therefore, there is also a problem in that it is not possible to develop a failure diagnosis expert system in parallel with the system development of the diagnosis target system.

Further, since the knowledge obtained by reliant on the empirical knowledge of the professional maintainer describes the isolation procedure only for the assumed test pattern, different test patterns occur. In some cases, there was a problem that it could not be dealt with.

The object of the present invention is to solve the above-mentioned conventional technical problems and to develop a fault diagnosis expert system in parallel with the system development of the system to be diagnosed, without directly relying on the empirical knowledge of a professional maintenance person. It is possible to provide a failure diagnosis device capable of dealing with any test pattern.

[0009]

In order to achieve the above object, in the present invention, a test is performed when a plurality of parts that have a possibility of failure are presented as suspected parts out of the parts that constitute a failure diagnosis target. By doing so, a fault diagnosis device for removing the suspected normal part from the plurality of suspected parts (this is called the suspected range) and narrowing down the number of suspected parts belonging to the suspected range. At

Identification names of various tests, parts to be tested,
A test definition knowledge table that stores the association degree indicating the degree of normality / abnormality that can be determined when a certain test is performed with a certain test target portion as the suspected portion, and the respective costs of various tests in association with each other. ,

With respect to a plurality of presented suspected parts, that is, a suspected range, it is possible to obtain a result of executing a test for each of various tests based on the degree of association of each suspected part obtained by referring to the table. A test gain calculation processing unit that calculates the amount of information that can be obtained as a test gain;

Considering the test gain of each test calculated by the test gain calculation processing section and the test cost of each test obtained by referring to the table, the test effectiveness of each test is determined. A test effectiveness calculation processing unit that calculates and positions various tests in descending order of effectiveness,

From among various tests, the test validity calculation processing unit selects a test determined to have the highest validity, causes a maintenance person to execute the test, and receives a result of the execution, a test selection processing unit,

Depending on the execution result from the maintenance person received by the test selection processing section, the degree of association in the test definition knowledge table is rewritten, and the test gain calculation processing section calculates the next stage test gain. A verification processing unit that stops the operation when there are no more tests to be performed by the maintenance personnel.

[0015]

After the suspected range, which is a group of devices with a possibility of failure, is specified in advance, the most effective test for narrowing down the failure cause candidate (failed device) is selected for the suspected range. The test effectiveness is calculated as the evaluation scale of the above using the test definition knowledge, and the tests are performed in order from the most effective test using this.

[0016]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the overall configuration of a failure diagnosis device as an embodiment of the present invention. In the figure, 11 is a test gain calculation processing unit, 12 is a test validity calculation processing unit, 13 is a test selection processing unit, 14 is a verification processing unit, 21 is a test definition knowledge table, and 22 is a maintenance person interface unit. .

The outline of the operation is as follows. First, a suspected device including a plurality of suspected devices is given in advance, and a suspected device belonging to the suspected device is removed from a suspected device by a test to remove the suspected device. (The number of suspected devices finally obtained is not limited to one, but is the minimum number possible) is the purpose of this operation.

Now, in the test definition knowledge table 21, the identification names of various tests, the test target device, and the degree to which normality / abnormality can be judged when a certain test is carried out using a certain test target device as the suspicious device Is stored in association with the respective degrees of association with the costs of various tests. The test gain calculation processing unit 11 performs a test for each of various tests on the basis of the degree of association for each of the presented suspected devices, that is, the suspected range, for each suspected device obtained by referring to the table 21. The amount of information that will be obtained as a result of execution is calculated as a test gain.

Next, the test effectiveness calculation processing section 12 is predetermined for each of the test gains calculated for each test by the test gain calculation processing section 11 and each of the various tests obtained by referring to the table. Considering the test cost, the test effectiveness is calculated for each of the various tests, and the various tests are positioned in descending order of effectiveness.

The test selection processing unit 13 selects a test determined to have the highest effectiveness by the test effectiveness calculation processing unit 12 from various tests, and the maintenance operator interface unit 22 is used to select the test. To execute and receive the execution result. The verification processing unit 14 rewrites the certainty factor of the suspected device, depending on the execution result from the maintenance person received by the test selection processing unit 13. Then, if there is no test to be executed by the maintenance person, the operation ends (search end), but if not, the test gain calculation processing unit 11 is prompted to calculate the test gain in the next stage.

The test gain calculation processing unit 11 calculates the test gain in the same manner as before from the certainty factor of the suspected device rewritten by the previous test result, and repeats the same as before. Thus, when there is finally no test to be executed by the maintenance person, the verification processing unit 14 ends the operation. Based on the test result obtained last from the maintenance person, the result of the final squeezing is obtained from the plurality of suspected devices first belonging to the suspected range.

Next, the main ones of the blocks shown in FIG. 1 will be specifically described. First, FIG. 3 is an explanatory diagram showing a specific example of the test definition knowledge table 21 in FIG.
As shown in FIG. 3, the test definition knowledge table 21 stores a test type, a test range (suspected device), a test cost, and a relevance as knowledge about updating the certainty factor based on the test execution result. Has been done.

That is, assuming that the test type is the "physical layer loopback test", the test cost (described later) is 5
0, and the suspected device is the “frame processing unit in the instruction card”
Then, when this is tested as an object (test range), the degree of association is 0.7. Generally, the test type and the test range are row-column relationships, the test cost is a value assigned for each row, and the knowledge of updating the certainty factor by the test execution result is a relevance value of 0.0 to 1.0. , Are stored in the squares on the table.

Here, the degree of association is a value given in advance, and if this test is carried out for this suspected device, it is judged whether the suspected device is normal or abnormal.
It shows the ratio of possible judgments such as "possible", "almost possible", "certainly possible", etc. 1.0 represents 100% possible and 0.0 represents 0% possible. .

Further adding the description with respect to FIG. 3, the horizontal row in the table indicates knowledge about the test and its test range,
Numerical values in the table show knowledge about updating the certainty factor based on the test execution result. For example, in the test type “clock card lamp confirmation test”, the test range is “instruction node clock” and “instruction node shelf power supply”, and the “instruction card frame processing unit” is outside the test range.

Further, the knowledge of updating by the test execution result is as follows:
The degree of association is represented by a numerical value of 0.0 to 1.0. For the test range (suspected device) "instruction node clock",
Knowledge that the "clock card lamp confirmation test" is updated with a degree of association of 0.5, and the test range (suspected device) "shelf power in the designated node" is updated with a degree of association of 0.6. It has been described. In addition, a cost value indicating the difficulty in carrying out the test is assigned to each test type.

Next, the test gain calculation processing section 11 in FIG. 1 will be described. First, referring to FIG. Figure 2
There are eight suspected objects (suspected devices) S1 to S8, and executable tests for them are T1 to T8, Ta,
It is explanatory drawing which shows the way of thinking when determining the order of test execution (namely, test selection), when there are 10 kinds of Tb.

In each of the tests T1 to T8, Ta, and Tb in FIG. 2, it is possible to check whether the suspicious object indicated by the shaded line is faulty or whether any other suspicious object is faulty. . Assuming, for example, there is one failure, the test T1 is
This indicates that it is possible to determine whether (S1) is out of order or the other suspected objects (S2 to S8) are out of order. (If there are multiple failures, the above cannot be determined completely.)

Here, if the failure probabilities of the eight suspected objects are all equal (1/8), the test Ta,
The test is performed in order of Tb, and the number of suspected objects is 4,
The method of reducing the number to two is considered preferable. But,
If the failure probability of the suspected object 1 (S1) is 90%, the test T1 is first performed to obtain 90%.
In the case of 1, it is possible to identify the failure location with one test. As described above, in order to measure the effectiveness of the test, it is important to consider the failure probability of each suspected object.

Further, generally, when a test is performed, the failure cause candidates are narrowed down from the suspected object and the search proceeds, but from another perspective, a certain amount of information is obtained as a result of the test execution. It can be considered that the search will proceed accordingly. That is, the larger the amount of information obtained, the more the search proceeds.

From this point of view, the entropy (information amount) in the intermediate state of the search is calculated from the failure probability of each suspected object, and the entropy before and after the test has the largest difference (obtained as a result of executing the test. The one with the most information) is selected as the optimum test. In order to realize these methods, the amount of information is defined as a test gain and calculated by the calculation formula described later.

As already described with reference to FIG. 3, as the test definition knowledge 21, a part (suspected device) capable of confirming whether the test is normal or abnormal when the test is performed is described for each test. The test range is defined, and the strength of the relationship between each test and the device in the test range (suspected device) is 0.0 to
It is expressed using the degree of association shown in the range of 1.0. Furthermore, the certainty factor shown in the range of -1.0 to 1.0 given to each device within the suspected range (each suspected device) and 0 given to each suspected device within the test range. .0 to 1.0
By using the degree of relevance indicated by the range, the change in the confidence level before and after the test execution is verified, and the amount of change in the entropy is calculated to obtain the test gain showing the effect of each test.

The certainty factor here means that an alarm is raised from various alarm devices set up for the test object, the suspicious range is determined by analyzing the alarm, and further, from the alarm analysis result, Among various suspected devices belonging to the suspected range, which device has a high suspicion of failure or low suspicion is used, and the certainty factor is used as a degree indicating the high or low suspicion of the failure. Confidence is-
It is shown in the range of 1.0 to 1.0, and the minus sign represents the direction of not being a failure. Therefore -1.0 means
There is no 100% suspicion of failure, and 1.0 means 100% suspicion of failure.

Now, in the following, the suspicious range and the confidence distribution before the test is performed are shown in FIG. 4, and the relation between the effective test and the test object for the suspicious range is as shown in FIG. 5, respectively. A calculation example is shown. In FIG. 4, the certainty factor of the suspected device 1 (h1) belonging to the suspected range is 0.8, and the suspected device 2 is
The certainty factor of (h2) is 0.5, the certainty factor of the suspected device 3 (h3) is -0.9, and the like. These results are already obtained from the analysis results of the alarms raised from various alarm devices. It is what has been.

FIG. 5 shows the suspected device 1 (h1) and the suspected device 2.
The effective test for (h2) is t1,
The degree of association between the test t1 and the suspected device 1 (h1) is 0.5,
The degree of association between the test t1 and the suspected device 2 (h2) is 0.4, and the ∘ mark indicates that either the suspected device 1 or 2 is determined by the test t1. The same applies to the other tests t2 and t3.

In order to calculate the test gain, the degree of association between the case where the test result is normal and the case where the test result is abnormal is considered as follows. If the test result is normal, it means that the device targeted by the test was normal by the test, and therefore, the degree of relevance shown in FIG. Confidence. If the test result is abnormal, it means that at least one of the devices targeted by the test is abnormal.
Among the degrees of association shown in FIG. 5, a value obtained by normalizing the degree of association of the device to be tested to 1 is set as the certainty factor after the test is performed.

As for the certainty factor of a device outside the suspected range, two types of certainty factors are assigned depending on the search mode. That is,
In the single failure mode, the purpose is to find one faulty device. Therefore, if the test result is abnormal, it means that there is a faulty device within the suspected range and it is judged to be normal outside the test range. And When the test result is normal, it is not possible to make a judgment outside the test range, so 0 is taken as the confidence factor, which means that no information is obtained.

In the multiple failure mode, it is assumed that the device to be searched has a plurality of failed devices. When considering the existence of multiple failures, it is not possible to determine whether a device outside the suspected range is completely abnormal or normal, whether the test result is normal or abnormal. ,
Abnormal 0 means that no information is available
Is the certainty factor.

The confidence factors after the test is executed when the test result is abnormal and when the test result is normal are as shown in FIGS. 6 and 7 in the single failure mode and the multiple failure mode, respectively.
First, referring to FIG. In FIG. 5, the degree of association between the device 1 (h1) and the test t1 was 0.5. Therefore, when the test t1 is executed and the result is normal, the degree of association is 0.5.
6 is shown in FIG. 6 as the certainty factor (certainty factor after the test) of the device 1 (h1). Test t
When 1 is executed and the result is abnormal, the relevance is 0.5
Is normalized to 1, that is, 0.56 is the certainty factor after the test is performed, and this is also shown in FIG. Figure 6
The rest of the above is read in the same way. 7 could be read in the same way, keeping in mind that this is for multiple failure modes.

Next, the certainty factor of the diagnosis target device after the test execution is calculated. The certainty factor is obtained by combining the degree of association between the test and the device under test with the certainty factor before the test is performed. CF1 is the degree of association and CF2 is the degree of certainty before the test is conducted.
, The confidence CF after the test execution is CF1 and CF.
The case is divided according to the value of 2 and is obtained by combining the following (Equation 1) to (Equation 4).

[0041]

[Equation 1]

[0042]

[Equation 2]

[0043]

[Equation 3]

[0044]

[Equation 4]

For example, the relevance of the device under test is 0.
8. As for the combined value (confidence factor after the test is executed) when the confidence factor before the test is −0.5, the above formula 3 is applied,
It is calculated according to the following equation (5).

[0046]

[Equation 5]

The calculation results of the single failure mode and the multiple failure mode are shown in FIGS. 8 and 9 for reference.
Next, the entropy is obtained from the values of the certainty factors before and after the test execution shown in FIGS. The calculation of the test gain value is performed by calculating the change in the information amount before and after the test execution as the entropy difference based on the knowledge stored in the test definition knowledge.

In the entropy calculation, the certainty factor of the suspected device needs to be (Σhi = 1, hi> 0). Therefore, the value between the “threshold value judged to be normal (normal threshold value)” and the “threshold value judged to be abnormal (abnormal threshold value)” is normalized. The normalization method is shown in FIG.
As shown in FIG. 2, please refer to it.

The values thus normalized and the entropy calculation results are shown in FIGS. 10 and 11 for the single failure mode and the multiple failure mode, respectively.
An example of calculation of entropy is shown below (in this example, normal threshold = −0.4, abnormal threshold = 0.9).
The calculation formula of entropy is shown by the following Formula 6.

[0050]

[Equation 6]

(Calculation of Entropy) (Single Failure, FIG. 1
0) E (HL) = − 0.28log 0.28−0.21lo
g 0.28-0.26 log 0.26-0.12 log
0.12-0.14 log0.14 = 2.26

E (t1, positive) =-0.27log0.2
7-0.16 log 0.16-0.29 log 0.29
-0.13 log 0.13-0.16 log 0.16 = 2.26

E (t1, different) =-1log1 = 0 E (t2, positive) =-0.46log0.46-0.07
log 0.07-0.43 log 0.43-0.02l
log 0.02-0.02 log 0.02 = 1.53

E (t2, different) =-0.48 log 0.4
8-0.52 log 0.52 = 1.00 E (t3, positive) =-0.33 log 0.33-0.27
log 0.27-0.11 log 0.11-0.13l
log0.13-0.16log0.16 = 2.19 E (t3, different) =-1log1 = 0

(Calculation of Entropy) (Multiple Faults, FIG. 1)
1) E (HL) =-0.28 log 0.28-0.
21 log 0.21-0.26 log 0.26-0.12 log 0.12-
0.14 log 0.14 = 2.26

E (t1, positive) =-0.27log0.2
7-0.16 log 0.16-0.29 log 0.29
-0.13 log 0.13-0.16 log 0.16 = 2.25

E (t1, different) =-0.37 log 0.3
7-0.32log 0.32-0.14log 0.14
−0.17 log 0.17 = 1.89 E (t2, positive) = − 0.46 log 0.46−0.06
log 0.06-0.43 log 0.43-0.02l
log 0.02-0.02 log 0.02 = 1.51

E (t2, different) =-0.27 log 0.2
7-0.25 log 0.25-0.23 log 0.23
-0.26 log0.26 = 2.00 E (t3, positive) =-0.32log0.32-0.27
log 0.27-0.11 log 0.11-0.13l
log 0.13-0.16 log0.16 = 2.19

E (t3, different) =-0.34 log 0.3
4-0.34log 0.34-0.14log 0.14
-0.17 log0.17 = 1.89

Next, the test gain value is calculated by the following equation (7).

[0061]

[Equation 7]

The test gain is obtained by the above equation (7).
The difference in the entropy at the normal time in the equation 7 is based on the value obtained by combining the certainty factor when the test result is normal to the certainty factor of the suspected device by the above equations 1 to 4. Next number 8
It is a value calculated using an equation, and the difference in entropy at the time of abnormality is a value obtained by combining the certainty factor when the test result is abnormal with the certainty factor of the suspected device by the above formulas 1 to 4. Originally, it is a value calculated by using the following Expression 9. .

[0063]

[Equation 8]

[0064]

[Equation 9]

The entropy E (HL) in the above equations 8 and 9 is calculated by the following equation 10.

[0066]

[Equation 10]

The calculation of the coefficient in the above formula 7 is obtained based on the certainty factor given before the test is carried out. Coefficients are suspect and consider only the test range. In the normal case, the coefficient is obtained by subtracting the coefficient from 1. The calculation of the test gain value is performed by the above formula 7, but as the method of giving the coefficient at the time of calculation, the sum of the calculation method using the average value of the certainty factors and the certainty factor normalized to 1 Two methods can be selected as the calculation method using the coefficient as. Below are examples of calculations for each of the multiple failure modes.

In this system, it is considered that the test itself has certainty. Therefore, in the test gain calculation formula (Formula 7), a coefficient is earned as the certainty factor of the test itself. As the calculation method, as shown in the equation (7), the difference between the entropy in the normal condition is multiplied by the value obtained by subtracting the coefficient from 1, and the difference in the entropy in the abnormal condition is multiplied by the coefficient. The test gain was used.

(Calculation method 1) A calculation method using the average value of the certainty factors as a coefficient. Coefficient: t1: (0.8 + 0.5) /2=0.65 t2: (0.5 + 0.1 + 0.2) /3=0.27 t3: (0.5 + 0.7) /2=0.60

Gain (t1) = (1.0-0.65)
(2.26-2.26) +0.65 (2.26-0) = 1.47

Gain (t2) = (1.0−0.27)
(2.26-1.53) +0.27 (2.26-1.0)
0) = 0.87

Gain (t3) = (1.0-0.60)
(2.26-2.19) +0.60 (2.26-0) = 1.38

From the calculated value of gain, the test order is as follows. Test gain order: t1, t3, t2 Therefore, the test is performed in the order of tests t1, t3, t2.

(Calculation method 2) Calculation method using the sum of certainty factors normalized to 1 as a coefficient coefficient: t1: 0.28 + 0.21 = 0.49 t2: 0.21 + 0.12 + 0.14 = 0.47 t3 : 0.21 + 0.26 = 0.47

Gain (t1) = (1.0−0.49)
(2.26-2.26) +0.49 (2.26-0) = 1.11.

Gain (t2) = (1.0−0.47)
(2.26-1.53) + 0.47 (2.26-1.0)
0) = 0.98

Gain (t3) = (1.0-0.47)
(2.26-2.19) +0.47 (2.26-0) = 1.10.

From the calculated value of gain, the order of tests is as follows. Test gain order: t1, t3, t2 Therefore, the test is performed in the order of tests t1, t3, t2. It can be seen that the result does not change by either of the calculation method 1 and the calculation method 2. This is the end of the description of the test gain calculation processing section 11 in FIG.

Next, the test validity calculation processing unit 1 in FIG.
2 will be described. The test effectiveness is defined as the test gain per unit cost of the test, that is, the test effectiveness = test gain / test cost. The test cost is calculated by the sum of the work time, the service interruption time, and the bias arbitrarily set by the maintenance person, that is, test cost = work time + service interruption time + bias.

In the execution of the test, it is necessary to consider not only the effectiveness of the test execution for fault isolation (squeezing of the suspected device) but also the cost of executing the test. As the cost, as shown in the above formula, the sum of the working time, the service interruption time, and the bias is defined as the test cost.

In selecting a test for isolating a failure (squeezing of a suspected device), the selection needs to take into consideration not only the test gain showing the effectiveness of the test itself but also the cost of the test. Therefore, as shown in the above equation,
The test gain per unit cost is defined as the test effectiveness, and in the test selection, the test with the highest test effectiveness has the highest test effectiveness.

For example, when the work time is 20, the service interruption time is 10, and the bias is 10, the test effectiveness of the test t1 calculated by the calculation method using the average value of the certainty factors as follows is as follows. . Test cost = work time + service interruption time + bias =
20 + 10 + 10 = 40 Test Effectiveness = Test Gain / Test Cost = 1.47 / 40
= 0.037

This is the end of the description of the test validity calculation processing section 12 in FIG. Test selection processing unit 1 in FIG.
3. Regarding the verification processing unit 14, there is no additional description.

[0084]

As described above, according to the present invention, the knowledge about the test, the test range, and the test cost is collected at the stage of collecting the knowledge about the isolation procedure by the test from the maintenance person involved in the failure diagnosis. By dividing the knowledge into knowledge and knowledge about updating the certainty factor based on the test execution results, the knowledge can be organized for each test and the independence can be increased, whereby knowledge acquisition and additional correction can be easily realized.

Further, by organizing the test division knowledge as the design knowledge at the time of the test design, it is possible to develop the failure search system without requiring the empirical knowledge from the professional maintenance personnel. This means that there is an advantage that a failure diagnosis system can be developed in parallel with the development of the diagnosis target system.

By using the test effectiveness as an evaluation scale for selecting the most effective test for the suspected range, the effectiveness of the test that is difficult to determine can be selected by a quantitative evaluation scale. In addition, since a test effective for division is selected as a test candidate, division corresponding to any test pattern can be executed.

[Brief description of drawings]

FIG. 1 is a block diagram showing an overall configuration of a failure diagnosis apparatus as an embodiment of the present invention.

FIG. 2 includes eight suspected objects (suspected devices) S1 to S8, and executable tests for them are T1 to T8.
It is explanatory drawing which shows the way of thinking when determining the order of test execution, when there are 10 types of Ta and Tb.

FIG. 3 is an explanatory diagram showing a specific example of a test definition knowledge table 21 in FIG.

FIG. 4 is an explanatory diagram showing a suspicious range and a certainty factor distribution before a test is performed.

FIG. 5 is an explanatory diagram showing a degree of association between an effective test and a test target for a suspected range.

FIG. 6 is an explanatory diagram showing the certainty factors after the test is executed when the test result is abnormal and when the test result is normal, in the case of the single failure mode.

FIG. 7 is an explanatory diagram showing the certainty factors after the test is executed when the test result is abnormal and when the test result is normal, in the case of the multiple failure mode.

FIG. 8 is an explanatory diagram showing a calculation result of a certainty factor in the single failure mode.

FIG. 9 is an explanatory diagram showing a calculation result of a certainty factor in the case of a multiple failure mode.

FIG. 10 is an explanatory diagram showing a value obtained by normalizing a certainty factor and a calculation result of its entropy in the case of a single failure mode.

FIG. 11 is an explanatory diagram showing a value obtained by normalizing a certainty factor and a calculation result of its entropy in the case of a plurality of failure modes.

FIG. 12 is an explanatory diagram showing a normalization method.

[Explanation of symbols]

11 ... Test gain calculation processing unit, 12 ... Test effectiveness calculation processing unit, 13 ... Test selection processing unit, 14 ... Verification processing unit, 21
… Test definition knowledge table, 22… Maintainer interface

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Katsumi Fujita 1-1-6 Uchisaiwaicho, Chiyoda-ku, Tokyo Nihon Telegraph and Telephone Corporation (72) Yukio Murasato 223-1 Yamashita-cho, Naka-ku, Yokohama-shi, Kanagawa No. NTT Software Corporation (72) Inventor Akira Nagasue No. 223-1, Yamashita-cho, Naka-ku, Yokohama-shi, Kanagawa Prefecture NTT Software Corporation (72) Inventor Hiroshi Zhang Inside NTT Software Corporation, 223-1, Yamashita-cho, Naka-ku, Yokohama-shi, Kanagawa

Claims (1)

[Claims] 1. Among the respective parts constituting the failure diagnosis target,
When more than one suspected part is presented as a suspected part, a test is performed to remove the suspected normal part from the plurality of suspected parts (this is called the suspected range). When performing a certain test using the identification name of each test, the test target part, and a certain test target part as the suspicious part in a failure diagnosis device for narrowing down the number of suspect parts belonging to the suspected range A test definition knowledge table that stores the degree of association indicating the degree of normality / abnormality that can be determined and the costs of various tests in association with each other, and the presented multiple suspected parts, that is, the suspected range Based on the degree of association for each suspected part obtained by referring to, calculate the amount of information that will be obtained as a result of executing the test for each of the various tests as the test gain. For each of the various tests, the test gain calculation processing unit, the test gain for each test calculated by the test gain calculation processing unit, and the test cost of each test obtained by referring to the table are taken into consideration. The test effectiveness calculation processing unit that calculates the test effectiveness and positions the various tests in descending order of effectiveness, and the test effectiveness calculation processing unit from among the various tests,
Depending on the execution result from the maintenance person received in the test selection processing unit and the test selection processing unit that receives the execution result by selecting the test determined to have the highest effectiveness and causing the maintenance person to execute it, By rewriting the degree of association in the test definition knowledge table, prompting the test gain calculation processing unit to calculate the test gain in the next stage, and when there is no test to be performed by the maintenance person, a verification processing unit that stops operation, A failure diagnosis apparatus comprising: a final result of a test performed by a maintenance person, and a final narrowed-down result from a plurality of suspected parts first belonging to a suspected range.