JPH08292811A - Method and device for diagnosing fault - Google Patents

Abstract

PURPOSE: To accurately and speedily diagnose the fault of an object system by deciding whether or not a final diagnostic result obtained by selecting low- order nodes in order on the basis of the path selection coefficients and degrees of conviction of the respective nodes is proper and updating the path selection coefficients of the nodes on the diagnostic path. CONSTITUTION: The inference part 21 of the diagnostic device 2 narrows down the cause of the system fault by sequentially selecting a next diagnostic node by using a degree Lk of learning conviction calculated by a conviction degree correction part 2 while referring to diagnostic trees 11a and 11b of a knowledge base 11. The conviction degree correction part 22 calculates the degree Lk of learning conviction from the degree Pk of conviction calculated by the inference part 21 and a learning coefficient Sk received from a learning coefficient storage part 13 and outputs it to the inference part 21. A learning coefficient update part 23 updates the learning coefficient Sk of the node stored in the learning coefficient storage part 13 on the basis of the diagnostic path stored in a path storage part 12 in the name of a node name and a proper/improper decision on the diagnostic result made by an operator. Then a next fault diagnostis is taken by using the updated learning coefficient.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a fault diagnosing device for diagnosing a fault in a complicated system or network, using a diagnostic tree in which diagnostic information is arranged in a hierarchical (tree-like) form as a diagnostic base.

[0002]

2. Description of the Related Art Examples of a system in which a large number of devices are complicatedly related include, for example, a communication network in which a large number of workstations are connected by a communication path, a manufacturing apparatus in which a plurality of manufacturing robots and a slicing device operate in association, and a road network. There is a traffic control device for a railroad network or an information processing device in which a plurality of memories and printer devices are connected in a complicated manner. In recent years, such a system has become larger, more complex, and wider in area, so that when a failure occurs, its impact is great. For this reason, it has become very important to diagnose a system failure accurately, quickly, and efficiently and to recover it early. Various diagnostic methods have been conventionally proposed.

In Japanese Patent Application No. 63-316037 (network fault diagnostic data description method), one diagnostic tree in which diagnostic information is hierarchically arranged is used as a diagnostic base, and it is most likely to be analyzed from the information above it. A method of selecting lower information has been proposed. In this method, one diagnostic tree with enormous diagnostic data is commonly used for all types of faults.

Japanese Patent Laid-Open Nos. 1-267742 and 4-326238 propose a method of providing a plurality of knowledge bases and using the knowledge bases individually to execute a fault diagnosis.

Further, Japanese Patent Application No. 5-336925 proposes a method in which a plurality of diagnostic bases are associated with each other to cooperate with each other, so that the method can be easily applied to a complicated diagnosis target.

[0006]

In the conventional diagnostic method,
In each case, since a fixed diagnostic knowledge base consisting of one or more diagnostic trees is used throughout the diagnosis, when the same fault occurs, the same diagnosis process is followed and the same diagnosis result is output. For this reason, for example, if a part that is prone to failure is changed due to aging of the device, original diagnostic knowledge is insufficient, or if there is a unique feature of the device, an incorrect diagnostic result should be output. However, there is a problem in that even if a correct diagnosis result is obtained, a large amount of time is spent by tracing an unnecessary diagnosis tree path.

SUMMARY OF THE INVENTION An object of the present invention is to provide a fault diagnosing method and device capable of promptly obtaining an accurate diagnostic result in view of the above problems.

[0008]

A fault diagnosis method according to the present invention stores a route selection coefficient (or a learning coefficient) assigned to all diagnosis nodes in a diagnosis tree, and stores them as subordinate nodes of the selected node. Based on the route selection coefficient and the certainty factor (that is, the strength of the relationship between the failure state inspection information of the lower node and the failure inspection result of the target system) assigned to the nodes, the nodes are sequentially selected from the higher order to the lower order. . Then, the route selection coefficient of the node related to the diagnostic route is updated according to the correctness of the fault diagnosis result when reaching the lowest node.

[0009]

Operation: From the upper node to the lower node according to the node hierarchical relationship of the diagnostic tree, the most probable node is selected according to the certainty factor of each node and the route selection coefficient (learning coefficient) to reach the lowest node. When the operator inputs a positive or negative result to the failure diagnosis result indicated by the lowest node, the learning coefficient of the node related to the diagnostic route is updated according to the positive or negative result. The next failure diagnosis is performed by using the updated learning coefficient.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of the fault diagnosis apparatus according to the present invention. The failure diagnosis apparatus of the present embodiment is for obtaining inspection data of a storage device 1, a diagnosis device 2 that executes a diagnosis process, an input device 3 such as a keyboard and a pointing device, a display device 4 such as a CRT, and a diagnosis target system. It comprises an inspection unit 5. The diagnostic device 2 diagnoses, based on the inspection data, a physical fault that occurs in the system to be diagnosed, a malfunction, a fault that occurs due to a program malfunction or parameter setting failure, and the diagnostic result and other necessary information are displayed on the display device 4. To display. The operator executes the diagnostic process through the input device 3, inputs various commands, and determines whether or not the diagnostic result is correct while observing the instructions and data displayed on the display device 4.

The storage device 1 includes a plurality of diagnostic trees 11a and 11a.
.., a path storage unit 12 that stores the node name of the diagnosis path that was traced at the time of diagnosis, and a learning coefficient storage unit 13 that stores the learning coefficient of each node used for diagnosis path selection. And consists of.

The diagnostic device 2 comprises an inference unit 21, a certainty factor correction unit 22, and a learning coefficient updating unit 23. Inference unit 21
As will be described later, while referring to the diagnostic tree of the knowledge base 11, the learning diagnostic factor Lk calculated by the certainty factor correction unit 22 is used to sequentially select the next diagnostic node to sequentially narrow down the cause of the system failure. Go out. The certainty factor correction unit 22
The certainty factor Pk calculated by the inference unit 21 and the learning coefficient storage unit 1
The learning certainty factor Lk is calculated based on the learning coefficient Sk received from No. 3, and is output to the inference unit 21. The learning coefficient updating unit 23 determines the learning coefficient Sk of the node Ck stored in the learning coefficient storage unit 13 based on the diagnosis route stored in the route storage unit 12 by the node name and the correctness determination of the diagnosis result by the operator. Update. This updated learning coefficient Sk
Since the certainty factor correction unit 22 is used when the learning certainty factor Lk is calculated, the inference unit 21 has a high possibility of selecting a diagnostic route in which the past diagnostic result is reflected.

FIG. 2 is a diagram for explaining the hierarchical structure of the diagnostic tree stored in the knowledge base 11. The diagnostic tree has a hierarchical structure of diagnostic nodes C1, C2, ...
The diagnosis is sequentially executed from the upper node to the lower node. Predetermined inspection items and other necessary information are described in each node, and the inference unit 21 receives failure information from the inspection unit 5 according to the inspection items of each node. The lower node describes more detailed inspection items than the upper node. Each node will be described in more detail.

FIG. 3 is a schematic diagram showing an example of the node description format. In the figure, in the name column 111, a name that clearly shows the failure state, for example, "call disabled", "power failure", or the like is described. In the condition column 112, n inspection items in the node are described. For example, in the case of a circuit switched network (telephone line, etc.)
"Noisy", "Low volume", "Intense interference",
"The power source uses XX volts", "to the exchange office"
An inspection item such as “a point of × km” is described. In the certainty factor data column 113, data such as numerical values indicating the inspection results of the inspection items described in the condition column 112 are described.

In the next-stage failure state name field 114, the next-stage failure state name, that is, the lower node name is described. In the lower node, more detailed failure information can be obtained by performing more detailed inspection items. For example, in the case of a circuit-switched network, in the condition column 112 of a certain node, a rough inspection item "call is not possible" is performed, and in a lower node thereof, "there is an abnormality only in a call in a certain area or a certain time zone" For "Noisy", perform more detailed inspections such as "Noise sound type" and "Noisy time period". When more detailed inspection information can be obtained in this way, a plurality of possible lower node names are described in the next-stage failure status name column 114.

In the previous-stage failure state name field 115, the upper-level node name one step before is described. This column 115 is used when returning to the previous stage and performing re-inspection (hereinafter, referred to as backtrack) when it is determined that the system failure state is unlikely to correspond to the inspection item of the node. To be done.

In addition, in the recovery method column 116, if there is a failure recovery method at the corresponding node, that recovery method is described. It is normal for a node in which this recovery method is described to have no lower node name in the next-stage failure status name field 114, but if emergency measures are needed, the contents of this field should be provided to the user. It is displayed by the display device 4. Remarks column 11
In item 7, if there is a need to ask or report a question to the user in a corresponding fault condition, the question and report contents are described. For example, the question taken to understand the content of the failure, how much processing time for failure diagnosis is required, or if the cause cannot be determined, let us know by saying that The description of the procedure is described.

Next, the operation of the above embodiment will be described.

First, at the start of diagnosis, the operator inputs the identification name of the diagnosis tree and the identification name of the node via the input device 3. The inference unit 21 refers to the knowledge base 11 using those identification names as keywords, starts diagnosis from the designated node of the designated diagnostic tree, and searches for a final cause of failure while sequentially selecting lower nodes. . In addition, the node identification names of the lower order nodes sequentially selected from the node that started the diagnosis are stored in the path storage unit 12 as the diagnosis path in the order of selection. The selection of subordinate nodes is performed by calculating the certainty factor and the learning certainty factor as described below.

Calculation of Confidence When the inspection result according to the inspection item of the node Ck is received from the inspection unit 5, the inference unit 21 first calculates the reliability Pk by the following formula 1. The certainty factor Pk is a numerical value indicating the degree of association between the failure state described in the selected node Ck and the actual system failure. For example, it is assumed that n inspection items are described in a certain node Ck, and coefficients R1 to Rn are prepared for each. For each inspection item, a corresponding coefficient Ri is given if an inspection result showing the fault condition of the inspection item is obtained, and a coefficient "0" is given if it cannot be obtained. Then, the certainty factor Pk is calculated by the following equation 1.

[0021]

Confidence factor: Pk = 1-Π (1-Ri) where Π (1-Ri) = (1-R1) x (1-R2) x ... x (1-Rn).

The certainty factor Pk calculated in this way is sent to the certainty factor correction unit 22.

Calculation of learning certainty factor When the certainty factor correction unit 22 receives the certainty factor Pk of the node Ck, the learning coefficient Sk corresponding to the node Ck is read from the learning coefficient storage unit 13, and the learning certainty factor Lk of the node Ck. Is calculated using either of the following equations 2 and 3.

[0024]

[Formula 2] Lk = Pk + Sk

Lk = Pk × Sk The learning certainty factor Lk thus calculated is returned to the inference unit 21.

Selection of Subordinate Node To simplify the explanation, node C of the diagnostic tree shown in FIG. 2 is used.
The procedure for selecting the lower node will be described by taking the case where the diagnosis is performed up to 3 and the lower nodes C5 and C6 exist in the node C3 as an example.

When the inference unit 21 learns from the next-stage failure status name column 114 of the node C3 that there are a plurality of subordinate nodes C5 and C6, it determines the confidence levels P5 and P6 for all the subordinate nodes by the above procedure. It is calculated and sent to the certainty factor correction unit 22. The certainty factor correction unit 22 calculates the learning certainty factors L5 and L6 for these certainty factors P5 and P6, respectively, and returns them to the inference unit 21.
When the learning certainty factors L5 and L6 are received from the certainty factor correction unit 22, the inference unit 21 selects the node having the highest learning certainty factor as the node to be checked next.

When the learning certainty factors of the lower nodes C5 and C6 are lower than a predetermined value, that is, when it is determined that the inspection result of the current system is different from the failure state indicated by the lower node, the inference unit 21 By referring to the previous-stage failure status name column 115 described in the node C3, the process returns to the upper node C1, and by referring to the next-stage failure status name column 114 of the upper node C1, a lower node C2 other than the node C3 is newly added. Select (Backtrack). Then, the same diagnosis process is performed for this node C2.

The node name thus selected is stored in the route storage unit 12, and if the recovery method column 116 or the remarks column 117 of the node is described, the contents thereof are displayed on the display device 4. Is displayed in.

The above-described node selection procedure is repeated until there is no lower node name in the next-stage failure status name column 114 of the selected node. When a node without a lower node name is reached, the failure information of that node becomes the final diagnosis result.

Determination of Correctness of Diagnosis Result When the diagnosis result is obtained in this way, the inference unit 21 causes the display device 4 to display a question as to whether or not the diagnosis result is correct. The operator inputs whether the question is correct or not via the input device 3, and when the diagnosis result is incorrect, the true cause is input. A method of inputting the character “y” or “n” from a keyboard is generally used as a method of inputting correctness. If the diagnosis result is incorrect, the node name that is the true cause may be input from the keyboard, or the diagnosis tree displayed on the display device 4 may be selected with the pointing device.

Update of Learning Coefficient Next, the learning coefficient updating operation of the learning coefficient updating unit 23 will be described with reference to FIGS. In FIGS. 4 to 6, the symbol “◯” indicates a correct diagnosis result, and “x” indicates an incorrect diagnosis result.

First, when there is no backtrack as shown in FIG. 4 and lower nodes are sequentially selected like nodes C1-C3-C5-C7 and a correct result C7 is obtained,
Α is added to the learning coefficient of all nodes of the route (+ α).

Further, as shown in FIG. 5, when the obtained diagnostic result C7 is erroneous and the correct diagnostic result C10 is reached by backtracking, β is obtained from the learning coefficient of the node C5 which gives an erroneous diagnostic result. Is subtracted (-β). In addition, the route C1-C3-that has obtained the correct diagnosis result C10
The learning coefficient in all the nodes C6-C10 is increased by α.

Further, when backtracking is performed or a plurality of correct diagnosis results are obtained as shown in FIG. 6, the learning coefficient of all the nodes from the start node C1 to the correct diagnosis result node is increased by + α, and Β is subtracted from the learning coefficients of the nodes C5 and C10 that have produced the diagnostic results C8 and C12.

The learning coefficient updated in this manner is stored in the learning coefficient storage unit 13 in association with each node of the diagnostic tree in the knowledge base 11. That is, the learning coefficient Sk is updated as Sk + α or Sk−β. Note that the learning coefficient Sk has an initial value of Equation 2 (Lk = Pk + S
When using k), '0' is given, and the equation 3 (Lk =
When Pk × Sk) is used, “1” is given. Further, as + α and −β, the confidence factor Pk of each node is' 1.
If 00 'is a perfect score, α = β =
When using 1 and Equation 3, numerical values such as α = β = 0.01 can be used. Further, the values of α and β may be changed according to the node, and the weights of α and β may be changed.

[0036]

As described above, in the fault diagnosing method and apparatus according to the present invention, the final diagnostic result is obtained by sequentially selecting the most probable lower nodes based on the route selection coefficient and the certainty factor of each node. To get By making a right / wrong determination on the diagnosis result, the route selection coefficient of the node in the diagnosis route is updated, and the next fault diagnosis, that is, the selection of the lower node is performed using the updated route selection coefficient.

Therefore, the lower node can be selected based on the past diagnosis record, the probability of selecting an unnecessary diagnosis route is reduced, and an accurate diagnosis result can be reached at high speed.
For this reason, for example, even if the site where failure is likely to occur changes due to aging of the target system, original diagnostic knowledge is inadequate, or there are system-specific characteristics,
Since the route selection coefficient (learning coefficient) of the node is updated according to the correctness of the diagnosis result, the possibility of misdiagnosis in the next failure diagnosis is extremely low, and an accurate diagnosis result can be quickly obtained.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a fault diagnosis apparatus according to the present invention.

FIG. 2 is a node hierarchical configuration diagram showing a hierarchical structure of a diagnostic tree in the present embodiment.

FIG. 3 is an explanatory diagram showing a description format of a node in this embodiment.

FIG. 4 is a node hierarchy configuration diagram for explaining a first example of a learning coefficient updating operation in the present embodiment.

FIG. 5 is a node hierarchy configuration diagram for explaining a second example of the learning coefficient updating operation in the present embodiment.

FIG. 6 is a node hierarchy configuration diagram for explaining a third example of the learning coefficient updating operation in the present embodiment.

[Explanation of symbols]

 1 Storage Device 2 Diagnostic Device 3 Input Device 4 Display Device 5 Inspection Device 11 Knowledge Base 12 Path Storage Unit 13 Learning Coefficient Storage Unit 21 Inference Unit 22 Confidence Correction Unit 23 Learning Coefficient Update Unit

Claims (8)

[Claims] 1. A fault diagnosis of a target system is performed by sequentially selecting nodes from a higher order to a lower order by using a diagnosis tree having a tree structure of a plurality of diagnosis nodes including information for checking a failure state. In the fault diagnosis method, the route selection coefficient assigned to each of the diagnostic nodes in the diagnosis tree is stored, the route selection coefficient assigned to the lower node of the selected node, and the fault state inspection information of the lower node. And a certainty factor indicating the strength of the relationship between the target system and the fault inspection result, the nodes are sequentially selected from the higher order to the lower order, and the diagnostic path formed by sequentially selecting the nodes is stored. Then, the route selection coefficient of the node related to the diagnostic route is updated according to the correctness of the fault diagnosis result indicated by the lowest node. Cross-sectional method. 2. The fault diagnosis method according to claim 1, wherein if the fault diagnosis result is positive, the route selection coefficient of all nodes related to the diagnosis route is increased. 3. The method according to claim 1, wherein when the failure diagnosis result is negative, the step of returning to the latest upper node and selecting another lower node of the latest upper node is repeated. Fault diagnosis method. 4. When the fault diagnosis result of the lowest node is positive, the route selection coefficient of all the nodes related to the shortest diagnostic route that has reached the fault diagnosis result of the lowest node is increased, 4. The fault diagnosis method according to claim 3, wherein the route selection coefficient of the node related to a route other than the shortest diagnostic route among the diagnostic routes that have reached the fault diagnostic result of 1. is reduced. 5. A fault for performing fault diagnosis by sequentially selecting nodes from an upper node to a lower node by using a diagnostic tree having a tree structure of a plurality of diagnostic nodes including information for inspecting a fault state. In the diagnostic device, first storage means for storing route selection coefficients assigned to all diagnostic nodes in the diagnostic tree, the route selection coefficient assigned to a lower node of the selected node, and the lower node of the lower node Inference means for sequentially selecting nodes from higher order to lower order based on the confidence level indicating the strength of the relationship between the failure status inspection information and the failure inspection result of the target system, and the node selected by the inference means Second storage means for storing a diagnostic route consisting of, and the route of the node related to the diagnostic route according to whether the fault diagnosis result indicated by the lowest node is correct or not. And coefficient updating means for updating 択係 number, fault diagnostic apparatus characterized by comprising a. 6. The fault diagnosis according to claim 5, wherein the coefficient updating means increases the route selection coefficient of all the nodes related to the diagnosis route when the fault diagnosis result is positive. apparatus. 7. The inference means, when the failure diagnosis result is negative, returns to the most recent upper node and repeats the step of selecting another lower node of the most recent upper node. Alternatively, the fault diagnosis device according to item 6. 8. The coefficient updating means, when the failure diagnosis result of the lowest node is positive, sets the path selection coefficients of all the nodes related to the shortest diagnosis path that has reached the failure diagnosis result of the lowest node. The fault diagnosis according to claim 7, wherein the fault selection result is increased and the route selection coefficient of the node related to a route other than the shortest diagnostic route among the diagnostic routes that has reached the failure diagnostic result of the no is decreased. apparatus.