JPH1055277A - Fault analysis instance storage device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To correctly reuse an analytic result by automatically generating and storing a process network on a physical phenomenon level reaching a fault phenomenon from a fault cause and showing the physical causality of a fault to a user who makes use of this fault analysis result. SOLUTION: This fault analysis instance storage device 1 previously stores a physical phenomenon knowledge base 2 with a qualitative state for causing a physical phenomenon and a qualitative state at the point of time when the phenomenon ends and when a fault phenomenon and the fault cause corresponding to the fault phenomenon are inputted in a pair from an input part 3, a qualitative inference part 4 refers to the physical phenomenon knowledge base 2 according to the pair, thereby generating a physical causality network as a chain of the physical phenomenon connecting the fault phenomenon and fault cause through qualitative inference. The generated physical causality network is stored in the fault analysis instance knowledge base 5 and can be retrieved by a fault analysis instance retrieval device 11.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for describing a failure analysis example such as a defect or a failure occurring in a design process and a result obtained by analyzing the failure in a format usable by a computer. The present invention relates to a failure analysis case accumulation device that enables an analyst such as a designer to search and reuse the same.

[0002]

2. Description of the Related Art Conventionally, in order to accumulate a failure analysis case, a failure phenomenon is associated with a failure cause obtained by analyzing the failure phenomenon by an analyst and registered in a database. By constructing such a database, when a designer encounters a trouble such as a failure, the cause of the failure can be estimated based on past analysis cases, and the trouble can be efficiently dealt with. For example, JP-A-6-259
In the design support system described in Japanese Patent Publication No. 294, when a design proposal is evaluated, past prototype data is registered in a prototype database, and a failure type, an evaluation item, and design specifications are associated by weighting. And when evaluating a new design proposal, it is evaluated in association with past cases,
In the case of a failure, the parameter causing the failure is extracted based on the above weighting.

[0003]

Generally, when a failure occurs, there are a plurality of corresponding failure causes. The designer implements a countermeasure by trial and error from possible fault causes, selects a countermeasure having a high improvement effect, and estimates the fault cause. In this case, the designer may catch only a superficial result, and a secondary effect may occur, causing another failure in a later process.

In the design support system described in Japanese Patent Application Laid-Open No. 6-259294, the cause of failure is found by weighting past failure cases and design parameters. However, there has been a problem that reliability cannot be obtained for the obtained result. In addition, even from the viewpoint of the designer who is the user, there is a problem that the process from the cause of the failure to the occurrence of the failure cannot be obtained, so that the validity cannot be confirmed. In addition, since the calculation results are obtained without knowing the weighted process, the evaluation process is not known, and the designer cannot obtain the knowledge of the designer who has experienced similar troubles in the past, and does not lead to skill improvement. There was a problem that.

The present invention focuses on the fact that the above-mentioned inconvenience is caused by the fact that the path from the cause of the failure to the physical phenomenon through which the failure occurs is not specified. However, even if the path is described by the analyst, the details of the description vary depending on the failure analyst, and although the man-hour of the analyst is required, it is not always useful information for the designer. It was done with a focus on something. Therefore, the present invention automatically generates and accumulates a process at a physical phenomenon level from a failure cause to a failure phenomenon as a network, so that a user using the failure analysis result can understand the physical causal relationship of the failure. The purpose is to present and correctly reuse the analysis results. It is another object of the present invention to present a physical causal relationship of a failure when utilizing an analysis result, thereby allowing a user to understand the mechanism of the failure and improving the skill.

[0006]

In order to achieve the above-mentioned object, in the failure analysis case accumulating apparatus according to the present invention, a qualitative state for the occurrence of a physical phenomenon is stored in a physical phenomenon knowledge base. The qualitative state at the time when the phenomenon has ended is stored in advance, and when the failure phenomenon and the failure cause corresponding to the failure phenomenon are input as a set from the input means, the qualitative inference means based on the set. By referring to the physical phenomenon knowledge base, a physical causal network, which is a chain of physical phenomena connecting the failure phenomenon and the cause of the failure, is generated by qualitative inference. Then, by accumulating and storing the generated physical causal network in the failure analysis case knowledge base, a network representing the causal relationship between the failure phenomenon and the cause of the failure is automatically created and accumulated and held in the database.

Further, in the failure analysis case accumulation device according to the present invention, the failure analysis case search means searches the failure analysis case knowledge base for a corresponding physical causal network based on the input failure phenomenon, and outputs the failure analysis case output. By means of outputting the retrieved physical causal network,
The analysis examples stored in the database are presented to the user to realize reuse. Further, in the failure analysis case accumulation device according to the present invention, prior to storing the generated physical causal network in the failure analysis case knowledge base, the physical causal network is presented to the analyst, and approval is obtained from the analyst. By storing the physical causal network in the failure analysis case knowledge base, the reliability of the accumulated analysis cases is improved.

In the fault analysis case accumulating device according to the present invention, the qualitative inference means determines the priority of the physical phenomena to be applied based on the change of the state variables at the inference start point and the inference end point, and according to the priority. By performing qualitative inference using a sequential inference method, inference efficiency is improved. Further, in the failure analysis case accumulating apparatus according to the present invention, the qualitative inference means performs qualitative inference in such a manner that the inference proceeds from both the inference start point and the inference end point in the horizontal direction, thereby narrowing the search space and reducing the inference efficiency. To improve. Further, in the failure analysis case accumulating apparatus according to the present invention, the input means also inputs a known intermediate state between the failure phenomenon and the failure cause, and the qualitative inference means infers for each section divided by the input intermediate state. Qualitative inference is performed by a method that advances the inference, thereby increasing the inference efficiency by dividing the inference section and limiting the search space.

[0009]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A failure analysis case accumulation device according to one embodiment of the present invention will be described with reference to the drawings. FIG. 1 shows a configuration of a failure analysis case accumulation device according to the present embodiment. The failure analysis case storage device 1 is a device that automatically generates and stores a causal network relating to a failure phenomenon. The failure analysis case storage device 1 of the present embodiment searches for and outputs a stored causal network. A case search device 11 is provided.

[0010] The failure analysis case accumulation device 1 corresponds to a physical phenomenon knowledge base 2 in which qualitative states of various physical phenomena are stored in advance, and a failure phenomenon input by an analyst and the failure phenomenon. The set of failure causes (ie,
A failure analysis case input unit 3 for receiving a failure case, and a physical causality which is a chain of physical phenomena connecting the failure phenomenon and the failure cause with reference to the physical phenomenon knowledge base 2 based on the input failure case. The system includes a qualitative inference unit 4 that generates a network by qualitative inference, and a failure analysis case knowledge base 5 that accumulates and stores the generated physical causal network. The failure analysis case input unit 3 also has a function of displaying and outputting to the analyst the physical causal network generated by the qualitative inference unit 4, and the qualitative inference unit 4 confirms that there is an input from the analyst for approval. The physical causal network generated as a condition is stored in the failure analysis case knowledge base 5.

The failure analysis case search device 11 searches for a corresponding physical causal network from the failure analysis case knowledge base 5 based on a failure phenomenon (that is, a case search request) input by a user. And a failure analysis case output unit 13 for displaying and outputting the searched physical causal network to the user. Note that the failure analysis case output unit 13 also has a function of receiving a case search request input from a user.

Here, an outline of the processing performed by the fault analysis case accumulation device having the above configuration will be described with reference to FIGS. 2 and 3. FIG. FIG. 2 shows an outline of processing when a failure analyzer performs a failure analysis and accumulates results obtained by using the failure analysis case accumulation device 1. First, the failure analyzer inputs a set of a failure phenomenon and a failure cause obtained by performing failure analysis on the failure phenomenon by the failure analyzer from the failure analysis case input unit 3 as a failure case. The input combination of the failure phenomenon and the failure cause is sent to the qualitative inference unit 4, and the qualitative inference unit 4 performs qualitative inference using the physical phenomenon knowledge stored in the physical phenomenon knowledge base 2, from the failure cause to the failure phenomenon. (Hereinafter referred to as a physical causal network) that represents a chain of physical causal relationships based on physical phenomena.

The generated physical causal network (result of inference) is sent to the failure analysis case input unit 3, where it is checked by a failure analyst and approved (ACK).
If not approved (NACK), the qualitative inference unit 4 performs re-inference, and is checked again by the failure analyzer. In this way, when approval is given by the failure analyzer, the physical causal network is stored in the failure analysis case knowledge base 5 as a failure analysis case.

FIG. 3 shows an outline of a process when a user such as a designer uses the failure analysis case knowledge base 5 in which the physical causal network is stored as described above. First, the designer inputs the occurring failure phenomenon from the failure analysis case output unit 13 as a case search request. The description of the failure phenomenon is sent to the failure analysis case search unit 12, and a failure analysis case (physical causal network) corresponding to the failure phenomenon input from the failure analysis case knowledge base 5 is searched. Then, the failure analysis case search unit 12 sends the search result to the failure analysis case output unit 13, and the failure analysis case output unit 13 displays and outputs the physical causal network to the designer.

As described above, a failure case (a set of a failure phenomenon and a corresponding failure cause) input from the failure analysis case input unit 3 is constituted by a failure analysis result input form as shown in FIG. The input items that are indispensable as a failure case are the failure cause and the failure phenomenon (that is, failure state), and the failure analyst uses the state variables representing the state of the system to be analyzed for each failure phenomenon and failure cause. Describe. That is, the failure phenomenon and the failure cause are represented as a list in which a state variable (qualitative variable) representing the state of the system and its qualitative value are paired.

On the other hand, descriptions of generated physical phenomena representing physical phenomena which must appear in a physical causal network and non-generated physical phenomena representing physical phenomena which should not appear are arbitrary. Each of these generated physical phenomena and non-generated physical phenomena includes a plurality of physical phenomena (Phy
1, Phy5, Phy20, etc.), and by describing these, as described later, the qualitative inference unit 4 can more efficiently generate a network that accurately reflects the intention of the failure analyst by qualitative inference. become.

Further, the physical phenomenon knowledge base 2 stores descriptions of various physical phenomena as shown in FIG. FIG. 2 shows only one physical phenomenon (physical phenomenon name: Phy1). The description of the physical phenomenon includes a qualitative initial state for the occurrence of the physical phenomenon and a qualitative end state when the physical phenomenon ends. That is, in the initial state, the state necessary for the physical phenomenon to occur is described as a relational expression and a list of a set of qualitative state variables and their qualitative values, and the relational expression causes the physical phenomenon to occur. It describes the relationships between the state variables needed for this. In the end state, a list of pairs of qualitative state variables and qualitative values at the time when the physical phenomenon ends is described.

Here, a physical phenomenon refers to a phenomenon that changes a certain state to another state. In the description of a physical phenomenon, a state variable described in an initial state but not described in an end state is a physical phenomenon. Is a condition necessary for the occurrence of a phenomena, and does not change due to the occurrence of a physical phenomenon. On the other hand, a state variable that is not described in the initial state but is described in the end state may be any variable in order to cause a physical phenomenon, and is a variable that defines the state after the end of the physical phenomenon.

FIG. 6 shows a description of a physical causal network generated by qualitative inference. A physical causal network is represented by a directed graph in which system states are nodes and physical phenomena are links.The starting point of the network is the fault cause input by the fault analyst, and the end point of the network is the fault input by the fault analyst. Phenomenon (failure state). In addition, from the cause of the failure to the failure phenomenon (failure state)
Some intermediate states are also created on the way to the network.

Here, it is possible for the designer to directly describe this physical causal network as an input.
Describing all processes from a failure cause to a failure phenomenon at a physical phenomenon level takes an enormous amount of time and is not realistic. In addition, analysts do not always consider at the physical phenomenon level, and there is also a problem that empirical elements are likely to be involved and the physical validity of the causal network is not guaranteed. Therefore, it is necessary to automatically generate a description of a physical phenomenon level from a partial description, and in the present invention, using a qualitative inference, a failure analyst obtains a failure cause and a failure phenomenon obtained as a result of a failure analysis. Enter a set of
Based on that, it automatically guides the physical causal network.

FIG. 7 shows a basic procedure of the qualitative inference processing performed by the qualitative inference unit 4. This qualitative inference process
This is performed by inferring the physical causal network for each node, and ends when a node that is the end point of the inference is reached (step S1). In this inference process, the qualitative inference unit 4 extracts all applicable physical phenomena from the physical phenomenon knowledge base 2 based on the current system state (qualitative variables and their values) (step S2), and Select one of the phenomena (step S3),
Apply to the current system state (step S4) to generate a new system state. Then, such processing is repeatedly performed until a failure state (inference end point) input by the failure analyzer can be reached.

If there are no applicable physical phenomena before the failure state can be reached (step S5),
Return to the previous system state (intermediate state node) and look for another applicable physical phenomenon and apply it. Here, as described above, when the occurrence physical phenomenon is described in the failure case input from the failure analysis case input unit 3, the application priority of this physical phenomenon is maximized. When a non-occurring physical phenomenon is described, the physical phenomenon is excluded from the search target when searching the physical phenomenon knowledge base 2.

When the physical causal network is formed in this way, this physical causal network is presented to the failure analyst, and after approval is received, the failure analysis case base 5
And stores the analysis result reflecting the intention of the failure analyst. Note that if the analysis result is different from the analysis intended by the failure analyst, that is, if the generated physical causal network is not approved by the failure analyst,
One of the following two processes is executed.

(1) By default, ordinary inference is performed in which the obtained physical causal network is regarded as an inference failure and returns to the previous inference to search for another physical phenomenon. At this time, the failure analyst can specify an acceptable part of the physical causal network obtained as a result of the inference so that the specified part is not re-inferred. this is,
A specific node in the physical causal network (intermediate state)
Alternatively, it can be performed by marking links (physical phenomena), whereby re-inference can be performed efficiently. (2) The failure analyst directly modifies the physical causal network. In this case, after the correction is completed, the qualitative inference unit 4
Performs a check to see if the physical causal network created by the failure analyst is correct. If the result is not correct, re-inference is performed, and a process of again requesting the failure analyzer for approval is performed.

By using the failure analysis case knowledge base 5 in which the physical network is stored as described above, a designer who has faced a failure can obtain effective case information for countermeasures. That is, when the designer inputs a case search request describing a failure state from the failure analysis case output unit 13, the failure analysis case search unit 12 searches the failure analysis case knowledge base 5 based on this request and finds the same failure state. The failure analysis case is taken out and output from the failure analysis case output unit 13 to the designer. As described above, when a failure occurs in a designed product, the designer can obtain a failure analysis case having the same failure state by inputting a case search request describing the failure state.

Even when a plurality of cases are searched, since the failure analysis case is described at the physical phenomenon level, the occurrence of a physical phenomenon or an intermediate state in the obtained physical causal network is determined. By checking, the designer can determine a correct failure analysis case. Further, even when a new design is performed or a design result is corrected, the possibility of occurrence of a failure can be checked in advance by searching the failure analysis case knowledge base 5 using those as a cause of the failure. it can.

In the above-described example, the qualitative inference unit 4 performs ordinary qualitative inference in which the inference is sequentially performed step by step from the inference start point (failure cause) to the inference end point (failure phenomenon) as shown in FIG. However, in this inference method, there is a concern that the efficiency is low, and particularly when the types of applicable physical phenomena increase, the amount of calculation increases and a realistic solution may not be obtained. Therefore, in another embodiment (FIGS. 9 to 12) according to the present invention, unlike the general qualitative inference method, an initial state (cause of failure) which is an inference starting point and an ultimate state which is an inference end point should be finally derived. By using the feature of the present invention that a state (failure phenomenon) is given, speeding up of the inference processing performed by the qualitative inference unit 4 is realized.

FIG. 9 shows an example in which priorities are set for applied physical phenomena and inference processing is performed. Qualitative reasoning part 4
When performing inference in, there are generally a plurality of physical phenomena that can be applied (can occur) in a certain state, and which of these physical phenomena is selected and applied depends on the inference efficiency. Has a significant effect. Here, when comparing both states of the failure cause and the failure phenomenon, a variable holding the same value is called an equivalent variable, and a variable not holding it is called a non-equivalent variable. That is, when there are a plurality of applicable physical phenomena in a certain state, the priority of the physical phenomenon that changes the value of the equivalent variable is set lower than the priority of the physical phenomenon that does not change the value of the equivalent variable, Attempt to apply from the higher priority physical laws. Of course, the value of the variable temporarily changes between the initial state and the end state, and may return to the original value again until the final state, but it is possible that the value will change as it is. Is higher.

In this example, the priorities are determined for the physical phenomena as described above, and inference is performed using the highest priority in the order of the highest priority. As a result, if it is not possible to reach the inference end point, a process of applying the next priority physical phenomenon and performing inference is performed. As shown in FIG.
In the inference in this example, inference is started from the cause of the failure, and the physical phenomena of the first priority are sequentially applied. The numbers given to the links in the figure are the priorities of the applied physical phenomena.

As a result, it is assumed that an intermediate state 2-1-1 is reached, and there is no applicable physical phenomenon.
In this case, inference is performed by applying the physical phenomenon of the second priority in the intermediate state 2-1. In this inference, it is assumed that there is no more applicable physical phenomenon in the intermediate state 2-1-2. In this case, the process returns to the intermediate state 2 and reaches the intermediate state 2-2 by applying the physical phenomenon of the second priority.
Eventually, the failure cause is reached. By applying the physical phenomena with high priority to inference in this way, the search space can be limited, and quick inference processing can be realized.

FIG. 10 shows an example in which inference processing is performed by performing a horizontal search one step at a time from both the failure cause and the failure phenomenon. There are two general inference methods, a vertical priority search and a horizontal priority search. In the case of the vertical priority, there is a problem that the search time becomes longer if a deep branch exists in any of the options, and there is a method of performing a horizontal priority search in response to the problem. In this example, the horizontal priority search is performed. However, in the case of simply performing the horizontal priority search, when the number of applicable physical phenomena increases, the search space rapidly increases and the search time increases. There are cases.

Therefore, in the present invention, since the inference start point and the inference end point are given, in this example, the horizontal search is performed one step at a time from both the inference start point and the inference end point. Is raised. That is, as shown in FIG. 10, first, inference is performed by one step from the cause of failure, intermediate state 1, intermediate state 2, and intermediate state 3 are respectively derived, and then inference is performed by one step in the reverse direction from the failure phenomenon. , Intermediate state Y, intermediate state 2-2-2, and intermediate state X, respectively. Next, this time, one step of inference is performed from the intermediate state 1 to derive the intermediate state 1-1, the intermediate state 1-2, and the intermediate state 1-3. Similarly, the inference is performed from the intermediate state 2 and the intermediate state 3, respectively. Is performed one step. Next, one-step inference is performed from the intermediate state Y, and similarly, the intermediate state 2
-2-2. One-step inference is performed from the intermediate state X. As a result, the intermediate state 2-2-2 reaches the intermediate state 2-2 derived from the intermediate state 2. At this point, the cause of the failure → the intermediate state 2 → the intermediate state 2-2 → the intermediate state 2 -2
The link of -2 → failure state is completed.

As described above, in this example, the search space can be limited, and N physical phenomena applicable to each intermediate state can be defined.
If the number of physical phenomena necessary to reach the failure phenomenon from the cause of the failure is M, in the normal lateral priority search,
At most {N × (NM-1) / (N-1)} searches are required. On the other hand, when inference is performed from both directions as in this example, at most ｛2 × N × (N (M / 2)
-1) / (N-1) searches may be performed, and the number of searches can be reduced by N (M / 2) times as compared with the normal horizontal priority search.

Here, when the failure analyst performs a failure analysis, if an intermediate state is obtained in addition to the failure cause and the failure phenomenon, the intermediate state is determined in accordance with the failure phenomenon and the failure cause. Can increase the efficiency of inference. FIG. 11 shows an input form used when the failure analyst inputs data to the failure case input section 3. This input form is an extension of the input form shown in FIG. is there. Any number of intermediate states can be described.

The intermediate states are classified as intermediate state 1, intermediate state 2, intermediate state 3,.
Then, the qualitative inference unit 4 receiving this input first performs inference with the failure cause as the inference start point and the intermediate state 1 as the inference end point. Then, the obtained partial physical causal network (the physical causal network from the failure cause to the intermediate state 1) is presented to the failure analyst, and if approval is obtained, the intermediate state 1 is subsequently inferred. Inference is performed using the start point and the intermediate state 2 as the inference end point. In this way, inference is performed with the intermediate state n as the inference start point and the failure phenomenon as the inference end point, and the inference result is approved by the failure analyst, so that the physical causal network from the failure cause to the failure phenomenon is obtained. To complete.

FIG. 12 shows a process of inference when an intermediate state is explicitly input by a failure analyzer. In this example, the above-described bidirectional inference is also performed. The intermediate state 2-2 in the figure is the intermediate state specified by the failure analyst. In this case, first, inference is performed with the failure cause as the inference start point and the intermediate state 2-2 as the inference end point. You. As a result, first, the intermediate state 1, the intermediate state 2, and the intermediate state 3 are derived from the cause of the failure, and then the intermediate state 2
By performing inference from −2, the intermediate state 2 is reached, and a partial physical causal network from the failure cause to the intermediate state 2-2 is completed. Similarly, inference is performed using the intermediate state 2-2 as the inference start point and the failure phenomenon as the inference end point, and a physical causal network from the intermediate state 2-2 to the failure phenomenon is obtained. In this way, finally, the failure phenomenon → intermediate state 2 → intermediate state 2-2 → intermediate state 2-2
2. A physical causal network leading to the failure phenomenon is obtained.

Each of the inference methods shown in FIGS. 9 to 12 may be implemented independently in the qualitative inference unit 4, or may be implemented in any combination in the qualitative inference unit 4. Good.

[0038]

As described above, according to the failure analysis case accumulating apparatus according to the present invention, the results of the failure analysis are stored in the form of a physical causal network. The physical causal relationship is also found out, so that the user does not reuse the wrong result, it is possible to take appropriate countermeasures, the failure mechanism can be grasped, and the designer's skill can be improved. Further, when searching for the stored analysis results, a physical phenomenon that is currently in the process of leading to a failure can also be used as a search key, so that search candidates can be narrowed down and a quick search can be realized.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a failure analysis case accumulation device according to an embodiment of the present invention.

FIG. 2 is a conceptual diagram illustrating a failure analysis case registration process.

FIG. 3 is a conceptual diagram illustrating a search process of a failure analysis case.

FIG. 4 is a diagram showing an example of an input form for a failure analyzer to describe a failure analysis result.

FIG. 5 is a conceptual diagram illustrating a description of a physical phenomenon stored in a physical phenomenon knowledge base.

FIG. 6 is a conceptual diagram illustrating a physical causal network.

FIG. 7 is a flowchart illustrating a basic processing procedure of a qualitative inference unit.

FIG. 8 is a diagram illustrating a process of general qualitative inference.

FIG. 9 is a diagram illustrating a process of qualitative inference in which a physical phenomenon is prioritized.

FIG. 10 is a diagram illustrating a process of qualitative inference in which inference is performed from both directions.

FIG. 11 is a diagram showing another example of an input form for a failure analyzer to describe a failure analysis result.

FIG. 12 is a diagram illustrating a process of qualitative inference in which inference is performed using an intermediate state.

[Explanation of symbols]

1 ··· Failure analysis case accumulation device 2 ··· Physical phenomenon knowledge base 3 · Failure analysis case input unit 4 · Qualitative inference unit 5 · Failure analysis case knowledge base 11 ·
..Fault analysis case suspension device, 12: Failure analysis case search unit, 13 ... Fault analysis case output unit,

Claims (6)

[Claims] 1. A failure analysis case accumulator for creating a network representing a causal relationship between a failure phenomenon and a failure cause and storing and storing the network in a database, wherein the failure phenomenon and the failure cause corresponding to the failure phenomenon are set as a set. An input means to be input, a physical phenomenon knowledge base storing a qualitative state for the occurrence of the physical phenomenon and a qualitative state at the time when the phenomenon ends, and the input failure phenomenon Qualitative inference means for generating, by qualitative inference, a physical causal network which is a chain of physical phenomena connecting the failure phenomenon and the failure cause by referring to the physical phenomenon knowledge base based on a set of And a failure analysis case knowledge base for storing and storing the generated physical causal network. 2. The failure analysis case storage device according to claim 1, wherein a failure analysis case search means for searching a corresponding physical causal network from a failure analysis case knowledge base based on the input failure phenomenon. And a failure analysis case output means for outputting a physical causal network. 3. The failure analysis case accumulating device according to claim 1, wherein the qualitative inference means presents the generated physical causal network to the analyst, and based on the approval from the analyst, A failure analysis case accumulation device characterized by storing a statistical causal network in a failure analysis case knowledge base. 4. The fault analysis case accumulation device according to claim 1, wherein the qualitative inference means is applied based on a change in state variables at an inference start point and an inference end point. Prioritize physical phenomena,
A failure analysis case accumulation device that performs qualitative inference by a method of sequentially inferring according to the priority. 5. The fault analysis case accumulating apparatus according to claim 1, wherein the qualitative inference means advances the inference from both the inference start point and the inference end point in the horizontal direction. A failure analysis case accumulation device that performs qualitative inference. 6. The fault analysis case accumulating device according to claim 1, wherein the input means also inputs a known intermediate state between the fault phenomenon and the fault cause, and performs qualitative inference. A failure analysis case accumulation device, wherein the means performs qualitative inference by a method of inferring for each section divided in the input intermediate state.