JPH05342003A - Diagnostic device - Google Patents

Abstract

PURPOSE:To always instruct optimum diagnosis to users by selecting the suitable instruction based on a diagnostic tree and executing a problem resolution process interactively with users. CONSTITUTION:This device is provided with a knowledge base 3 storing knowledge required for diagnosis, and a diagnostic tree constructing means 5 to construct the diagnostic tree, which respectively expresses the instruction applied to the user and an answer from the user with an instruction node and an answer node and constitutes the combination of possible diagnosis in the shape of a tree corresponding to the instruction node and the answer node, based on the knowledge in the knowledge base 3. Then, the suitable instruction is selected based on the diagnostic tree, and the problem resolution process is executed interactively with the user. On the other hand, the diagnostic tree constructing means 5 calculates the distribution of cause probability at each answer node of the diagnostic tree and the probability of recovery at each instruction node based on the knowledge in the knowledge base 3 and expresses the instruction node having the recovery probability higher than a certain value as a problem resolving state. Therefore, the optimum instruction can be always applied to the user.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a diagnostic device that utilizes a knowledge base to diagnose a malfunction or illness of a diagnostic object such as a machine or a patient.

There is known a diagnostic device that utilizes a knowledge base to make a diagnosis when a failure or illness occurs in a diagnosis target such as a machine or a patient. In this type of conventional diagnostic device, based on the knowledge base and the information of the symptoms associated with the failure or illness, the possible cause of failure and its certainty factor is inferred, and the user is asked questions and tests. After narrowing down the causes and determining that a certain cause is the true cause,
Eventually, the necessary troubleshooting measures were presented.

[0003]

However, in the conventional diagnostic apparatus as described above, the failure treatment is uniquely instructed based on the determination of the cause of the failure, so that the user is not always required to instruct the optimum failure treatment. There was a problem that I could not give. Specifically, the conventional diagnostic device has a problem in that the failure action corresponding to the cause with the highest probability is uniquely instructed even if the action is not effective. Further, the conventional diagnostic apparatus has a problem that it is not possible to perform a diagnosis flexibly corresponding to a diagnostic strategy required by a user.

It is an object of the present invention to provide a diagnostic device which can optimally give a diagnostic instruction to a user and can realize a diagnostic flexibly corresponding to a diagnostic strategy required by the user. There is.

[0005]

In order to achieve the above object, the invention according to claim 1 has a knowledge base in which knowledge necessary for diagnosis is stored, an instruction given to a user and a response from the user. , And a diagnostic tree constructing means for constructing a diagnostic tree in the form of a tree of possible diagnostic combinations of the indicative and reply nodes, based on the knowledge in the knowledge base. However, it is characterized in that an appropriate instruction is selected based on the diagnosis tree and a problem solving process is interactively performed with the user. As a result, the user can always be given the optimum instruction.

According to the second aspect of the invention, the diagnostic tree construction means obtains the distribution of the cause probability at each response node of the diagnostic tree and the restoration probability at each instruction node based on the knowledge in the knowledge base. The feature is that the indicated node having a recovery probability equal to or greater than the value is expressed as a problem solving state. By fixing the problem solving state of the diagnosis to fix the failure, the user can be given a proactive instruction for the treatment, and the user only needs to perform the work according to the instruction from the system, so that even a beginner can easily perform it. Can be used.

According to the third aspect of the invention, an evaluation value obtained by digitizing the distance to the problem solving state is given to each instruction node of the diagnostic tree, and an appropriate instruction is selected based on the evaluation value. The feature is that it is given to the user. This makes it possible to realize a diagnosis flexibly corresponding to the diagnosis strategy required by the user.

The invention according to claim 4 is characterized in that the evaluation value at the designated node of the diagnostic tree is set as an expected value of the cost from the designated node to the problem solving state. As a result, the instruction can be selected accurately.

In the invention according to claim 5, the cost is expressed by a plurality of cost values, a cost distribution ratio is set between the plurality of cost values, and the optimum cost in the diagnostic tree is set based on the cost distribution ratio. It is characterized in that it is adapted to make a calculation and determine a diagnostic strategy. Thereby, the diagnostic strategy can be flexibly determined.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an embodiment of the diagnostic device according to the present invention. The diagnostic device 1 of the present embodiment is adapted to interactively perform diagnosis, that is, problem solving for the diagnosis target OBJ by giving an instruction to the user US and receiving a response from the user US. A user interface unit 2 that functions as an interface, a knowledge base 3, a knowledge base management unit 4 that manages the knowledge base 3, an instruction given to the user US and a response from the user US are expressed by an instruction node and a response node, respectively. A diagnostic tree construction unit 5 for constructing a diagnostic tree, which is a tree-like configuration of possible diagnostic combinations based on these, from information collected from the knowledge base 3.
And a diagnosis control unit 6 that performs control related to diagnosis.

Here, a numerical value of the distance to the diagnosis goal (solved state) is given to each of the instruction nodes of the diagnostic tree as an evaluation value, and the diagnostic device 1 determines the most based on this evaluation value. An appropriate instruction is selected and given to the user US.

More specifically, in the diagnostic device 1 of the present embodiment, in order to diagnose a certain failure phenomenon (an event expressing an abnormal state or an abnormal operation) occurring in the diagnosis target OBJ, the cause hypothesis is assumed. ci and a priori probability P (ci), symptom si and condition probability P (si | cj), fault treatment ti
Information such as the condition probability P (OK (ti, cj) | cj) is stored in the knowledge base 3, and the information can be collected by designating the failure phenomenon to be diagnosed.

In the above information, the causal hypothesis ci is a cause that may cause a failure phenomenon, and when the number of causes is n, there are n causal hypotheses ci of {c1, c2, ..., cn}. It is stored in knowledge base 3. It is assumed that the failure phenomenon is always caused by one cause hypothesis and that several cause hypotheses cannot be established at the same time.

The a priori probability P (ci) is the probability that the causal hypothesis ci is estimated to be the true cause when there is no information, and each causal hypothesis ci {c1, c2, ... ，
a priori probability P (ci) {P (c1), P for cn}
(C2), ..., P (cn)} are given and stored in the knowledge base 3. The prior probability P (c1),
It is assumed that the sum of P (c2), ..., P (cn) is “1”.

Further, the symptom si is various events caused by the causal hypothesis ci. When the number of symptom is m, m symptom Si of {s1, s2, ..., sm} are stored in the knowledge base 3. It is stored. However, the symptom si does not include a failure phenomenon. In addition, it is assumed that each symptom is an independent event.

The conditional probability P (si | cj) is the probability that the symptom si is likely to occur when the causal hypothesis cj holds, and the conditional probability P (si | cj) is given. For example, symptom s1
, The conditional probability P (s1 | cj): {P (s1 | c
1), P (s1 | c2), ..., P (s1 | cn)}, and the symptom sm has a conditional probability P (sm | c
j): {P (sm | c1), P (sm | c2), ..., P
(Sm | cn)}, and these are knowledge base 3
It is stored in.

Further, the fault remedy ti is a work of correcting (solving) a fault phenomenon caused by the certain hypothesis ci by performing a certain work on the hypothesis ci. , {T1, t2, ..., Tr} r fault measures ti are stored in the knowledge base 3. In addition, it is assumed that each failure treatment is an independent event.

The conditional probability P (OK (ti, cj) |
cj) is the probability that the failure is fixed by executing the failure treatment ti when the cause hypothesis cj is established. In addition, O
K (ti, cj) is a fault treatment t for the cause hypothesis cj.
This represents an event in which the failure phenomenon is corrected by executing i. If there are n causal hypotheses cj, for example, the condition probability P (OK (t1, cj) |
cj): {P (OK (t1, c1) | c1), P (OK
(T1, c2) | c2), ..., P (OK (t1, cn)
| Cn)} is given, and for the failure treatment tr, the condition probability P (OK (tr, cj) | cj): {P (OK (t
r, c1) | c1), P (OK (tr, c2) | c
2), ..., P (OK (tr, cn | cn)} are given and stored in the knowledge base 3.

Further, the fault countermeasure ti is applied to the cause hypothesis cj.
Event that the failure phenomenon is corrected by executing
Corresponding to (ti, cj), an event NG (ti, cj) | cj) that the failure phenomenon is not corrected even if the failure treatment ti is executed for the cause hypothesis cj is defined. In this case, the following relationship is established between the two.

[0020]

## EQU1 ## P (NG (ti, cj) | cj) + P (OK (ti, cj) | cj) = 1

As a specific example, if the failure phenomenon is "the engine is not started", the cause hypothesis ci is 3 of {battery abnormality, spark plug abnormality, fuel supply abnormality}.
Three causal hypotheses are given, and three symptoms of {no horn, no sparks, decrease of fuel injection pressure} are given as symptoms si, and {battery charging, spark plug cleaning, fuel Replace the filter}. In this case, the relationship between the failure phenomenon and the causal hypothesis ci is set as shown in FIG. Here, the numerical value attached to the arrow indicating the correspondence is the a priori probability P (c
i) is represented. The relationship between the symptom si and the causal hypothesis ci is set as shown in FIG. Here, the numerical value attached to the arrow indicating the correspondence is the conditional probability P (sj | c
i) is represented. In addition, the cause hypothesis cj and the failure treatment t
The relationship with i is set as shown in FIG. Here, the numerical value attached to the arrow indicating the correspondence is the conditional probability P (O
K (ti, cj) | cj) is represented.

The knowledge base 3 stores various kinds of information as described above for failure phenomena.

Further, the diagnosis control unit 6 of the diagnostic apparatus 1 has a confirmation instruction cki for knowing the fact of the symptom effective for narrowing down the cause, or a treatment instruction doi for making the diagnosis target normal, and further those instructions. The cost is given to the user US through the user interface unit 2.
On the other hand, the user US returns the confirmed symptom to the confirmation instruction cki and the result of the treatment to the treatment instruction doi to the diagnosis control unit 6 via the user interface unit 2. There is. Here, confirmation instruction ck
i, the treatment instruction doi and their costs are defined as follows.

First, the confirmation instruction cki means that the diagnostic device 1 asks the user US a question or causes the user US to perform a test to confirm the fact of the symptom to be diagnosed. {Ck1, ck2, ..., ckp}
It is assumed that p confirmation instructions cki are prepared. One confirmation instruction has a plurality of symptom candidates that are mutually exclusive and can be confirmed. When a confirmation instruction is received, the user replies with one of the plurality of symptom candidates for the confirmation instruction. ing. That is, one confirmation instruction cki
Multiple candidates for symptoms that can be confirmed by answer (ck
i): If it is defined as {plurality of symptom candidates}, the user is supposed to reply one of them.

The treatment instruction doi means that the diagnostic device 1 instructs the user US to perform a fault repair work, and the diagnostic device 1 indicates q treatment instructions {do1, do2, ..., Doq}. It is assumed that doi is prepared. Upon receipt of one treatment instruction, the user takes action according to the treatment instruction, and as a result, the user replies that the failure is fixed (OK) or the failure is not fixed (NG). That is, if the user's response candidate for one treatment instruction doi is defined as answer (doi): {OK, NG}, the user US will reply OK or NG.

As a concrete example, corresponding to FIG.
{Does the horn ring? ， Is the spark ignited? ， What is the measured value of fuel injection pressure? } If the three confirmation instructions cki are prepared, answer (whether the horn sounds?) = {Sounds or not} for the confirmation instruction "whether the horn sounds?"
The candidate of the symptom is presented to the user, and in response to the confirmation instruction of "is the spark igniting?", The answer (is the spark igniting?) = {Non-ignition, ignition} symptom candidate Is presented to the user, and in response to the confirmation instruction “What is the measured value of fuel injection pressure?”, Answer (What is the measured value of fuel injection pressure?)
= {Specified value, less than specified value, more than specified value} symptom candidates are presented to the user.

Further, in the case where three treatment instructions doi of {charging of battery, cleaning of spark plug, replacement of fuel filter} are prepared corresponding to FIG. 2C, for example,
In response to the “charge battery” treatment instruction, the user is presented with two reply candidates, answer (charge battery) = {OK, NG}.

Further, the confirmation instruction cki and the treatment instruction doi have the money cost money-
cost (cki) and time cost time-cost (cki) are given. For convenience of explanation, the cost for the treatment instruction doi is also money-cost (cki), time-cost (ck
i).

As a specific example, the diagnostic control unit 6 of the diagnostic device 1
To the user US via the user interface unit 2,
For example, the money cost and time cost instructions required to measure the fuel injection pressure are presented as money-cost (what is the measured value of the fuel injection pressure?) And time-cost (what is the measured value of the fuel injection pressure?). In response to this, the user can reply with 0 (yen) and 10 (minute), respectively. In addition, for example, it is possible to present monetary cost and time cost instructions required to charge the battery as money-cost (battery charge) and time-cost (battery charge), respectively. The user can reply 1000 (yen) and 30 (minute), respectively.

Next, a diagnostic processing mode of the diagnostic apparatus 1 of this embodiment will be described. The diagnostic device of the present embodiment is intended to instruct the user of an optimal diagnostic procedure in consideration of recovery probability and cost by constructing a diagnostic tree and presenting (diagnosing) instructions based on the diagnostic tree. .. FIG. 3 is a diagram showing an outline of diagnosis. Referring to FIG. 3, first, a user inputs a failure phenomenon to be diagnosed and a cost distribution ratio to the diagnosis apparatus 1 (A1). The cost allocation ratio is used as a parameter for determining the diagnostic strategy by cost calculation. For example, if the ratio of the time cost is increased, the diagnosis will eventually select the instruction that shortens the time. Alternatively, if the ratio of the monetary cost is increased, the diagnosis is performed so that the monetary cost can be resolved without spending as a result.

Next, the diagnostic device 1 collects all the above-mentioned information regarding the input failure phenomenon from the knowledge base or the like (A2) and constructs a diagnostic tree (A3). afterwards,
Based on the constructed diagnosis tree, the optimum instruction is selected and given according to the response from the user (A4).

At this time, the diagnostic tree constructing unit 5 constructs a diagnostic tree by the procedure shown in FIG. That is, first, the diagnosis tree is developed (generated) with two types of nodes, that is, a designated node and a response node, starting from the start node (step T1).

Here, the instruction node expresses a confirmation or treatment instruction presented by the diagnostic apparatus 1, and the response node expresses a symptom candidate or a treatment result (OK / NG) state obtained by confirmation. There is. FIG. 5 is a diagram showing an example of the expanded (generated) diagnostic tree, and FIG. 6 is a diagram for explaining the process of expanding the diagnostic tree. Referring to FIGS. 5 and 6, from the start node, the instruction nodes corresponding to all the confirmation instructions and the treatment instructions are developed. That is, one indicating node is 1
It is developed as one corresponding to one confirmation instruction or treatment instruction. Also, the response node is expanded as a continuation node of the designated node. Here, if the instruction node is the confirmation instruction ck (for example, “what is the value of the fuel injection pressure?”), The symptom candidate answ.
er (ck): {below the specified value, above the specified value, above the specified value}, or when the instruction node is a treatment instruction (for example, "replacement of fuel filter"), from the response candidate {OK, NG}, Expand the reply node. In addition, a designated node is expanded from the response node.

However, in the expansion of the response node, answer
If the symptom of (ck) has already appeared from the starting node to the current node, do not expand it. Further, when all the treatments appear from the start node to the parent node (treatment instruction node), the response node is not expanded. Therefore, the treatment instruction node becomes the terminal node. For example, in FIG. 6, all the treatments appear from the indication nodes # 1 to # n−1, and it is meaningless to continue the diagnosis after performing all the treatments. Does not expand the response node,
The designated node #n is the terminal node.

In the expansion of the indication node, if the confirmation instruction or the treatment instruction is the same as the confirmation instruction or the treatment instruction that has already appeared in the process from the start node to the response node, the same instruction is given in the diagnosis. Do not repeat the directive node so that it does not repeat. Also, the designated node is not expanded from the OK response node.

In step T1, after the diagnosis tree is expanded as described above, the transition (change) of the cause probability is calculated (step T2). Here, the cause probability changes with the progress of the diagnosis process due to the confirmation of the symptoms and the execution of the treatment. Therefore, the transition of the cause probability is obtained by calculating the cause probability at each response node of the diagnosis tree. Specifically, first, the a priori probability of the cause is set to the start node, and the cause probability is calculated at each response node as follows while tracing the node from the start node to the end of the diagnostic tree. , Hold the result.

That is, in the case of the response node of the confirmation instruction, the cause probability P (cj) of the upper response node and the condition probability P
From (si | cj) (known), Bayes' theorem is used to calculate P (cj | si) by the following equation, and this is taken as the cause probability at this response node. Note that this probability P (cj
| Si) is the condition probability of the failure cause cj obtained by confirming the symptom si.

[0038]

[Equation 2]

Further, in the case of the response node of the treatment instruction, the cause probability P (cj) and the condition probability P (OK) of the upper response node are obtained.
From (ti, cj) | cj) (known), Bayes' theorem is used to calculate P (cj | NG (ti, cj)) by the following equation, and this is taken as the cause probability at this response node.
This probability is the condition probability of the failure cause cj when the failure action ti is not corrected by the failure treatment ti. Therefore,
It is not necessary to perform this calculation at OK nodes.

[0040]

[Equation 3]

FIGS. 7A and 7B are diagrams showing an example of calculation of the cause probability at the response node, and FIG. 7A shows an example of calculation of the cause probability at the response node of the confirmation instruction ck. FIG. 7B shows a calculation example of the cause probability at the response node of the treatment instruction do. In addition, FIG.
In the example of (b), it is assumed that there are two causal hypotheses, c1 and c2. In the example of the response nodes s1 and s2 of the confirmation instruction ck shown in FIG. 7A, the cause probability P (cj) of the upper response node s0 is {0.7, 0.3}, and the condition probability P (si | Cj) is as shown in the figure,
Thus, the cause probabilities P (cj | s1) and P (cj | s2) at the response nodes s1 and s2 of the confirmation instruction ck are set as {0.95,0.05} and {0.2,0.8}, respectively. I want it. Further, in the example of the response node of the treatment instruction do shown in FIG. 7B, the cause probability P (c
j) is {0.7, 0.3} and the conditional probability P (OK
When (ti, cj) | cj) is as shown in the figure, the response node of the treatment instruction do, that is, OK
Cause probability P at the NG node of the nodes and the NG nodes
(Cj | NG (ti, cj)) is obtained as {0.2, 0.8}.

After calculating the cause probability at the response node in this way, the diagnostic tree construction unit 5 then calculates the restoration probability of the treatment instruction node (step T3). Here, the recovery probability (P OK) is the probability that the failure has been fixed at some point in the diagnosis. Therefore, this recovery probability (P O
K) becomes larger as the treatment is performed and changes at the treatment instruction node of the diagnosis tree. Here, the recovery probabilities at all the treatment instruction nodes of the diagnostic tree are calculated. In this calculation procedure, the following calculation is performed at each treatment instruction node while tracing the node from the start node to the end node. That is, the treatment instruction node doi
In (treatment ti), the success probability P that the failure is fixed as a result of the treatment is P.
(OK (ti, *)) is the cause probability P of the upper response node
(Cj) and known P (OK (ti, cj) | cj) are used to obtain from the following equation.

[0043]

[Equation 4]

Further, based on the success probability, the failure probability P
(NG (ti | *)) is obtained from the following equation.

[0045]

## EQU00005 ## P (NG (ti, *)) = 1-P (OK (ti, *))

Now, from the start node, the processing instruction node do
The designated nodes appearing in the process up to n are {do1, do2
, ..., Don-1}, the probability that all of them have failed as a result of performing these measures is the product of the failure probabilities P (NG (ti, *)) of these nodes, and is represented by the following equation.

[0047]

[Equation 6] P NG (do0) = 1 P NG (doi) = P NG (doi−1) × P (NG (ti | *)) (1 ≦ i ≦ n)

Then, the recovery probability P which is the probability that the failure is resolved by the node don. OK is expressed by the following equation.

[0049]

[Equation 7] P OK (don) = 1-P NG (don)

In this case, for example, P OK (doi)>
A treatment instruction node satisfying the condition of 0.95 can be defined as a diagnosis goal. In FIG. 8, the treatment instruction node do1
An example of calculation of the above-mentioned restoration probabilities from to to don is shown.

After calculating the recovery probability at the treatment instruction node in this way, the diagnostic tree construction unit 5 calculates the cost of each instruction node (step T4). That is, the diagnostic tree construction unit 5 costs the cost cost (do) required to execute the instruction (cki or doi) at each instruction node.
i) is calculated by the following equation.

[0052]

[Equation 8]

Here, w is the cost allocation ratio, Total Time C
ost is the sum of the time costs of all instructions, Total Money C
ost is the sum of the monetary costs of all instructions.

Next, the diagnostic tree construction unit 5 calculates the evaluation value of each designated node (step T5). Here, the evaluation value of each designated node is defined as the expected value of the cost required from that node to the goal node. A goal node (problem solving state) is defined as a node at which the restoration probability is a certain value (for example, 0.95) or more. Therefore, if the diagnosis progresses to the goal node, it is highly likely that the failure has been corrected. In addition, a node with a small evaluation value is considered to have a low cost from that point to recovery from the failure. The evaluation value defined in this way, that is, the expected value of the cost required to reach the goal node is calculated from the probability of the continuous response node, the evaluation value of the lower designated node, and the cost of the designated node by the following formula. The reason why the smallest evaluation value is selected and used from the lower order designation nodes is that the diagnostic device 1 selects the designation node having the smallest evaluation value at the time of diagnosis.

[0055]

[Equation 9]

FIG. 9 shows a calculation example of the evaluation value of each designated node.

When the diagnostic tree is constructed as described above,
Diagnosis can be performed based on this. Next, a diagnostic processing procedure based on a diagnostic tree in the diagnostic device 1 of the present embodiment will be described. First, when the user US inputs a failure phenomenon, a cost distribution ratio, etc. through the user interface unit 2, the diagnosis control unit 6 builds a diagnosis tree corresponding to the input of the failure phenomenon, the cost distribution ratio, etc. from the user US. To issue a command to the diagnostic tree construction unit 5. Upon receiving this command, the diagnostic tree construction unit 5 extracts necessary information from the knowledge base 3 and performs the processing shown in FIG.
Build a diagnostic tree.

In the actual diagnosis, in the above-described diagnosis tree, the designated node having a low evaluation value is selected from the start node and the node is moved toward the end node. When the user moves to the confirmation instruction node, the user is instructed about the candidate of the symptom to be confirmed and the confirmation method, and when a response is obtained, the user moves to the corresponding response node. When moving to the node of the treatment instruction, the contents of the treatment (procedures, precautions, etc.) are indicated, and as a result, a reply of OK or NG is received. If the trouble is fixed, OK
The diagnosis ends because the node is at the end. When the node moves to the reply node, the node having the lowest evaluation value is selected and moved among the following designated nodes. Unless the fault is corrected by the movement of this node, all the measures appear until the end.

In this way, one optimum instruction can be given to the user according to the diagnosis tree, the user replies to the system with the result of the instruction, and the system selects the optimum instruction in response to the reply. And give it to the user. By repeating such instructions / replies, the diagnosis process proceeds and the solution is reached. Since the instruction includes the troubleshooting, the user can correct the failure only by performing the work according to the instruction.

As described above, in this embodiment, a diagnosis tree is constructed from the knowledge base information expressing / storing the knowledge of the expert, and the cause of the failure phenomenon is narrowed down or divided based on this diagnosis tree. By showing the user an instruction to confirm the symptom and an instruction to perform a troubleshooting for correcting the abnormality causing the failure, it is possible to interactively perform a problem solving process of restoration (correction) of the failure. At this time, the diagnostic tree is
The instruction given by the system and the response from the user are expressed by the instruction node and the response node, respectively, and the possible combinations of diagnoses are structured in a tree shape.
An evaluation value (a numerical value of the distance to the problem solving state) is given to each instruction node of this diagnostic tree. Therefore, select the most appropriate instruction based on this evaluation value and give it to the user. You can

Further, the a priori probability of the cause, the conditional probability of the symptom,
Probability calculation is performed by using the information of the conditional probability of restoration by measures, the distribution of the cause probability at each response node of the diagnostic tree and the restoration probability at the designated node are obtained, and the designated node having the restoration probability greater than a certain value is diagnosed. It is designed to be expressed as a goal (problem-solving state) node in the tree. By fixing the problem-solving state of the diagnosis to fix the failure, it is possible to positively give the user instructions for treatment.
Since the user only has to work according to the instruction from the system, even a beginner can easily use it.

Further, the costs associated with the confirmation of symptoms and the execution of treatment are expressed by a plurality of types of cost values (for example, time cost, monetary cost), a cost allocation ratio is set for these, and the cost allocation ratio The optimal cost calculation in the diagnostic tree is performed to determine the diagnostic strategy. Considering multiple costs in the selection of diagnostic instructions and the simple expression of the cost allocation ratio, the diagnostic strategy Can be flexibly determined.

The evaluation value at the designated node of the diagnostic tree is defined as the expected value of the cost required from the designated node to the goal node, and the expected value of the cost is the cost of the designated node and the probability of each reply node (symptoms). Occurrence probability) and the evaluation value of the indicator node of its lower class based on the minimum value selected,
The evaluation value is calculated at each instruction node while bottoming up the tree from the goal node. By selecting cost and failure solution (recovery) as evaluation values, the instruction can be selected accurately. ..

[0064]

As described above, according to the first to fifth aspects of the invention.
According to the described invention, a knowledge base in which knowledge necessary for diagnosis is stored, an instruction given to a user and a response from the user are expressed by an instruction node and a response node, respectively, and it is possible to use the instruction node and the response node. Diagnostic tree constructing means for constructing a diagnostic tree, which is a tree-like combination of various diagnostics, based on the knowledge in the knowledge base, and selects an appropriate instruction based on the diagnostic tree, and communicates with the user. Since the problem solving process is carried out interactively, it is possible to always give the user an optimum diagnosis instruction.

In particular, in the invention according to claim 2, the diagnosis tree construction means obtains the distribution of the cause probability at each response node of the diagnosis tree and the recovery probability at each instruction node based on the knowledge in the knowledge base. An instruction node having a recovery probability equal to or higher than a value is expressed as a problem solving state, and by deciding that the state of the problem solving of the diagnosis will be fixed, it is possible to positively give the user instructions for treatment. The user only has to perform the work according to the instruction from the system, and even a beginner can easily use it.

Further, in the invention according to claim 3, an evaluation value obtained by digitizing the distance to the problem solving state is given to each instruction node of the diagnostic tree, and an appropriate instruction is selected based on the evaluation value. Since it is designed to be given to the user, the most appropriate instruction can be selected based on this evaluation value and given to the user to realize a diagnosis flexibly corresponding to the diagnosis strategy required by the user.

In the invention according to claim 4, since the evaluation value at the indication node of the diagnostic tree is set as the expected value of the cost from the indication node to the problem solving state, the instruction can be selected accurately. it can.

In the invention according to claim 5, the cost is represented by a plurality of cost values, a cost distribution ratio is set between the plurality of cost values, and the optimum cost in the diagnostic tree is set based on the cost distribution ratio. Since the calculation strategy is used to determine the diagnostic strategy, it is possible to flexibly determine the diagnostic strategy by considering multiple costs in selecting the diagnostic instruction and by using a simple expression of a cost allocation ratio. be able to.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an embodiment of a diagnostic device according to the present invention.

2A is a diagram for explaining a relationship between a failure phenomenon and a causal hypothesis, FIG. 2B is a diagram for explaining a relationship between a symptom and a causal hypothesis, and FIG. It is a figure for explaining the relation with.

FIG. 3 is a diagram for explaining an outline of a diagnostic process of the diagnostic device of FIG.

FIG. 4 is a flowchart showing a procedure for constructing a diagnostic tree in the diagnostic device of FIG.

FIG. 5 is a diagram showing an example of a developed diagnostic tree.

FIG. 6 is a diagram for explaining a process of developing a diagnostic tree.

7A and 7B are diagrams showing an example of calculation of a cause probability at a response node.

FIG. 8 is a diagram illustrating a calculation example of a restoration probability at a designated node.

FIG. 9 is a diagram illustrating a calculation example of an evaluation value of a designated node.

[Explanation of symbols]

 1 Diagnostic Device 2 User Interface Section 3 Knowledge Base 4 Knowledge Base Management Section 5 Diagnostic Tree Construction Section 6 Diagnostic Control Section

Claims (5)

[Claims] 1. A knowledge base in which knowledge necessary for diagnosis is stored, an instruction given to a user and a response from the user are expressed by an instruction node and a response node, respectively, and possible diagnosis is made by the instruction node and the response node. And a diagnosis tree constructing means for constructing a diagnosis tree having a combination of the above in a tree shape based on the knowledge in the knowledge base, selecting an appropriate instruction based on the diagnosis tree, and causing a problem with the user. A diagnostic device characterized in that a solution process is performed interactively. 2. The diagnostic device according to claim 1, wherein the diagnostic tree constructing means is based on knowledge in the knowledge base,
A diagnostic device characterized in that a distribution of cause probabilities at each response node of a diagnostic tree and a restoration probability at each designated node are obtained, and a designated node having a restoration probability of a certain value or more is expressed as a problem solving state. 3. The diagnostic device according to claim 1, wherein each instruction node of the diagnostic tree is provided with an evaluation value which is a numerical value of a distance to a problem solving state, and an appropriate instruction is given based on the evaluation value. A diagnostic device characterized by being selected and given to a user. 4. The diagnostic device according to claim 3, wherein the evaluation value at the instruction node of the diagnostic tree is set as an expected value of the cost from the instruction node to the problem solving state. 5. The diagnostic device according to claim 4, wherein the cost is represented by a plurality of cost values, a cost distribution ratio is set between the plurality of cost values, and the cost distribution ratio is set based on the cost distribution ratio.
A diagnostic device, characterized in that an optimal cost calculation in a diagnostic tree is performed to determine a diagnostic strategy.