JPH04299740A - Knowledge information processor - Google Patents

Abstract

PURPOSE:To acquire the knowledge information in response to a user request by setting the priority of knowledge information for the trouble treatment. CONSTITUTION:When various types of information on the trouble diagnosis are inputted through an information input means with the operating input of a user, a factor inference means infers the trouble factor hypotheses and each assurenace of these hypotheses based on those input information and the knowledge information stored in a knowledge base. These inferred trouble factor hypotheses are exclusive to each other end one of them is considered as a genuine trouble factor. Then a treatment detection means detects the knowledge information of at least a trouble treatment to each of those inferred trouble factor hypotheses. A priority deciding means decides the priority of the knowledge information on the detected trouble treatments based on the relative enclosing relation among the knowledge information set by each of relevant contents. Thus the user can carry out the trouble treatment based on the decided priority and without using the user's own judgement.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a knowledge information processing device used for equipment failure diagnosis.

0002

[Prior Art] Currently, AI (Artificial
Fault diagnosis systems, which are knowledge information processing devices that infer the causes of failures of various devices by using intelligence (Intelligence), have been developed. In such a failure diagnosis system, experts generally store various knowledge information related to failure diagnosis of target equipment in advance in a knowledge base, and general users can input various information related to failure diagnosis as desired. Input using information input means such as a keyboard. Then,
The cause inference means infers the cause of the failure from the input knowledge information and the knowledge information in the knowledge base, and the action detection means instructs the knowledge information of at least one failure treatment for each of the plurality of inferred failure cause hypotheses. is detected from the knowledge base,
This knowledge information on how to deal with the detected failure is output from a display or the like. In other words, in such a fault diagnosis system, by having an expert input the necessary knowledge information into the knowledge base in advance, even a general user without specialized knowledge can easily input information without consulting an expert. You can obtain specialized knowledge information.

0003

SUMMARY OF THE INVENTION The above-mentioned fault diagnosis system can easily provide specialized knowledge information to general users.

However, in the conventional fault diagnosis system, at least one fault treatment knowledge information is output for each of the plurality of fault cause hypotheses inferred as described above. The amount of knowledge information may become so large that general users may not be able to discern what is valid knowledge information. In other words, if a general user executes a fault action based on one of the multiple pieces of knowledge information output in this way, and the fault condition is not resolved as a result, knowledge of the effective fault action to be executed next Information is unclear.

[0005] Furthermore, when troubleshooting is performed sequentially until the fault condition is resolved according to a plurality of pieces of output knowledge information, for example, the monetary cost, time cost, success probability, etc. may be limited depending on the user's situation. However, conventional fault diagnosis systems that cannot meet such conditions have difficulty providing knowledge information that meets user requirements.

[0006]

[Means for solving the problem] The invention according to claim 1 includes:
A knowledge base is provided that stores various kinds of knowledge information related to failure diagnosis of target equipment in a freely updatable manner, an information input means is provided for inputting various kinds of information related to failure diagnosis, and knowledge information inputted from this information input means is provided. A cause inference means is provided for inferring a plurality of failure cause hypotheses that are assumed to include the true cause of failure from knowledge information stored in a knowledge base, and a degree of certainty for each of the failure cause hypotheses. A treatment detection means is provided for detecting at least one failure treatment knowledge information from the knowledge base for each of the plurality of failure cause hypotheses detected by the treatment detection means, and an information input means is provided for determining the priority of the failure treatment knowledge information detected by the treatment detection means. A ranking determining means is provided to generate rankings according to knowledge information input from the system.

[0007] In the invention as claimed in claim 2, the priority order of the knowledge information of the failure treatment detected by the treatment detection means is partially omitted based on the relative inclusion relationship between the knowledge information set in advance according to the content of each piece of knowledge information. generate.

[0008] The invention according to claim 3 is a result prediction method that appropriately calculates monetary costs, time costs, and success probabilities that are sequentially updated when failure treatment knowledge information is sequentially executed according to the priority order generated by the order determining means. Information input means is provided for inputting a maximum cost value that limits one of the monetary cost and time cost calculated by the result prediction means, and a minimum probability value that limits the success probability. A best treatment selection means is provided for selecting the knowledge information of a failure treatment that satisfies the input minimum probability value and shows the other cost being the minimum value according to the maximum cost value.

[0009]

[Operation] In the invention as set forth in claim 1, various kinds of knowledge information related to failure diagnosis of target equipment are stored in a knowledge base in a freely updatable manner, and various kinds of information related to failure diagnosis are inputted from an information input means, From this input knowledge information and the knowledge information stored in the knowledge base, the cause inference means infers a plurality of failure cause hypotheses that are assumed to include the true cause of the failure and the confidence level of each of the failure cause hypotheses. A treatment detecting means detects at least one failure treatment knowledge information from the knowledge base for each of the plurality of inferred failure cause hypotheses, and a ranking determining means determines the priority of the detected failure treatment knowledge information. By generating the knowledge information input from the input means and executing troubleshooting according to the determined priority order, general users can execute troubleshooting in an effective order without having to make their own judgments. .

[0010] In the invention as set forth in claim 2, the priority determining means determines the priority of the knowledge information of the failure treatment detected by the treatment detecting means based on the relative inclusion relationship between the knowledge information set in advance according to the content of each piece of knowledge information. By omitting and generating the knowledge information, it is possible to reduce the number of pieces of knowledge information for setting the priority order.

[0011] In the invention as set forth in claim 3, the result predicting means appropriately calculates the monetary cost, time cost, and success probability that are sequentially updated when the knowledge information of failure treatment is sequentially executed according to the priority order generated by the ranking determining means. A maximum cost value that limits one of the calculated monetary cost and time cost and a minimum probability value that limits the probability of success are input from the information input means, and the input minimum probability value is satisfied and By selecting the knowledge information of the failure treatment whose cost is the minimum value according to the maximum cost value, the best treatment selection means can provide the best knowledge information that satisfies various conditions.

0012

[Embodiment] An embodiment of the present invention will be explained based on the drawings. First, the failure cause diagnosis system, which is a knowledge information processing device, has a knowledge base that stores various knowledge information related to failure diagnosis of target equipment in a RAM (Random).
An information input means, such as a keyboard, is provided for inputting various information related to failure diagnosis, such as a maximum cost value and a minimum probability value, which will be described later. Furthermore, this fault diagnosis system includes a cause inference means for inferring multiple failure cause hypotheses and the reliability of the failure cause hypotheses from the various kinds of knowledge information as described above, and a cause inference means for inferring failure cause hypotheses inferred by this cause inference means. A treatment detection means for detecting knowledge information on failure treatment from a knowledge base, a ranking determination means for generating a priority order of knowledge information on failure treatment detected by this treatment detection means, and a priority determination means for generating a priority order of the knowledge information on failure treatment detected by this treatment detection means. Result prediction means that appropriately calculates the monetary cost, time cost, and success probability when knowledge information is executed sequentially, and best treatment selection means that selects knowledge information for troubleshooting according to the limited value of the calculation result of this result prediction means, etc. For example, if the CPU (Central
Processing Unit) is formed by firmware, etc.

[0013] With such a configuration, in this fault diagnosis system, when various types of information related to fault diagnosis are input from the information input means through the user's operation input, the input knowledge information and the information stored in the knowledge base are A cause inference means infers a plurality of failure cause hypotheses and the confidence level of each failure cause hypothesis from the knowledge information. At this time, the multiple inferred failure cause hypotheses are mutually exclusive, and one of them is considered to be the true cause of the failure. Next, for each of the plurality of inferred failure cause hypotheses, at least one failure treatment knowledge information is detected from the knowledge base by the treatment detection means, and the detected failure treatment knowledge information is prioritized. The determining means generates the knowledge information according to the relative inclusion relationship between the knowledge information set in advance according to each content.

[0014] Here, in this failure diagnosis system, when the knowledge information of failure treatment is sequentially executed according to the priority order generated as described above, the result is predicted based on the sequentially updated monetary cost, time cost, and success probability. When the means appropriately calculates the calculated maximum cost value that limits one of the calculated monetary cost and time cost, and the minimum probability value that limits the probability of success, this The knowledge information of a failure treatment that satisfies the input minimum probability value and shows that the other cost is the minimum value according to the maximum cost value is selected by the best treatment selection means and outputted by displaying on a display, printing on a printer, or the like.

[0015] By doing this, in this fault diagnosis system, priority is set for the knowledge information of a large number of fault actions, so that even general users without specialized knowledge can effectively carry out multiple fault actions. Can be executed in order. Furthermore, in this fault diagnosis system, when troubleshooting is performed sequentially until the fault condition is resolved according to a plurality of pieces of knowledge information output according to the priority order, financial costs, time costs, success probability, etc. are calculated in advance according to the user's situation. Since the information can be limited, it is possible to obtain knowledge information on troubleshooting that corresponds to the user's request.

[0016] Various operations of such a failure diagnosis system will be described in detail below. First, in this fault diagnosis system, as described above, the set CS of knowledge information ci of multiple fault causes, which is a hypothesis inferred by the cause inference means, and the confidence level P(cj) of each fault cause hypothesis are expressed as CS={ ci: Cause of failure; c1, c2...cm} P(cj): Cause of failure cj is the possibility of being the true cause of failure, and a specific example is: CS={low battery voltage, defective plug...}P( c.
j)=0.9, 0.8, etc.

[0017] Here, the set AS of knowledge information ai of failure treatments detected from the knowledge base by the treatment detection means for each of the failure cause hypotheses as described above includes all failure treatments effective for each failure cause. is the union of knowledge information ai, and AS={ai: troubleshooting; a1, a2...am},
A specific example is AS={charging the battery, cleaning the plug...}. At this time, the information accompanying the knowledge information ai of each failure treatment is as follows: TC(ai) : Time cost required to execute failure treatment ai MC(ai) : Monetary cost required to execute failure treatment ai E(ai , cj): Success probability when troubleshooting ai is executed for failure cause cj. As a specific example, TC (plug cleaning) = 10 minutes MC (plug cleaning) = 1000 yen E (plug cleaning, plug cleaning) (defective) = 0.9, etc. Note that the success probability in this case is E(ai, cj)=P(a
i=OK;cj) and {P(ai=OK;cj)+
P(ai=NG;cj)=1.0}.

[0018] Next, the ranking determining means generates a priority ranking for the above-mentioned fault treatment knowledge information according to a relative inclusion relationship set in advance according to the content. here,
The inclusion relationship between the knowledge information of this failure treatment will be specifically explained. For example, the troubleshooting action "battery replacement" includes the roles of "battery charging" and "battery fluid replenishment," so if the fault condition is not resolved even after "battery replacement" is performed, This means that there is no possibility that the fault condition will be resolved even if you perform "battery charging" or "battery fluid replenishment." In other words, when the knowledge information ai and aj of troubleshooting are included in the knowledge information ak of troubleshooting, ai<ak aj<ak, and in the above specific example, "battery charging"<"batteryreplacement""battery fluid replenishment" ” < “Battery replacement”. Therefore, based on the order of such inclusion relationships, a set POAS (Partial
Order Action Set), POAS
={(ai, aj<ak), (an<am)...}.

Next, as described above, knowledge information (failure treatment sequence) is created in which priorities are determined for the set of failure treatment hypotheses obtained with their respective degrees of certainty as a result of inference. This is carried out according to a three-step procedure: 1st step: parameter specification, 2nd step: generation of fault action sequence, 3rd stage: selection of optimal fault action sequence. Therefore, each procedure will be explained in detail below along with specific examples.

First, the first step of specifying parameters is to obtain the necessary parameters through the user's operation input using the information input means, as described above, in order to form a troubleshooting sequence corresponding to the user's request. It has become. For example, when prioritizing time cost for troubleshooting, set the time limit TLimit as the maximum cost value and the success probability POK (0<POK<1) that is desired to be achieved within this time.
In addition, when giving priority to monetary cost, specify a limit amount MClimit as the maximum cost value and a success probability POK that is desired to be achieved within this amount. Specifically, [TClimit=60min, POK=0.
98] (Repair the failure with a probability of 98% within one hour) [MClimit=50000, POK=0.80] (
There is an 80% chance of repairing the malfunction for less than 50,000 yen.

[0021] Next, in the generation of a fault treatment sequence, which is the second step procedure, knowledge information is extracted from the set AS of knowledge information of failure treatments effective for eliminating the cause of the failure and a permutation is created. Permutations that conflict with the partial order relationship based on the inclusion relationship between knowledge information on failure treatment are deleted. In other words, set A
If S consists of 3 pieces of knowledge information, the number of permutations is 6 (
=3! ), but those that do not satisfy POAS are detected and deleted from this failure treatment sequence. Specifically, AS=
(a1, a2, a3), POAS=(a1<a2), the permutation S1 created from AS is S1={(a1, a2, a3), (a1, a3, a2)
, (a2, a1, a3), (a2, a3, a1), (a
3, a1, a2), (a3, a2, a1)}, but
This includes S2={(
a2, a1, a3), (a2, a3, a1), (a3,
a2, a1)} exists, so the set AQ of failure treatment sequences
S (Action Queue Set) is AQS=S1-S2={(a1, a2, a3), (a1
, a3, a2), (a3, a1, a2)}. Therefore, from the result of such a generation step, a set AQS of failure treatment sequences is obtained as AQS={AQ1, AQ2...AQi...AQm}. Note that AQi in the above equation is AQi={a
i1, ai2, ai3...ain(i)}(n(x)is
length of AQx).

[0022]The third step, which is the selection of the optimal fault treatment sequence, involves appropriately calculating the probability of success when the failure treatments in each failure treatment sequence are sequentially executed, and calculating the time cost and monetary cost in each case. One failure treatment sequence that is most suitable for the user's request is selected in consideration of cost. For example, the set CS of knowledge information ci of failure causes assumed for a failure phenomenon by the inference of the cause inference means and the confidence level P(cj) of each failure cause hypothesis are CS={c1, c2, c3, c4 } P(c1)=0.6, P(c2)=0.2, P(c3)
=0.1, P(c4)=0.1 {P(c1)+P(c2
)...+P(c4)=1.0}, and at this point, as mentioned above, the set AQS of failure treatment sequences is obtained as AQS={AQ1, AQ2, AQ3, AQ4, AQ5}
AQ1={a1, a2, a3, a4, a5} AQ2={
…} … Assume that it is obtained as follows. And AQ as mentioned above
When AQ1 is selected and presented as the most appropriate fault action sequence from S, the user will execute fault actions a1 to a5 until the fault condition of the target device is resolved, as illustrated in FIG. .

[0023] Next, when troubleshooting is performed sequentially as described above, by appropriately recalculating the certainty factor of the cause of the fault, the success probability, time cost, and The means for determining the monetary cost and the means for selecting a failure treatment sequence that optimizes these success probabilities and each cost will be described in detail below.

[0024] First, assuming an actual situation, what the user wants to know about the failure treatment sequence is the probability of success at which failure treatment will be executed to eliminate the failure condition. As mentioned above, for each failure treatment ai, the measure of effectiveness against the failure cause cj is the success probability E(ai
, cj). Furthermore, if the failure state is not resolved even after a certain failure treatment is executed (NG), the reliability of each failure cause changes before and after this, so recalculation is necessary. This calculation is calculated using the following formula according to Bayes' theorem, using the confidence level P (cj) that the failure cause cj is the true failure cause, and the success probability E (ai, cj) of failure treatment. It will be done.

Probability that the fault condition will be resolved as a result of executing the fault action ai.

[0026]

[Math 1]

Probability that the fault condition will not be resolved as a result of executing fault action ai.

[0028]

[Math 2]

[0029] Confidence level of failure cause cj when failure condition is not resolved after execution of failure treatment ai.

[0030]

[Math 3]

[0031] Therefore, a specific example of the above calculation will be illustrated below with reference to Table 1 below. First, troubleshooting a1
The probability that the fault condition will be resolved when this is executed is as shown below.

[0032]

[Math 4]

[0033]The probability that the fault condition will not be resolved when fault action a1 is executed is P(a1=NG)=1-P(a1=OK)=1.00-
Since 0.54=0.46, the confidence level of the cause of failure cj when the failure condition is not resolved after execution of failure treatment ai is as illustrated below.

[0034]

[Math 5]

[0035] Here, an example of success probabilities of failure treatment obtained as the above calculation results is illustrated as Table 1 below.

[0036]

[Table 1]

[0037] Next, we will calculate the probability that the fault condition will be resolved if the fault treatment is executed continuously, but this event is contrary to the phenomenon in which the fault condition is not resolved if the failure treatment is executed continuously. Therefore, the probability that this fault state will not be resolved is calculated to obtain the probability that the fault state will be resolved. In reality, the certainty of the cause of the failure when the fault condition is not resolved by the first failure treatment a1 is recalculated, and based on this calculation result, the cause of the failure is determined if the failure condition is not resolved by the second failure treatment a2. In this way, the conditional probability that the fault condition is not resolved is calculated for all fault actions in the fault action sequence. Here, the conditional probability of failure of such a failure treatment is P(a1=NG) P(a2=NG;a1=NG) P(a3=NG;a2=NG;a1=NG)... Specifically, as illustrated in FIG. 2, the total value is 1.0. Therefore, the probability of failure in resolving a fault when the n-th fault action is executed in the fault action sequence AQi is determined by a mathematical formula as exemplified below.

[0038]

[Math 6]

Therefore, from this calculation result, the failure treatment sequence AQ
The probability of success when the failure treatment in i is executed up to the nth time is: POK^(AQi, n) = 1-PNG^(AQi, n)
It is required as.

In addition, as described above, the time cost T when executing up to the n-th failure treatment in the failure treatment sequence AQi is
C and the monetary cost MC are each calculated using the following formulas.

[0041]

[Math 7]

[0042]

[Math. 8]

Note that, as illustrated in FIG. 3 and Table 2 below, the time cost TC and monetary cost MC obtained as described above inevitably increase as the troubleshooting is performed sequentially. It has become.

[0044]

[Table 2]

[0045] By the procedure explained so far, when the failure treatment sequence AQi is executed up to the n-th failure treatment, the success probability POK^ (AQi, n) Time cost TC^ (AQi, n) Monetary cost MC^ Since (AQi,n) is calculated, one optimal fault action sequence is selected from the set AQS of fault action sequences using these calculation results. Here, in this failure state system, two types of failure treatment sequence selection strategies are set as illustrated below, and one of these is executed according to the user's request.

Selection strategy 1. To minimize monetary costs on the condition that a failure state is resolved within a limited time.

Selection strategy 2. To minimize time costs on the condition that a failure state is resolved within a limited amount of money.

[0048] First, if the user prioritizes the time cost over the monetary cost for troubleshooting, selection strategy 1 is adopted, and this user selects the limit that is the maximum cost value in the first step procedure, which is the parameter specification. The time TClimit and success probability POK will be specified. Then, Figure 4
As illustrated in , it is determined whether the set AQS of failure treatment sequences is an empty set, and if it is not an empty set, an arbitrary failure treatment sequence aq is selected. Therefore, the maximum value of n within the time cost less than the time limit is calculated for this failure treatment sequence aq, and the success probability in this case is sufficiently higher than the success probability of other failure treatment sequences calculated in advance. It is sequentially confirmed that the monetary cost is the minimum value. Therefore, when these are confirmed, the failure treatment sequence aq is updated as the optimal failure treatment sequence BestAq, the minimum monetary cost is updated, and the process returns to checking whether the failure treatment sequence set AQS is an empty set. Therefore, the above-described operations are sequentially repeated, and BestAq at the time when the failure treatment sequence set AQS becomes an empty set is output as the optimal failure treatment sequence together with the minimum monetary cost.

If the user prioritizes monetary cost over time cost for troubleshooting, selection strategy 2 is adopted, and the limit amount MClimit, which is the maximum cost value, and the probability of success POK are specified. Then, as illustrated in FIG. 5, similar to the selection strategy 1 described above, the maximum value of n within the monetary cost less than the limit amount is calculated for the arbitrarily selected failure treatment sequence aq, and the success probability is calculated. The update of the time cost and time cost are sequentially repeated, and BestAq is outputted as the optimal failure treatment sequence with the minimum time cost.

[0050]

[Effects of the Invention] The invention as set forth in claim 1 provides a knowledge base storing various knowledge information related to failure diagnosis of target equipment in a freely updatable manner, and an information input means into which various information related to failure diagnosis is input. A plurality of failure cause hypotheses that are assumed to include the true cause of failure and the confidence level of each of the failure cause hypotheses are inferred from the knowledge information input from this information input means and the knowledge information stored in the knowledge base. A cause inference means is provided for detecting at least one failure treatment knowledge information from the knowledge base for each of the plurality of failure cause hypotheses inferred by the cause inference means. By providing a ranking determining means for generating the priority of knowledge information of the troubleshooting that has been performed according to the knowledge information inputted from the information input means, the troubleshooting can be executed according to the priority determined by the ranking determining means. General users can perform troubleshooting in an effective order without having to make their own judgments, and even if the number of pieces of inferred troubleshooting knowledge information is enormous, it is possible to provide good work procedures. It is something that you have.

[0051] The invention according to claim 2 partially omits the priority order of the knowledge information of the failure treatment detected by the treatment detecting means based on the relative inclusion relationship between the knowledge information set in advance according to the content of each piece of knowledge information. By providing a generation ranking determining means, it is possible to reduce the number of pieces of knowledge information for which priorities are set, thereby shortening the calculation processing time and speeding up the output operation of knowledge information. It is effective.

[0052] The invention as claimed in claim 3 is a result prediction method that appropriately calculates monetary costs, time costs, and success probabilities that are sequentially updated when failure treatment knowledge information is sequentially executed according to the priorities generated by the ranking determining means. Information input means is provided for inputting a maximum cost value that limits one of the monetary cost and time cost calculated by the result prediction means, and a minimum probability value that limits the success probability. By providing the best treatment selection means for selecting the knowledge information of the failure treatment that satisfies the input minimum probability value and shows the other cost being the minimum value according to the maximum cost value, it is possible to select the best knowledge information that satisfies various conditions. This has the advantage of being able to respond to various requests from users.

[Brief explanation of the drawing]

FIG. 1 is a process explanatory diagram of a processing operation showing an embodiment of the present invention.

FIG. 2 is a process explanatory diagram showing an example of a calculation process.

FIG. 3 is a process explanatory diagram showing an example of a calculation process.

FIG. 4 is a flowchart.

FIG. 5 is a flowchart.

Claims (3)

[Claims] Claims 1: A knowledge base storing various knowledge information related to failure diagnosis of target equipment in a freely updatable manner; information input means into which various information related to failure diagnosis is input; and information input means input from the information input means. cause inference means for inferring a plurality of failure cause hypotheses that are assumed to include the true cause of failure and a confidence level of each of the failure cause hypotheses from knowledge information stored in the knowledge base and knowledge information stored in the knowledge base; a treatment detection means for detecting at least one failure treatment knowledge information from the knowledge base for each of the plurality of failure cause hypotheses inferred by the inference means; and a priority order of the failure treatment knowledge information detected by the treatment detection means. a knowledge information processing device comprising a ranking determining means for generating a ranking according to knowledge information inputted from the information inputting means. 2. Ranking determining means for generating the priority of the knowledge information of the failure treatment detected by the treatment detecting means by partially omitting the priority of the knowledge information on the relative inclusion relationship between the knowledge information set in advance according to the content of each piece of knowledge information. 2. The knowledge information processing device according to claim 1, further comprising a knowledge information processing device. 3. Result prediction means for appropriately calculating monetary costs, time costs, and success probabilities that are sequentially updated when failure treatment knowledge information is sequentially executed according to the priority order generated by the ranking determination means; Information input means is provided into which a maximum cost value that limits one of the monetary cost and time cost calculated by the prediction means and a minimum probability value that limits the success probability are input, and the probability input by the information input means is input. 3. The knowledge information processing apparatus according to claim 1, further comprising a best treatment selection means for selecting knowledge information of a failure treatment that satisfies a minimum value and has a minimum cost according to the maximum cost value. .