Maintenance support system, maintenance support method, and program

Abstract

To eliminate distrust a user can feel for a trouble shooting that utilizes a prediction model of a Bayesian network.SOLUTION: A maintenance support system comprises: a relevance information retaining unit 11 that retains a level of an index indicating a level of structural understanding on a relevancy between a next inquiry and a failure cause as relevance information; a boundary determination unit 12 that determines a boundary of the index indicating the level of structural understanding at which user's understanding of the relevancy between the next inquiry and the failure cause changes from a structural one to a nonstructural one, for the level of the index indicating the level of structural understanding; and a relevancy determination unit 13 that determines whether the relevancy between the next inquiry and the failure cause is a structural one or a nonstructural one on the basis of the boundary determined as the level of the index indicating the level of structural understanding on the relevancy between the next inquiry and the failure cause. On the basis of a determination result, information is displayed to a user.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present invention relates to a maintenance support system, a maintenance support method, and a program for supporting the maintenance work of a user who uses troubleshooting. [Background technology]

[0002] In recent years, artificial intelligence (AI) has been put to practical use in various fields. For example, in the industrial field, a repair recommendation technology such as that disclosed in Patent Document 1 has emerged as a maintenance support system for facilities and equipment. This technology supports maintenance work by inferring, using an inference model, recommendation information including multiple repairs recommended for input information related to an event and a confidence level indicating the degree of accuracy of each repair, and outputting the recommendation information.

[0003] On the other hand, there are cases where Bayesian networks are used for various recommendations. In such cases, a recommendation can be presented based on a comprehensive judgment by collectively performing inference on multiple factors. For example, Patent Document 2 discloses a recommendation technology using a Bayesian network, which presents information useful for a recipient to decide whether or not to accept a recommendation target selected by a prediction model. This technology has a prediction model based on a Bayesian network including multiple nodes that are influential factors for determining a recommendation target, and determines the main influence node that has had the greatest influence on the decision along with the decision of the recommendation target, and presents the main influence node to the recipient as the basis for the recommendation. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2021-22205 [Patent Document 2] JP 2006-127452 A Summary of the Invention [Problem to be solved by the invention]

[0005] By the way, there are cases where troubleshooting involves repeatedly asking questions to a user using a predictive model based on a Bayesian network, which has a complex inference process. In troubleshooting, when a user faces a problem, symptoms, problem locations, and causes are explored through interviews, and measures to solve, avoid, alleviate, or replace the problem are proposed. In such cases, if information derived from a predictive model based on a Bayesian network is continuously presented, users (e.g., maintenance personnel) who have a certain level of knowledge about the target equipment are likely to distrust the content and order of individual questions, which can hinder the speed of troubleshooting. In particular, in the initial phase of introducing a predictive model, users' distrust of the predictive model can lead to an increase in the amount of work required to confirm the order of questions and to provide explanations about the predictive model, which can also cause a decrease in ease of introduction.

[0006] The technology described in Patent Document 2 merely presents information derived from a predictive model in response to a user's distrust of the predictive model, and is not able to grasp when a user's distrust of the predictive model arises or to fundamentally eliminate that distrust.

[0007] Given the above situation, there was a demand for a method to eliminate the distrust that users felt toward troubleshooting that uses Bayesian network predictive models. [Means for solving the problem]

[0008] In order to solve the above problems, a maintenance support system according to one embodiment of the present invention includes a Bayesian network model that estimates a cause of a failure based on the content of a question asked to a user and the answer to the question, and determines a next question; a relevance information storage unit that stores a level of an index indicating a degree of structural understanding of the relationship between the next question and the cause of the failure as relevance information; a boundary determination unit that determines a boundary of the index indicating the degree of structural understanding at which the user's understanding of the relationship between the next question and the cause of the failure changes from structural to non-structural, based on the level of the index indicating the degree of structural understanding of the relationship between the next question and the cause of the failure and the boundary determined by the boundary determination unit, and is configured to display information to the user based on the determination result. Effect of the Invention

[0009] According to at least one aspect of the present invention, it is possible to eliminate the distrust felt by users in troubleshooting using a Bayesian network prediction model, thereby making it possible for users to be convinced of the content and order of troubleshooting questions. Problems, configurations and effects other than those described above will become apparent from the following description of the embodiments. [Brief description of the drawings]

[0010] [Figure 1] 1 is a diagram showing an example of the configuration of an entire system including a troubleshooting recommendation system (TSRS) to which a relevance determination system according to a first embodiment of the present invention is applied. [Diagram 2] 1 is a diagram illustrating an example of the configuration of a relevance determination system according to a first embodiment of the present invention. [Diagram 3] 3 is a diagram showing an example of the structure of a relevance information table according to the first embodiment of the present invention; FIG. [Figure 4] FIG. 4 is a diagram showing an example of a relevance assessment value and a threshold value according to the first embodiment of the present invention. [Diagram 5]FIG. 2 is a diagram showing an example of the structure of an explanation table according to the first embodiment of the present invention. [Figure 6] FIG. 4 is a sequence diagram showing an outline of troubleshooting using the TSRS and the relevance determination system in the first embodiment of the present invention. [Figure 7] 5 is a flowchart showing an example of a procedure for a troubleshooting process performed by the TSRS and the relevance determination system in the first embodiment of the present invention. [Figure 8] FIG. 2 is a diagram showing an example of a user interface screen according to the first embodiment of the present invention. [Figure 9] FIG. 4 is a diagram showing another example of a user interface screen according to the first embodiment of the present invention. [Figure 10] 10 is a flowchart showing an example of a procedure of a troubleshooting process performed by a TSRS and a relevance determination system according to a second embodiment of the present invention. [Figure 11] FIG. 11 is a diagram showing an example of the structure of a relevance table according to the second embodiment of the present invention. [Figure 12] FIG. 11 is a diagram showing an example of a user interface screen according to the second embodiment of the present invention. [Figure 13] 13 is a flowchart showing an example of a procedure of a troubleshooting process performed by a TSRS and a relevance determination system according to a third embodiment of the present invention. [Figure 14] FIG. 13 is a diagram showing an example of a relevance information function according to the third embodiment of the present invention. [Figure 15] FIG. 13 is a diagram showing an example of a test question for a comprehension test in the fourth embodiment of the present invention. [Figure 16] FIG. 13 is a diagram showing an example of an answer result to a test question in the fourth embodiment of the present invention. [Figure 17] FIG. 13 is a diagram illustrating an example of a threshold set in the fourth embodiment of the present invention. [Figure 18] FIG. 13 is a diagram showing an example of a question group list in the fourth embodiment of the present invention. [Figure 19] FIG. 13 is a diagram showing an example of a correct answer rate and a threshold value in the fourth embodiment of the present invention. [Figure 20] FIG. 23 is a diagram showing an example of a department and years of employment of a user in the fifth exemplary embodiment of the present invention. [Figure 21] FIG. 23 is a diagram showing an example of a department, a required number of years of employment, and a correction threshold value according to the fifth exemplary embodiment of the present invention. [Figure 22] FIG. 13 is a diagram illustrating an example of the relationship between years of service and a threshold value in the fifth embodiment of the present invention. [Diagram 23] FIG. 13 is a diagram illustrating an example of a correction threshold value taking into consideration the expertise of each department in the fifth embodiment of the present invention. [Figure 24] FIG. 13 is a diagram showing an example of a threshold set before and after correction in the fifth embodiment of the present invention. [Diagram 25] FIG. 2 is a block diagram showing an example of the hardware configuration of a computer included in each device according to the embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, examples of modes for carrying out the present invention (hereinafter referred to as "embodiments") will be described with reference to the accompanying drawings. In this specification and the accompanying drawings, identical components or components having substantially the same functions are designated by the same reference numerals, and duplicated descriptions will be omitted.

[0012] <First embodiment> [Maintenance support system configuration] First, the overall configuration of a maintenance support system including a troubleshooting recommendation system (hereinafter referred to as "TSRS") to which a relevance determination system according to a first embodiment of the present invention is applied will be described with reference to FIG.

[0013] Fig. 1 is a diagram showing an example of the overall configuration of a maintenance support system including a TSRS to which the relevance determination system according to the first embodiment is applied. As shown in Fig. 1, the maintenance support system 100 is configured by applying a relevance determination system 10 to a TSRS 20 using a Bayesian network.

[0014] The TSRS 20 includes an interview processing unit 21 and a Bayesian network model 22 used by the interview processing unit 21. The interview processing unit 21 transmits and receives questions and answers between the TSRS 20 and the user terminal. The interview processing unit 21 also transmits the cause of the failure and the question calculated by the Bayesian network model 22 to the relevance determination system 10, and receives the determination result from the relevance determination system 10.

[0015] Connected to the TSRS 20 are an administrator terminal used by a model administrator 30 who manages the Bayesian network model 22, and user terminals used by users 41 to 43 of the TSRS 20. The users 41 to 43 have different levels of proficiency in maintenance work. For example, user 41 is a beginner, user 42 is an intermediate user, and user 43 is an expert.

[0016] In the present embodiment, the TSRS 20 performs troubleshooting by repeatedly asking and answering questions between the Bayesian network model 22 and the user via the medical interview processing unit 21, thereby inferring the cause of a failure.

[0017] [Configuration of relevance determination system 10] Next, the configuration of the relevance determination system 10 will be described with reference to FIG. 2 is a diagram showing an example of the configuration of the relevance determination system 10. The relevance determination system 10 is roughly divided into three blocks: a relevance information holding unit 11, a threshold value determination unit 12, and a relevance determination unit 13.

[0018] [Relationship information storage section] The association information storage unit 11 has an association information registration function 111 and an association information storage function 112. In this embodiment, the association means the association between the content of a question in troubleshooting (survey items on the maintenance support system side) and the cause of a failure, and is assumed to be quantified.

[0019] The relationship information registration function 111 is a function that receives input from an external device such as an administrator terminal and registers and updates relationship information. The association information holding function 112 has a function of holding and storing registered association information (association information table 300 shown in FIG. 3, which will be described later).

[0020] [Relationship Information Table] 3 is a diagram showing an example of the structure of the relevance information table 300. The relevance information table 300 includes failure cause group information 301, which is information that groups failure causes, and question content information 302 for narrowing down the failure causes. The question content information 302 is a survey item in which the actual question content is expressed in short words such as keywords so that it can be easily handled by the system. The relevance evaluation value 303 for each combination of the failure cause group information 301 and the question content information 302 is a numerical representation of the user's structural understanding of the relevance between the question content and the failure cause. The relevance evaluation value is set based on information sources that are the basis of the user's knowledge, such as manuals, design documents, and training records.

[0021] When explaining the relationship between the investigation items and the cause of the failure, the more knowledge that the user does not have is included, the less likely it is that the user will be able to understand or accept it. Relationships that are easy for users to understand are assumed to have physical and informational connections between the question content and the cause of the failure. For example, if the "question content" is about the cable connection status and the "cause of the failure" is a broken cable or poor contact, there is a physical connection (structural relationship). Also, if the "question content" is about checking on the BIOS screen and the "cause of the failure" is a faulty OS, there is an informational connection (structural relationship).

[0022] On the other hand, if the "question" is about whether the fan sound changes intermittently and the "cause of the fault" is incorrect wiring, there is no physical / informational connection (non-structural relationship). Relationships that cannot be understood without knowledge of the Bayesian network model are also treated as non-structural relationships.

[0023] The contents of the relevance information (relevance information table 300 ) are not changed by the algorithm of the Bayesian network model 22 .

[0024] [Threshold determination part] The threshold determination unit 12 (an example of a boundary determination unit) has a learning model for threshold determination 121 and a threshold holding function 122. The threshold determination learning model 121 calculates a threshold for a relevance evaluation value using the relevance information stored in the relevance information storage function 112 of the relevance information storage unit 11 and the user's technical understanding or proficiency. The threshold is calculated by a statistical method using the relevance information as learning data. The calculated threshold is stored in the threshold storage function 122.

[0025] The threshold for the relevance evaluation value is an index value (criterion) that indicates the user's structural understanding. Basically, the relevance evaluation value and the threshold are expressed as numbers, but are not limited to numbers. For example, when the relevance evaluation value is classified as "high", "slightly high", "medium", "slightly low", and "low", the threshold (boundary) is set between the relevance evaluation values. In this way, it becomes easier for the model manager 30 and the like to understand the contents of the relevance evaluation value and the threshold, and the relationship between them. Hereinafter, the learning model for threshold determination may be simply referred to as a "learning model". The learning model for threshold determination 121 may be constructed using machine learning such as a neural network or deep learning.

[0026] [Examples of relevance assessment values ​​and thresholds] FIG. 4 is a diagram showing an example of the relevance evaluation value and the threshold according to the present embodiment. FIG. 4 shows an example of the relevance evaluation value and the threshold expressed by the numbers 1 to 5. For example, when the threshold is set between 2 and 3, the meaning of each of the relevance evaluation values ​​1 to 5 can be defined according to the threshold. In FIG. 4, the relevance evaluation value "5" is "easy to explain structurally", the relevance evaluation value "4" is "relatively easy to explain structurally", the relevance evaluation value "3" is "experienced person can understand", the relevance evaluation value "2" is "expert person can understand", and the relevance evaluation value "1" is "difficult even for an expert to understand". As a result, when the relevance evaluation value is 3 to 5, the structural understanding of the user is determined to be "structural", and when the relevance evaluation value is 1 to 2, the structural understanding of the user is determined to be "non-structural".

[0027] [Relevance determination section] The relevance determination unit 13 has a question content receiving function 131, a relevance determination function 132, and a determination result sending function 133. In the process in which the user repeats a medical interview in the TSRS 20, the Bayesian network model 22 determines the next question each time an answer is received from the user. The question content receiving function 131 receives this next question information. The relevance determination function 132 determines the relevance between the question content and the cause of a failure based on the next question information received by the question content receiving function 131, the relevance information in the relevance information storage unit 11, and the threshold value in the threshold value determination unit 12. The determination result sending function 133 sends the result of the determination made by the relevance determination function 132 to the medical interview processing unit 21 of the TSRS 20.

[0028] In the TSRS 20, the medical interview processing unit 21 receives the determination result from the relevance determination unit 13 of the relevance determination system 10, and transmits the next question and an explanatory text regarding the relevance between the content of the question and the cause of the failure to the user terminal.

[0029] [Description table] 5 is a diagram showing an example of the structure of an explanation table 500 provided in the TSRS 20. The explanation table 500 includes explanation information 503 for each combination of failure cause group information 501 and question content information 502. The explanation information 503 registers, for each combination of a failure cause group and question content, an explanation when the relationship between the question content and the failure cause is structural and an explanation when the relationship is non-structural.

[0030] The medical interview processing unit 21 refers to the explanation table 500 and selects an explanation to be notified to the user based on the question content, the fault cause (fault cause group in this example), and the judgment result. For example, for a combination of question content 1 and fault cause group 1, "structural" explanation 1-1 and "non-structural" explanation 1-1 are prepared as explanations. If the relationship between question content 1 and fault cause group 1 is structural, the "structural" explanation 1-1 is selected, and if the relationship is non-structural, the "non-structural" explanation 1-1 is selected.

[0031] [Troubleshooting Overview] Fig. 6 is a sequence diagram showing an overview of troubleshooting using the TSRS 20 and the relevance determination system 10. This sequence diagram shows the exchange of information between the relevance determination system 10 shown in Fig. 1, the user terminals used by the users 41 to 43, the medical interview processing unit 21 of the TSRS 20, and the Bayesian network model 22 of the TSRS 20. The troubleshooting sequence shown in Fig. 6 shows one scene in which questions and answers are repeatedly asked.

[0032] First, in step S601, users 41 to 43 use their user terminals to log in to the TSRS 20, and also transmit information linked to the user information to the relevance determination system 10. After processing in step S601, the process proceeds to step S602. The information linked to the user information is information associated with information that identifies the user (such as name or employee number), such as information that indicates the user's proficiency and the user's level of technical understanding. Information that indicates proficiency is a career history such as the department to which the user belongs and the number of years of service (see FIG. 20 described later). The level of technical understanding is, for example, the results of a comprehension test (see FIG. 15 described later).

[0033] In step S602, in the relevance determination system 10 that has received the user information, a threshold for the relevance assessment value corresponding to the user information is set in the threshold determination unit 12. Thereafter, the sequence from step S603 onwards is carried out.

[0034] In step S603, the users 41 to 43 who have received a question from the medical interview processing unit 21 of the TSRS 20 operate the user terminal to answer the question in a selection format. The answers of the users 41 to 43 to the question are then received by the medical interview processing unit 21 of the TSRS 20. After the process of step S603, the process proceeds to step S604.

[0035] In step S604, the medical interview processing unit 21 of the TSRS 20 passes the answer received from the user terminal to the internal Bayesian network model 22.

[0036] In step S605, the Bayesian network model 22 determines the next question based on the answer result and passes the determined next question and information on the cause of the failure that the Bayesian network model 22 suspects at that time (candidate cause of the failure) to the interview processing unit 21.

[0037] In step S606, the medical interview processing unit 21 transmits the information on the cause of the failure and the next question to the relevance determination unit 13 of the relevance determination system 10. In the relevance determination system 10, the question content receiving function 131 receives the information on the cause of the failure and the next question. After the processing of step S606, the process proceeds to step S607.

[0038] In step S607, in the relevance determination system 10, the relevance determination function 132 of the relevance determination unit 13 determines whether the relevance between the next question and the cause of the failure is structural / non-structural for the user, based on the information on the cause of the failure and the next question received from the TSRS 20. Then, the determination result transmission function 133 transmits the determination result to the TSRS 20. After the processing of step S607, the process proceeds to step S608.

[0039] In step S608, the medical interview processing unit 21 of the TSRS 20 displays information (e.g., an explanatory text) based on the determination result of the relevance determination system 10 and information on the next question on the user terminals of the users 41 to 43. The information based on the determination result is, for example, an explanatory text regarding the relevance between the cause of the failure and the content of the question.

[0040] [Troubleshooting Process] Fig. 7 is a flowchart showing an example of a procedure for troubleshooting processing by the TSRS 20 and the relevance determination system 10. The flowchart in Fig. 7 shows an example of a case where an announcement is displayed on a user terminal to inform the user that the relevance between a question and a cause of a failure is difficult to understand.

[0041] First, after starting the TSRS 20 (step S701), a threshold value for the relevance evaluation value in the relevance determination unit 13 is set. In the example of Fig. 7, the threshold value is set to "3" (step S702). After that, troubleshooting (questioning) using the Bayesian network model 22 is started (step S703), and a suspected fault cause is determined (step S704). In the example of Fig. 7, a fault cause belonging to fault cause group 1 is suspected.

[0042] Thereafter, a question to be presented to the user is determined by the Bayesian network model 22 (step S705), and information on the failure cause group and the question is transmitted to the relevance determination unit 13 (step S706). In the example of Fig. 7, the question to be presented is determined to be question content 1.

[0043] The relevance determination unit 13 loads the relevance information table 300 from the relevance information storage unit 11. The relevance determination function 132 refers to the relevance evaluation value of the row of the question content corresponding to the question in the failure cause group column to which the failure cause belongs, based on the information on the failure cause and question received from the TSRS 20 (step S707). In the example of Fig. 7, the relevance evaluation value corresponding to the combination of failure cause group 1 and question content 1 determined in steps S704 and S705 is obtained.

[0044] The relevance determination function 132 compares the referenced relevance evaluation value with a threshold value (step S708), and if the relevance evaluation value is 3 or more (YES judgment in step S708), it determines that the relevance between the question and the fault cause group is structural (understandable) to the user. On the other hand, if the relevance evaluation value is less than 3 (NO judgment in step S708), the relevance determination function 132 determines that the relevance between the question and the fault cause is unstructured (incomprehensible and distrustful) to the user.

[0045] Next, after the determination in step S708 is completed, the determination result transmission function 133 of the relevance determination unit 13 transmits the determination result to the TSRS 20 (step S709).

[0046] Then, if it is determined that the relationship between the question and the fault cause (fault cause group) is structural for the user (YES judgment in step S708), the interview processing unit 21 of the TSRS 20 presents only the question information to the user terminal (step S710).

[0047] Furthermore, when it is determined that the relationship between the question and the fault cause (fault cause group) is unstructured for the user (NO judgment in step S708), the medical interview processing unit 21 notifies the user terminal that the relationship is difficult to understand by presenting the next question and sending a message (explanatory text) as shown in FIG. 9 (step S711).

[0048] [User interface screen] Here, the user interface screen displayed on the user terminal will be described with reference to FIGS.

[0049] (Example of a structural relationship) Fig. 8 is a diagram showing an example of a user interface screen. Fig. 8 shows an example of a medical interview screen 800 in the case where the relationship between the question and the fault cause is structural (YES in step S708). The medical interview screen 800 has display areas for a fault cause candidate 801, an explanation 802, a question 803, and a confirmation result selection 804.

[0050] The fault cause candidate 801 displays the fault causes currently suspected (estimated with a high degree of certainty) based on the results of inquiries during troubleshooting. In the explanation 802, an explanation selected by the medical interview processing unit 21 from the explanation table 500 based on the question content, the fault cause (fault cause group), and the determination result is displayed. A question that can be answered in a multiple choice format is displayed in question 803. In the example of FIG. Icons showing the numbers of answer options to the question are displayed in confirmation result selection 804. When the user selects an icon with a number matching the answer option to the question and performs an operation such as clicking or touching, the answer to the question is transmitted from the user terminal to TSRS 20. Note that confirmation result selection 804 is not limited to icons of options, and may also have a function that allows the user to freely input text such as "comments / memo."

[0051] In this example, the explanation 802 states, "*The relationship between this question and the cause of the failure is based on page 25 of the manual." When the relationship between the question and the cause of the failure is structural, as described in step S710 of Fig. 7, essentially only the question 803 is displayed, and explanation 802 is not necessary. However, when explanation 802 is displayed, a user (especially a beginner) can easily understand the relationship because he or she knows that the basis for the structural relationship is "page 25 of the manual."

[0052] In this manner, in this embodiment, it is possible to express a relevance that matches the user's sense regarding the relevance between the question content and the cause of a failure by using the relevance information (relevance evaluation value) in the relevance information table 300. Then, by determining the relevance between the question content and the cause of a failure in step S708 every time a question is presented, the system can detect any question that makes the user feel suspicious.

[0053] 8, a diagnostic result list is displayed below the display area of ​​the failure cause candidate 901. The diagnostic result list displays questions that the TSRS 20 has previously presented to the user and the user's answers. This allows the user to visually check the questions and the order in which they were asked.

[0054] (Example of a non-structural relationship) Fig. 9 is a diagram showing another example of the user interface screen. Fig. 9 shows an example of an inquiry screen 900 in the case where the association between the question and the fault cause is non-structural (NO in step S708). The inquiry screen 900 has display areas for a fault cause candidate 901, an explanation 902, a question 903, and a confirmation result selection 904.

[0055] In this example, the explanation 902 states, "*The relevance between this question and the cause of the failure is not stated in the manual, but the relevance has been recognized from past interview results." In this way, by switching the information presented to the user based on the judgment result (for example, an explanation different from that in FIG. 8), the user can recognize the possibility that the information may contain a relevance derived from a learning model that the user himself / herself does not know. This has the effect that, particularly when the user is a beginner, the user can continue the troubleshooting interview without worrying about the relevance.

[0056] As described above, the maintenance assistance system (maintenance assistance system 100) according to the first embodiment includes a Bayesian network model that estimates a cause of a failure based on the content of a question and the answer given to a user and determines a next question, an association information storage unit (association information storage unit 11) that stores the level of an index indicating a structural understanding level regarding the association between the next question and the cause of the failure as association information (e.g., a numerical value or a degree such as high, medium, low, etc.), a boundary determination unit (threshold determination unit 12) that determines a boundary (e.g., a threshold, a boundary of a degree, etc.) of the index indicating the structural understanding level at which the user's understanding of the association between the next question and the cause of the failure changes from structural to non-structural based on the level of the index indicating the structural understanding level obtained from the association information storage unit, and an association determination unit (association determination unit 13) that determines whether the association between the next question and the cause of the failure is structural or non-structural based on the level of the index indicating the structural understanding level regarding the association between the next question and the cause of the failure and the boundary determined by the boundary determination unit, and is configured to display information (question, explanation regarding the determination result for the association, etc.) to the user based on the determination result.

[0057] In addition, in the maintenance support system according to the first embodiment, the level of the index indicating the structural understanding of the relevance information is set based on the information source on which the knowledge of the user is based. For example, the information source is at least a manual and a design document related to the maintenance target. In this way, by setting the level of the index indicating the structural understanding of the relevance information based on technical information that the user should know for the job, such as the contents of technical documents created for the maintenance target, the level of the index indicating the structural understanding can be set in stages according to the technical level of the user.

[0058] Moreover, in the maintenance support system according to the first embodiment, the relevance information storage unit (relevance information storage unit 11) quantifies and stores the level of an index indicating the structural understanding of the relevance between the next question and the cause of the failure as relevance information (for example, the relevance information table 300), and the boundary determination unit (threshold determination unit 12) determines a numerical value (threshold) as a boundary of the index indicating the structural understanding. The relevance determination unit (relevance determination unit 13) compares the numerical value of the level of the index indicating the structural understanding of the relevance between the next question and the cause of the failure with the numerical value as the boundary of the index indicating the structural understanding determined by the boundary determination unit. In this way, by quantifying the level and boundary of the index indicating the structural understanding, the maintenance support system (relevance determination unit) can clearly determine whether the relevance between the next question and the cause of the failure is structural or non-structural.

[0059] The maintenance support system according to the present embodiment having the above-mentioned configuration can eliminate the distrust felt by users in troubleshooting using a Bayesian network prediction model, and can make users feel convinced of the content and order of troubleshooting questions. This leads to faster troubleshooting and reduced man-hours in operating the maintenance support system. Furthermore, an increase in users who feel convinced and use the system (improved user coverage) improves the ease of introducing the maintenance support system.

[0060] <Second embodiment> Next, a troubleshooting process performed by the TSRS 20 and the relevance determination system 10 according to the second embodiment will be described with reference to FIG. Fig. 10 is a flowchart showing an example of a procedure of a troubleshooting process by the TSRS 20 and the relevance determination system 10 according to the second embodiment. In Fig. 10, step S711 in Fig. 7 described in the first embodiment is changed to steps S1001 to S1003.

[0061] 10, after performing the processes of steps S701 to S709 in FIG. 7, the TSRS 20 judges whether the question and the cause of the failure were linked in the initial learning model (when the Bayesian network model 22 was constructed) (step S1001). If the question and the cause of the failure were linked in the initial learning model (YES judgment in step S1001), the medical interview processing unit 21 presents an explanation and question information based on the judgment result (step S1002).

[0062] On the other hand, if the question and the cause of the failure were not linked in the initial learning model (NO judgment in step S1001), the medical interview processing unit 21 announces that the relationship is a newly linked relationship created by the Bayesian network model 22 (prediction model) that the user is not aware of (not described in the manual, etc.), together with the question information (step S1003), as shown in Figure 12.

[0063] Here, the determination of the link between the question and the fault cause and the initial learning model will be described with reference to FIG. Fig. 11 is a diagram showing an example of the structure of an association table 1100. To determine whether a question and a fault cause are linked to an initial learning model, it is preferable to include fault cause information 1101 and question content information 1102 as shown in Fig. 11 and set the value of the combination of squares according to the link. In the association table 1100, if a link is set to the initial learning model, it is set to "1", and if no link is set, it is set to "0". Such an association table 1100 may be used to determine whether a question and a fault cause are linked to an initial learning model.

[0064] (New associations generated by the Bayesian network model) Fig. 12 is a diagram showing an example of a user interface screen. An interview screen 1200 shown in Fig. 12 has display areas for a failure cause candidate 1201, an explanation 1202, a question 1203, and a confirmation result selection 1204. The explanation 1202 states, "*The relationship between this question and the failure cause is newly generated within the prediction model."

[0065] In this manner, in this embodiment, it is possible to express a relevance that matches the user's sense regarding the relevance between the question content and the fault cause by using the relevance information (relevance evaluation value) in the relevance information table 300. Then, by determining the relevance between the question content and the fault cause (fault cause group) in step S708 every time a question is presented, the maintenance support system can detect all questions that the user finds suspicious.

[0066] Furthermore, in step S1003, by presenting that the association is newly generated in the Bayesian network model 22 (prediction model), the user can recognize that the association between the question and the cause of the failure is not included in the initial learning model. This allows a novice user to continue the medical interview without worrying about the association. Also, when an experienced user feels something is wrong with the association, the user can understand that the reason is that the corresponding association is not included in the initial learning model, and can continue the medical interview with a sense of satisfaction.

[0067] <Third embodiment> Next, a troubleshooting process performed by the TSRS 20 and the relevance determination system 10 according to the third embodiment will be described with reference to FIG. Fig. 13 is a flowchart showing an example of the procedure of a troubleshooting process by the TSRS 20 and the relevance determination system 10 according to the third embodiment. Fig. 13 shows an embodiment in which a relevance information function 1400 shown in Fig. 14 is used. In Fig. 13, steps S707 and S708 in Fig. 7 described in the first embodiment are changed to steps S1301 to S1303.

[0068] 14 is a diagram showing an example of the relationship information function 1400. The relationship information function 1400 is stored in the relationship information storage function 112 as many times as the number of failure cause groups. The relationship information function 1400 is a function having a coefficient a weighted by the failure cause group and the structural understanding of each question. For a variable Y (dependent variable) of one relationship information function 1400, there are variables X as many as the number of questions, and the coefficient a applied to each variable X is determined by the structural understanding.

[0069] 13, the processing of steps S701 to S706 in Fig. 7 is performed. However, in step S706, information on the cause of the failure and the question is transmitted to the relevance determination unit 13. After performing step S706, the relevance determination function 132 of the relevance determination unit 13 calls the relevance information function 1400 corresponding to the cause of the failure from the information on the cause of the failure and the next question received from the TSRS 20 (step S1301).

[0070] Next, the relevance determination function 132 assigns "1" only to the variable X corresponding to the question information, and assigns "0" to the other variables X (step S1302). The variable Y of the relevance information function 1400, which is the assignment result, has the same value as the coefficient a of the variable X corresponding to the question information.

[0071] Thereafter, the relevance determination function 132 compares the threshold value set in step S702 with the value of variable Y of the relevance information function 1400 (step S1303), and if the value of variable Y is 3 or more (YES judgment in step S1303), it determines that the relevance between the question and the cause of the failure is structural (understandable) to the user. On the other hand, if the value of variable Y is less than 3 (NO judgment in step S1303), the relevance determination function 132 determines that the relevance between the question and the cause of the failure is unstructured (incomprehensible and distrustful) to the user.

[0072] The determination result transmission function 133 of the relevance determination unit 13 transmits the determination result to the TSRS 20 (step S709).

[0073] If it is determined that the relationship between the question and the fault cause (fault cause group) is structural (YES in step S708), the interview processing unit 21 of the TSRS 20 presents only the question information to the user terminal (step S710).

[0074] Furthermore, if it is determined that the relationship between the question and the fault cause (fault cause group) is non-structural (NO in step S708), the medical interview processing unit 21 notifies the user terminal that the relationship is difficult to understand by presenting the next question and sending a message (explanatory text) as shown in FIG. 9 (step S711).

[0075] In this manner, in this embodiment, unlike the first embodiment, the relationship information is held as the relationship information function 1400, which makes it possible to express not only a one-to-one relationship between a question and a failure cause group, but also relationships between multiple questions and failure cause groups.

[0076] Moreover, by executing the process of calling the association information function 1400 corresponding to the failure cause in step S1301, it is possible to express an association that matches the user's sense. Then, by determining the association between the question content and the failure cause (failure cause group) in step S1303 every time a question is presented, the maintenance support system can detect all questions that make the user feel suspicious.

[0077] Next, two embodiments will be described as specific methods of utilizing the threshold determination learning model 121. In the above-mentioned first to third embodiments, the thresholds are determined collectively for all the fault cause groups according to the user's proficiency level information (for example, three-level indicators of "beginner", "intermediate", and "expert"). In the following two embodiments, the threshold determination learning model 121 is used to set the thresholds in detail for each user and each fault cause group.

[0078] In operation, a threshold is calculated in advance for each user using the threshold determination learning model 121, and is stored in the threshold storage function 122. Then, when the user uses the TSRS 20, a threshold specific to that user is determined based on the login information. As embodiments in which the threshold is determined using the threshold determination learning model 121, a fourth embodiment and a fifth embodiment will be described.

[0079] <Fourth embodiment> Fig. 15 is a diagram showing an example of test questions 1500 for the comprehension test in the fourth embodiment. The test questions 1500 may be created for each failure cause group. In the example of Fig. 15, there are 4 questions for each failure cause group 1 to 3, for a total of 12 questions. As an example, it is assumed that questions will be presented in a format that is easy to tally, such as multiple choice questions (e.g., five-choice format).

[0080] 16 is a diagram showing an example of answer results 1600 to test questions in the fourth embodiment. The answer results 1600 are, for example, a table having items such as "question number," "answer," and "correct / incorrect." The answer results 1600 reflect the user's answer and whether the answer is correct for each question number. The answer results are input information for the threshold determination learning model 121.

[0081] Fig. 17 is a diagram showing an example of a threshold set 1700 in the fourth embodiment. In the threshold set 1700, a user-specific threshold is set for each failure cause group. For example, in Fig. 17, the communication-related threshold is set to "4", the hardware-related threshold is set to "3", and the software-related threshold is set to "3" as the thresholds for user A. It can be inferred that user A has an understanding of communication-related matters, but is not good at software-related matters.

[0082] Fig. 18 is a diagram showing an example of a question group list in the fourth embodiment. The question group list includes information on each test question and the association between the question content and the relationship between the cause of failure. For example, Fig. 18 shows an example in which question numbers 1-1 to 1-4 are classified as communication-related, question numbers 2-1 to 2-4 as hardware-related, and question numbers 3-1 to 3-4 as software-related. Test questions 1500 are questions to check the level of understanding of each group of failure causes.

[0083] Fig. 19 is a diagram showing an example of the correct answer rate (%) and the threshold value in the fourth embodiment. For example, Fig. 18 shows an example in which the threshold value is set to "2" when the value is 0 or more and less than -50, the threshold value is set to "3" when the value is 50 or more and less than -80, and the threshold value is set to "4" when the value is 80 or more. For example, since a user with a high correct answer rate is considered to have a high level of understanding, the threshold value for structural understanding is set relatively high.

[0084] In this embodiment, the user is given the understanding test shown in Fig. 15, and thresholds are created for each user and each fault cause group using the answer results. Input and output information for the threshold determination learning model 121 (hereinafter sometimes simply referred to as "model") is defined as follows. Input information: User's answers to test questions (Figure 16) Information held by the model: Relationship between accuracy rate and threshold (setting value) (Fig. 19), linking each test question with the relationship between the question content and the cause of failure (Fig. 18) Output information: A set of thresholds specific to each user (Figure 17)

[0085] A procedure for determining a user-specific set of thresholds will now be outlined. (1) Have the user answer test questions that were prepared in advance to measure their understanding of the question content and the failure cause groups (Figure 15). (2) The learning model for threshold determination 121 holds a list of test questions (Fig. 18) that organizes the test questions to confirm the degree of understanding of the relationship between the question content and the failure cause group, and a method for determining the correct answer rate and the threshold (Fig. 19). The correct answer rate for each test question group is calculated from the user's test answer results, which are input information. The threshold for each user-specific failure cause group is calculated from the correct answer rate (Fig. 17). (3) The thresholds set when a user logs in can be finely tuned according to the results of user comprehension tests for each failure cause group (Figure 17).

[0086] <Fifth embodiment> In a fifth prevailing mode, a proficiency level is predicted from user information, and a threshold value is set according to the prediction result.

[0087] Fig. 20 is a diagram showing an example of a department and years of employment of a user in the fifth embodiment. A department table 2000 has items of "department" and "years of employment". Fig. 20 gives examples of departments to which a certain user belongs, such as "software development department", "hardware development department", and "information and communications department".

[0088] FIG. 21 is a diagram showing an example of the department, required years of employment, and correction threshold in the fifth embodiment. The correction threshold table 2100 has the items of "department," "required years of employment," and "correction threshold." The "required years of employment" is the number of years of employment required in the target department in order for the user to increase his / her proficiency in the target technology or knowledge. The "correction threshold" is a correction value for the threshold when the user has been in the target department for the required number of years or more. When the correction threshold is "1," for example, in the threshold set 1700 in FIG. 17, the threshold for the corresponding failure cause group is increased by "1." It is expected that the user's proficiency increases, and the user can structurally understand the relationship between the question and the failure cause (failure cause group).

[0089] Fig. 22 is a diagram showing an example of the relationship between years of service (years) and the threshold value in the fifth embodiment. In Fig. 22, as an example, the relationship between years of service (years) and the threshold value is expressed linearly. The longer the years of service, the higher the threshold value. In this figure, when the years of service is "6 years", the threshold value is set to "3".

[0090] 23 is a diagram showing an example of a correction threshold value considering the expertise of each department in the fifth embodiment. A correction threshold value table 2300 has items of "Employed department", "Required years of employment", and "Correction threshold value". In "Correction threshold value", a correction value for the threshold value is set for each combination of an employed department and a failure cause group. For example, if the employee has been in the software development department or hardware design department for four years, or in the information and communications department for three years, the base threshold value in FIG. 22 is adjusted by "+1."

[0091] Fig. 24 is a diagram showing an example of threshold sets before and after correction in the fifth embodiment. For example, a base proficiency level (threshold) is set based on information such as six years of tenure (top of Fig. 24). Then, the proficiency level (threshold) of each failure cause group is corrected based on information about the department to which the user belongs. Compared to the threshold set 2401 before correction, the thresholds for software-related failure cause groups in the corrected threshold set 2402 are increased by "1". For example, since user B has a long history of working at a software company, the software-related thresholds are increased (bottom of Fig. 24).

[0092] The input and output information for the threshold determination learning model 121 (hereinafter sometimes simply referred to as the "model") is defined as follows. Input information: User's work history, department information, etc. (Figure 20) Information held by the model: Relationship between work history and threshold (proficiency) (Figure 22), expertise of each department (indicator showing strengths / weaknesses for each failure cause group) (Figure 23) Output information: A set of thresholds specific to each user (Figure 24)

[0093] A procedure for determining a user-specific set of thresholds will now be outlined. (1) Prepare information about the user’s work history and the departments in which they have worked in the past (Figure 20). (2) The learning model for threshold determination 121 holds the relationship between work history and thresholds (proficiency) (Fig. 22) and information on the expertise of each department (Fig. 23). A base threshold is calculated from the work history, which is the input information (top of Fig. 24), and the threshold for each failure cause group is corrected based on the information of the department to which the user belongs. The corrected threshold for each failure cause group is set as a threshold set specific to the user (bottom of Fig. 24). (3) The thresholds set when a user logs in can be finely tuned according to the user’s history and by fault cause group (lower part of Figure 24).

[0094] The fourth and fifth embodiments may be combined. That is, the threshold determination learning model 121 determines the numerical value (threshold) of the index indicating the structural understanding level based on the user's technical understanding level (e.g., test results) and proficiency level. In this case, the user's skills in maintenance work can be more accurately reflected.

[0095] [Hardware configuration of TSRS and relevance determination system] Next, the hardware configuration of the computer provided in the TSRS 20 and the relevance determination system 10 will be described with reference to FIG.

[0096] Fig. 25 is a block diagram showing an example of the hardware configuration of a computer provided in the TSRS 20 and the relevance determination system 10. A calculator 2500 shown in Fig. 25 is hardware used as a so-called computer.

[0097] The computer 2500 includes a CPU (Central Processing Unit) 2501, a ROM (Read Only Memory) 2502, a RAM (Random Access Memory) 2503, a non-volatile storage 2506, and a network interface 2507, each of which is connected to a system bus.

[0098] The CPU 2501 is an example of a calculation device serving as a control unit. The ROM 2502 and the RAM 2503 are examples of memories.

[0099] The nonvolatile storage 2506 is a nonvolatile storage element having a larger capacity than a memory. Programs for realizing the functions of the TSRS 20 and the relevance determination system 10 in each of the above-mentioned embodiments are stored in the respective nonvolatile storages 2506. The nonvolatile storage 2506 is an example of a computer-readable non-transient recording medium. The programs may be stored in the ROM 2502. When at least the functions of the processing blocks in each system are realized by software, the CPU 2501 executes the programs stored in the ROM 2502 or the nonvolatile storage 2506 to realize the functions of the processing blocks in each system.

[0100] Furthermore, the Bayesian network model 22 of the TSRS 20 may be stored in the non-volatile storage 2506. Furthermore, the relevance information holding function 112 of the relevance determination system 10 is realized by using the RAM 2503 or the non-volatile storage 2506. Furthermore, the threshold determination learning model 121 of the relevance determination system 10 may be stored in the non-volatile storage 2506. Furthermore, the threshold holding function 122 is realized by using the RAM 2503 or the non-volatile storage 2506.

[0101] The network interface 2507 is configured with a communication device or the like that controls communication with other devices through a communication network. For example, the question content receiving function 131 and the determination result transmitting function 133 of the relevance determination system 10 are realized by the network interface 2507. In addition, the transmitting and receiving functions of the medical interview processing unit 21 are realized by the network interface 2507.

[0102] Furthermore, the present invention is not limited to the above-described embodiment, and various other applications and modifications are possible without departing from the gist of the present invention described in the claims. For example, the above-described embodiment has been described in detail and specifically in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to having all of the components described. It is also possible to replace a part of the configuration of one embodiment with a component of another embodiment. It is also possible to add a component of another embodiment to the configuration of one embodiment. It is also possible to add, replace, or delete other components to or from a part of the configuration of each embodiment.

[0103] In the example of FIG. 1, the relevance determination system 10 and the TSRS 20 are configured as separate systems, but the relevance determination system 10 may be incorporated into the TSRS 20 to form an integrated system.

[0104] In addition, the above-mentioned configurations, functions, processing units, etc. may be realized in part or in whole by hardware, for example, by designing them as integrated circuits. As the hardware, a broad processor device such as an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit) may be used.

[0105] In this specification, the processing steps describing chronological processing include not only processing performed chronologically according to the order described, but also processing that is not necessarily performed chronologically but is performed in parallel or individually (for example, processing by objects). Furthermore, the processing order of the processing steps describing chronological processing may be changed as long as it does not affect the processing result. [Explanation of symbols]

[0106] 10...Relevance determination system, 11...Relevance information storage unit, 12...Threshold determination unit, 13...Relevance determination unit, 20...TSRS, 21...Interview processing unit, 22...Bayesian network model, 100...Maintenance support system, 111...Relevance information registration function, 112...Relevance information storage function, 121...Learning model for threshold determination, 122...Threshold storage function, 131...Question content receiving function, 132...Relevance determination function, 133...Determination result sending function, 300...Relevance information table, 301...Failure cause group information, 302...Question content information, 303...Relevance evaluation value, 500...Explanation table, 501...Failure cause group information, 502...Question content information, 503...Explanation information, 800, 900...Interview screen, 801, 901...Failure cause candidate, 802,902...Explanation, 803,903...Question, 804,904...Select confirmation result

Claims

1. A Bayesian network model that infers the cause of a failure based on the questions and answers given to the user and determines the next question to be asked; and an association information storage unit that stores, as association information, a level of an index indicating a structural understanding degree regarding an association between a next question and a cause of a failure; a boundary determination unit that determines a boundary of the index indicating the structural understanding level at which the user's understanding of the relationship between the next question and the cause of the failure changes from structural to non-structural, with respect to the level of the index indicating the structural understanding level obtained from the relationship information storage unit; a relevance determination unit that determines whether the relevance between the next question and the cause of the failure is structural or non-structural based on a level of the index indicating the structural understanding degree regarding the relevance between the next question and the cause of the failure and the boundary determined by the boundary determination unit, Display information to the user based on the judgment result Maintenance support system.

2. The level of the structural understanding index of the relevance information is set based on the information source on which the user's knowledge is based. The maintenance support system according to claim 1 .

3. The information sources are at least manuals and design documents related to the maintenance target. The maintenance support system according to claim 2 .

4. the relevance information storage unit stores, as the relevance information, a level of an index indicating the structural understanding regarding the relevance between the next question and the cause of the failure, in a numerical form; The boundary determination unit determines a numerical value as the boundary of the index indicating the structural comprehension degree, The relevance determination unit compares a level value of the index indicating the structural understanding degree regarding the relevance between the next question and the cause of the failure with a boundary value of the index indicating the structural understanding degree determined by the boundary determination unit. The maintenance support system according to claim 1 .

5. The boundary determination unit has a learning model created using the relevance information and the user's technical understanding, and the learning model determines a numerical value of an index indicating the structural understanding based on the user's technical understanding. The maintenance support system according to claim 4.

6. The boundary determination unit has a learning model created using the relevance information and the proficiency of the user, and the learning model determines a numerical value of an index indicating the structural understanding level based on the proficiency of the user. The maintenance support system according to claim 4.

7. A maintenance support method in a maintenance support system, which uses a Bayesian network model to infer a cause of a failure based on a question to a user and the answer to the question, and to determine a next question, comprising the steps of: A process of determining a boundary of an index indicating a structural understanding level at which a user's understanding of the relationship between the next question and the cause of the failure changes from structural to non-structural with respect to a level of the index indicating a structural understanding level of the relationship between the next question and the cause of the failure; a process of determining whether the relationship between the next question and the cause of the failure is structural or non-structural based on a level of the index indicating the structural understanding degree regarding the relationship between the next question and the cause of the failure and a boundary of the index indicating the structural understanding degree; and displaying information to the user based on the result of the determination. Maintenance support method.

8. The maintenance support system uses a Bayesian network model to infer the cause of a failure based on the questions and answers given to the user, and to determine the next question to ask. A step of determining a boundary of an index indicating a structural understanding level at which a user's understanding of the relationship between the next question and the cause of the failure changes from structural to non-structural with respect to a level of the index indicating a structural understanding level of the relationship between the next question and the cause of the failure; a step of determining whether the relationship between the next question and the cause of the failure is structural or non-structural based on a level of the index indicating the structural understanding degree regarding the relationship between the next question and the cause of the failure and a boundary of the index indicating the structural understanding degree; A procedure for displaying information to the user based on the judgment result; A program for executing the above.

Images