JP2020080024A - Fault symptom detection system and program - Google Patents

Abstract

To provide a fault symptom detection system and program capable of improving efficiency in reviewing a design document by automatically analyzing the design document.SOLUTION: A collection of words N-gram (or word string) cut out from a word string of fault information (phenomenon and cause) is stored in failure list storage means 34 in association with a fault management number, check information related to past faults is stored in check list storage means 61 in association with the fault management number, each of the words N-gram cut out from a word string of the design document and the words N-gram (or words N-gram cut out from a word string) of fault information stored in the failure list storage means 34 are matched, and check information stored in the check list storage means 61 is acquired and output by using the fault management number corresponding to the word N-gram of the matched fault information.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a failure sign detection system and program configured by a computer that executes a process of detecting a sign of failure occurrence from the description contents of a design document, for example, an automatic analysis of a design document created in design development of a system. It can be used when performing.

  Generally, in system design and development, the designer or developer assigned as the person in charge of the development project performs the design and development based on the requirement definition according to the purpose of the user, and the design document created in the design and development is Development is carried out by a system in which reviewers such as line officers of the department in charge of development perform reviews. Therefore, the quality of system design and development depends on the skills of designers and reviewers. At this time, the skills of reviewers such as line officers are based on past development experience, fault information accumulated as knowledge, fault response experience, and the like.

  In recent years, along with the sophistication and complexity of information technology that accompanies rapid technological innovation, the annual development volume at the site of system design and development has increased. Not only normal reviewers such as seats, but also third-party reviews by line managers of the headquarters (departments other than the department in charge of development) with more extensive past development experience, trouble information, trouble handling experience, etc. It has been implemented.

  In addition, as a technique related to the present invention, a document inspection method is known that can effectively prevent omission of description of necessary items regarding a document created during software development (see Patent Document 1). In this document inspection method, morphological analysis is performed, but word N-gram (word N-gram) is not used.

  Further, there is known a knowledge extraction device that extracts knowledge information from an electronic document, specifically, a knowledge extraction device that extracts knowledge information related to stock trading from distributed news information. It is also possible to perform extraction (see Patent Document 2). This knowledge extraction device uses N-grams and morphological analysis, but does not use word N-grams.

  Note that the word N-gram may refer to a method of dividing a word string into continuous N words, but in the present invention, a word group generated by cutting out the word string while shifting the word string by one word at a time. (Consecutive N words) shall be called a word N gram.

Japanese Unexamined Patent Publication No. 2018-26029 Re-table 2013-179346 gazette (paragraph [0162])

  As mentioned above, in the conventional system design and development, the quality of the development system is ensured whether it is a normal review by the line officers or a third party with more experience and knowledge. Was a system that depended on the reviewer's skills.

  Therefore, it is difficult to standardize reviews because it depends on the experience and knowledge of the reviewers. Also, because of the convenience of the reviewer, it may be difficult to conduct an immediate review when the design is completed. Furthermore, it has become more difficult to secure the time for reviewers due to an increase in development volume and an increase in development projects. Therefore, efficient review of design documents has been desired.

  Not only design documents created in system design and development, but also various design documents such as design documents for products and products such as home appliances, vehicle design documents such as automobiles and aircraft, and architectural design documents. As for the above, if a similar development system is adopted, similar problems may occur.

  An object of the present invention is to provide a failure sign detection system and program capable of automatically analyzing a design document and improving the efficiency of reviewing the design document.

The present invention is a failure sign detection system configured by a computer that executes a process of detecting a sign of failure occurrence from the description of a design document,
Original failure list storage means for storing failure information including a phenomenon and a cause of past failures in association with a failure management number,
Using the part including at least the phenomenon and the cause of the failure information stored in the original failure list storage means, a word string of failure information for a word N-gram having N equal to 2 or more is generated, or Failure information word string generation means for executing a process of generating a set of failure information word N-grams that are cut out one by one from the failure information word string,
Failure list storage means for storing the failure information word string or the set of failure information word N-grams generated by the failure information word string generation means, together with the failure information, in association with the failure management number.
A check list storage unit that stores check information including check items related to past failures in association with a failure management number,
Design document text extraction means for executing a process of extracting a text from the data of the design document which is the target of the failure sign detection,
Using the design text extracted by the design text extraction means, a word string of the design text for the word N-gram with the same number N as the word N-gram of the fault information is generated, or Design document word string generation means for executing a process of generating a set of word N-grams of the design document that are cut out from the word string one by one.
Design document word string storage means for storing a design document word string or a set of design document word N-grams generated by the design document word string generation means;
A word N gram of the design document cut out by shifting one word from the word string of the design document stored in the design document word string storage means, or a word N gram of the design document acquired from a set of the word N gram of the design document. Word N gram unit detection processing means for executing the word N gram matching request processing, which includes each of them in the query for the word N gram matching search request, and sends the query result for the query;
A fault received by the query sent from the word N-gram unit detection processing means, and the word N-gram of the design document included in the query, which is cut out by shifting one word at a time from the word string of the failure information stored in the failure list storage means. The word N-gram of information or the word N-gram of fault information acquired from the set of the word N-grams of fault information determines whether there is a word N-gram of fault information that matches, and the word of fault information that matches When there is an N-gram, the word N-gram of the failure information or the failure management number stored in the failure list storage means in association with the word string obtained by cutting out the word N-gram or the failure management number is dealt with in addition. Search means for executing the word N-gram matching search processing for returning the failure information to the word N-gram unit detection processing means as a search result for the query;
Using the failure management number received by the word N-gram unit detection processing means as the search result by this search means, the check information stored in the check list storage means is acquired and the acquired check information is output, or this check information is output. In addition to the check information, a detection result output means for executing a process of outputting fault information corresponding to the fault management number is provided.

  Here, the "fault management number" may be composed of numbers such as 0, 1, 2, 3, etc., and is composed of characters such as alphabets, Greek characters, Japanese (including hiragana, katakana, and kanji). It may be configured as a combination of numbers and letters, and is used as identification information for identifying failure information in the present invention.

  In such a failure sign detection system of the present invention, whether the design information is automatically analyzed by using the failure information about the past failure and the description content of the design statement, and the check information including the check item associated with the past failure is output? , Or the fault information about past faults and the check information including the check items associated with the fault information are output, so that the design document review can be automated.

  In addition, by automating the review of design documents, it is possible to move away from the quality assurance system that depends on people. The reviewer who performs the review can reduce the burden on the reviewer because the time constraint is reduced or eliminated accordingly, and the need to consider the convenience of the reviewer is reduced. Since it disappears, it is possible to conduct an immediate review.

  Further, by realizing the automation of the review of the design document, a certain level of review quality can be secured and the review can be standardized as compared with the case of relying on the experience and knowledge of the reviewer.

  Then, since the check information including the check items associated with the past failure is output, it is possible to prevent the failure from occurring.

  In addition, since matching is performed using N-grams (N consecutive words) with N equal to 2 or more, matching with simple N-grams (N consecutive characters instead of words) or words with N being 1 Compared to the case of matching by N-gram (word unigram: word Uni-gram), it is possible to improve the accuracy of automatic analysis of the description content of the design document, that is, the accuracy of failure sign detection. To be done.

  <Configuration that executes frequent word skip processing>

In the failure sign detection system described above,
The failure information word string generation means is
It is configured to also execute a process of obtaining the cumulative number of appearances of each word extracted from the fault information stored in the fault list storage means, and storing the obtained cumulative number of appearances of each word in the word cumulative number of appearances storage means,
Of the words stored in the word cumulative appearance count storage means, a frequent word acquisition unit that executes a process of acquiring a frequent word in which the cumulative occurrence count exceeds a threshold value or is equal to or more than a threshold value,
The word N-gram unit detection processing means is
Prior to executing the word N-gram matching request process, the frequently-used word acquired by the frequently-used word acquisition means in the N words constituting the word N-gram of the design document that is the target of inclusion in the query and is sent out. It is desirable to determine whether or not is included, and when it is included, it is preferable to be configured to execute the frequent word skip processing without performing the word N-gram matching request processing.

  When the frequent word skip processing is executed as described above, it is possible to realize the processing in which the frequent words are regarded as noise, and it is possible to improve the accuracy of failure sign detection. Therefore, when viewed in terms of a word N-gram as a word group (a word string in which N consecutive words are arranged in that order), a word group that has a large number of appearances is automatically identified by the contents described in the design document. While it is considered to be an important factor in performing analysis, it is possible to realize the handling of frequently-used words when viewed with one word rather than with a word group, which is regarded as noise. Therefore, the accuracy improvement effect of the automatic analysis of the description contents of the design document, that is, the accuracy improvement effect of the failure sign detection, by adopting the matching by the word N gram (N consecutive words) in which N is 2 or more, It can be even more prominent.

  <Configuration for executing program ID/job ID reference determination processing>

Furthermore, in the failure sign detection system described above,
The failure list storage means is
In addition to the failure information word string or the set of failure information word N-grams, the failure information is associated with a failure impact degree indicating the magnitude of the failure and a program ID and/or job ID as failure information. It is configured to remember,
Failure information including a program ID and/or job ID stored in the failure list storage means in association with a predetermined large failure impact degree, and a corresponding failure management number are acquired to store the program ID/job ID. A program ID/job ID acquisition means for executing the processing to be stored in the means,
The word N-gram unit detection processing means is
In addition to word N-gram matching request processing,
Before sending the word N-gram of the design document included in the query, whether the program ID and/or job ID stored in the program ID/job ID storage means is included in the word N-gram of the design document It is determined whether or not it is included, and if it is included, a program ID/job ID reference determination process that adopts the fault management number corresponding to the program ID and/or the job ID as the fault management number of the output target is executed. Is configured,
The detection result output means is
Using the failure management number obtained by the word N-gram matching request processing and the failure management number obtained by the program ID/job ID reference determination processing, the check information stored in the checklist storage means is obtained. It must be configured to output the check information acquired by using each failure management number side by side, or to execute the processing of outputting the failure information corresponding to each failure management number in addition to these check information. Is desirable.

  Here, the term “program ID and/or job ID” means that the fault information management system that manages the source data of the fault information stored in the fault list storage means stores both the program ID and the job ID. When it is handled (when both the program ID and the job ID are used for managing the failure information), it means that either one of the program ID and the job ID may be used, or both may be used. Further, when the fault information management system originally handles only one of the program ID and the job ID, it is only necessary to handle one of them.

  When the program ID/job ID reference determination process is executed in this way, it is possible to use the program ID and/or the job ID to associate the description in the design document with the past failure. Therefore, the automatic analysis of the design document created in the design and development of the computer system can be performed with higher accuracy, and the accuracy of failure sign detection can be further improved.

  <Structure for calculating score using weight for each obstacle impact>

In the failure sign detection system described above,
The failure list storage means is
In addition to the word string of the failure information or the set of the word N-grams of the failure information, a failure impact degree indicating the magnitude of the failure is stored as the failure information in association with the failure management number,
The word N-gram unit detection processing means is
The word N-gram matching request process is configured to perform a process of including the weight for each failure impact degree in the query together with the word N-gram of the design document and transmitting the weight.
Search method is
As the word N-gram matching search process, when there is a word N-gram of fault information that matches the word N-gram of the design document included in the query, the word N-gram of the fault information or a word string obtained by cutting out this word N-gram is used. A score indicating the degree of matching is calculated by using the weights of the failure impact levels stored in association with each other in the failure list storage means, and the calculated score is added to the failure management number or this failure management number, and corresponding failure information. At the same time, it is configured to execute processing for returning to the word N-gram unit detection processing means as a search result for the query.
The detection result output means is
It is preferable that the score is used to output the check information, or to execute the process of outputting the check information and the failure information corresponding to the failure management number.

  Here, in the "detection result output means", "to output check information using a score or to output fault information corresponding to a fault management number in addition to this check information" means the size of the score. According to the above, the check information and the like are output side by side, the check information and the like are output together with the value of the score, the check information and the like corresponding to the fault management number having the higher score are output, and the score is determined by the user. This includes outputting check information or the like corresponding to the failure management number in the designated order, and combining these outputs.

  When the score is calculated by using the weight for each failure influence level in this way, when the word N-gram of the design document included in the query matches the word N-gram of the failure information, the failure information If the word N-gram corresponds to a large obstacle influence degree, the score becomes relatively large, and if it corresponds to a small obstacle influence degree, the score can be adjusted to become relatively small. Therefore, it becomes possible to further improve the accuracy of failure sign detection.

  <Configuration for performing matching using a group of related words>

Furthermore, in the failure sign detection system described above,
The word groups cut out while shifting one word at a time from the word string generated by using the failure information including the phenomena and causes about the past failures stored in the original failure list storage means Using a set as learning data, a related word group learning means for performing a learning process of a relationship between word groups,
A learning result storing means for storing a word vector as a distributed expression of a word group obtained as a learning result by the related word group learning means,
The word N-gram unit detection processing means is
As the word N-gram matching request process, using the word vector stored in the learning result storage means, a related word group associated with the word N-gram of the design document is obtained, and the obtained related word group is obtained as the word N of the design document. It is configured to execute the process of sending it by including it in the query together with Gram,
Search method is
As the word N-gram matching search processing, if there is a word N-gram of failure information that matches the word N-gram of the design document included in the query, or if there is a word N-gram of failure information that matches the related word group included in the query. , Or there is a word N-gram of failure information that matches the word N-gram of the design document included in the query, and a word N-gram of failure information that matches the related word group included in the query, the failure The failure management number stored in the failure list storage means in association with the word N-gram of information or the word string obtained by cutting out this word N-gram, or failure information corresponding to this failure management number in addition to this failure management number is used as the search result for the query. It is desirable to be configured to execute a process of returning to the word N-gram unit detection processing means.

  In the case where the matching is performed using the related word group in this way, in addition to matching the word N-gram of the design document and the word N-gram of the failure information, the related word group and the word N-gram of the failure information are also matched. Since matching can be performed, it is possible to further improve the accuracy of failure sign detection.

  <Structure for calculating score using weight for word N-gram of design document and weight for related word group>

Also, in the case where the matching is performed using the related word group,
The word N-gram unit detection processing means is
As the word N-gram matching request process, the weight for the word N-gram in the design document and the weight for the related word group smaller than this weight are included in the query together with the word N-gram in the design document and the related word group and transmitted. It is configured to execute processing,
Search method is
As the word N-gram matching search process, when there is a word N-gram of failure information that matches the word N-gram of the design document included in the query, a score indicating the degree of matching using the weight of the word N-gram of the design document. Is calculated, and if there is a word N-gram of the fault information that matches the related word group included in the query, a score indicating the matching degree is calculated using the weight of the related word group, and the design document included in the query is calculated. If there is a word N-gram of failure information that matches the word N-gram of the above-mentioned word and there is a word N-gram of failure information that matches the related word group included in the query, the word N-gram of the design document and the related word group A score indicating the degree of matching is calculated by using both weights of, and the calculated score, together with the failure management number or the failure information corresponding to the failure management number, is detected as a word N-gram unit detection process as a search result for the query. It is configured to execute the process returned to the means,
The detection result output means is
It is preferable that the score is used to output the check information, or to execute the process of outputting the check information and the failure information corresponding to the failure management number.

  When the score is calculated using the weight for the word N-gram of the design document and the weight for the related word group as described above, the word N-gram of the design document included in the query is the word of the failure information. It can be adjusted so that the score becomes relatively large when it matches the N-gram, and the score becomes relatively small when the related word group included in the query matches the word N-gram of the failure information. It is possible to further improve the accuracy of failure sign detection.

  <Using the word group appearance frequency and the word group rarity, the weight for the word N gram of the design document and the weight for the related word group are calculated and determined, and the sending of the query including the word N gram of the same design document is repeated. Configuration to avoid

Furthermore, in the case where the score is calculated using the weight for the word N-gram of the design document and the weight for the related word group,
Design document word string generation means,
In the set of word N-grams in the design document, a word group appearance frequency indicating the frequency of occurrence of the same word group is obtained, and a word group rarity is obtained using this word group appearance frequency. And the word group rarity are configured to execute a process of storing the word group appearance frequency storage means,
The word N-gram unit detection processing means is
As the word N-gram matching request process, a word group appears for the same word group as the word N-gram of the design document, using the weight for the word N-gram of the design document as a reference weight for relative weighting with the related word group. While calculating and deciding by multiplying the word group appearance frequency and word group rarity stored in the frequency storage means,
The weight of the related word group is used as a reference weight for relative weighting with the word N-gram of the design document, and the word group appearance stored in the word group appearance frequency storage means for the same word group as the related word group. Calculated by multiplying the frequency and word group rarity,
It is determined whether or not a query including the word N-gram of the same design document has already been transmitted, and if it has not been transmitted yet, the weight for the word N-gram of the design document that has been calculated and determined, and the related word group The weight is preferably included in the query together with the word N-gram of the design document and the related word group, and is transmitted. If the weight has already been transmitted, it is preferable that the processing not execute the query is executed.

  In this way, by using the word group appearance frequency and the word group rarity, the weight for the word N gram of the design document and the weight for the related word group are calculated and determined, and the query including the word N gram of the same design document is transmitted. When it is configured to avoid the repetition of the above, it is possible to avoid the repetition of the same matching processing by the search means, and it is possible to improve the calculation speed.

  <Configuration for executing word N-gram extended matching request processing>

Further, in the above-described configuration for executing the program ID/job ID reference determination process,
The program ID/job ID storage means
By associating the failure information including the program ID and/or the job ID acquired from the failure list storage means by the program ID/job ID acquisition means, the failure management number, and the word string of the failure information or the set of the word N-grams of the failure information. It is configured to remember,
The word N-gram unit detection processing means is
In addition to word N-gram matching request processing and program ID/job ID reference determination processing,
Fault information words cut out while shifting one word at a time from the fault information word string stored in the program ID/job ID storage means in association with the fault management number of the output target adopted in the program ID/job ID criterion determination process. Each of the N-grams or each of the word N-grams of the fault information acquired from the set of the word N-grams of the fault information is included in the query for the word N-gram extended matching search request and is sent out. Configured to execute a word N-gram extended matching request process that receives a search result for a request query,
Search method is
The query for the word N-gram extended matching search request sent from the word N-gram unit detection processing means is received, and the word N-gram of the failure information included in the received query is the word of the failure information stored in the failure list storage means. Whether there is a word N-gram of failure information that is cut out by shifting one word from the column, or whether there is a matching word N-gram of failure information among the word N-grams of failure information acquired from a set of word N-grams of failure information. If there is a matching word N-gram of failure information, the failure management number stored in the failure list storage means in association with the word N-gram of the failure information or the word string obtained by cutting out the word N-gram. Alternatively, in addition to the fault management number, the corresponding fault information is returned to the word N-gram unit detection processing means as a search result for the query for the word N-gram extended matching search request, and the word N-gram extended matching search process is executed. ,
The detection result output means is
Using the failure management number obtained by the word N-gram matching request processing, the failure management number obtained by the program ID/job ID reference determination processing, and the failure management number obtained by the word N-gram expanded matching search processing The check information stored in the check list storage means is acquired, and the check information acquired by using each failure management number is arranged and output, or in addition to these check information, each failure management number is corresponded to. It may be configured to execute a process of arranging and outputting the failure information.

  When the word N-gram extended matching request process is executed in this way, the word string of the fault information or the word N-gram of the fault information corresponding to the fault management number obtained by the program ID/job ID reference determination process. Is used to create a query for a word N-gram extended matching search request including a word N-gram of failure information, and the word N-gram of failure information included in the query and the failures stored in the failure list storage means. Match the word N-gram of information. Therefore, the word N-gram extended matching request process (matching between the word N-grams in the fault information) includes the program ID/job ID reference determination process and the word N-gram matching request process (the word N-gram in the design document and the fault information). It can be said that the processing is a combination of (matching with a word N gram), and therefore the word N gram unit detection processing means performs two different systems of processing and a processing of merging them so that three systems related to each other can be processed. The processing makes it possible to obtain the fault management number of the output target.

  <Invention of program>

  Further, the program of the present invention causes a computer to function as the failure sign detection system described above.

  The above program or a part thereof is, for example, a magneto-optical disk (MO), a compact disk (CD), a digital versatile disk (DVD), a flexible disk (FD), a magnetic tape, a read-only memory (ROM). , Electrically erasable and rewritable read-only memory (EEPROM), flash memory, random access memory (RAM), hard disk drive (HDD), solid state drive (SSD), flash disk, etc. It can be stored and distributed as a storage medium, and can be transmitted using a transmission medium such as a wired network such as LAN, MAN, WAN, Internet, intranet, and extranet, or a wireless communication network, or a combination thereof. It is also possible to carry it on a carrier wave. Further, the above program may be a part of another program, or may be recorded in a recording medium together with a separate program.

  As described above, according to the present invention, the design information is automatically analyzed using the failure information about the past failure and the description content of the design document, or the check information including the check item associated with the past failure is output, Or, it outputs fault information about past faults and check information including check items associated with it, so it is possible to automate the review of design documents. It is possible to reduce the burden, perform immediate reviews, ensure a certain level of review quality, standardize reviews, and prevent failures from occurring. In addition, matching is performed using the word N-gram with N of 2 or more, so failure sign detection There is an effect that the accuracy of can be improved.

The whole block diagram of the failure sign detection system of one embodiment of the present invention. The block diagram of the checklist memory|storage means of the said embodiment and its source data. The block diagram of the failure list storage means of the said embodiment and the failure list file which is the source data. The figure which shows the output example of the failure sign detection result of the said embodiment. The figure of the flow chart which shows the flow of registration processing (a part of preparation processing) of a checklist of the above-mentioned embodiment. The figure of the flow chart which shows the flow of registration processing (a part of preparation processing) of a failure list of the above-mentioned embodiment. The figure of the flow chart which shows the flow (the 1) of failure sign detection processing of the above-mentioned embodiment. The figure of the flow chart which shows the flow (the 2) of failure sign detection processing of the above-mentioned embodiment. The figure which shows the example of generation of the word N gram (word 3 gram) of the said embodiment. Explanatory drawing of the determination process of the fault management number of an output object based on the program ID and job ID of the said embodiment. The detailed explanatory view of the processing of the frequently-used word of the embodiment. The figure which shows the example of generation of the query (query from the word N gram unit detection processing means to the search engine) of the said embodiment. The figure which shows the specific example of the response result with respect to the query by the search engine of the said embodiment. The detailed explanatory view of the process of the score (coincidence) included in the response result to the query of the embodiment. Explanatory drawing of the learning process of the related word group of the said embodiment. Explanatory drawing of the word N gram extended matching search process and the word N gram extended matching search process which show the modification of this invention.

  An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 shows the overall configuration of a failure sign detection system 10 of this embodiment. FIG. 2 shows the structure of the checklist storage means 61 and the checklist which is the source data thereof, and FIG. 3 shows the structure of the failure list storage means 34 and the failure list file which is the source data thereof. .. Further, FIG. 4 shows an output example of the failure sign detection result by the detection result output means 56. Further, FIG. 5 shows a flow of checklist registration processing (a part of preparation processing), FIG. 6 shows a failure list registration processing (a part of preparation processing), and FIGS. The flow of the failure sign detection process is shown in each flowchart. Further, FIG. 9 is a process of generating a word N gram (word 3 gram), FIG. 10 is a process of determining a fault management number of an output target based on a program ID/job ID, and FIG. 11 is an explanatory diagram of a process of a frequent word. Is. FIG. 12 shows a generation example of a query (a query from the word N gram unit detection processing means 55 to the search engine 30), and FIG. 13 shows a specific example of a response result to the query by the search engine 30. There is. Further, FIG. 14 is an explanatory diagram of a process of a score (coincidence degree) included in a response result to a query (inquiry), and FIG. 15 is an explanatory diagram of a related word group learning process by the related word group learning means 43.

  <Overall configuration of failure sign detection system 10>

  In FIG. 1, a failure sign detection system 10 includes a failure sign detection server 20 that executes various kinds of processing related to failure sign detection and stores data necessary for various kinds of processing. The failure sign detection server 20 includes the network 1. Via shared file server 70, failure information management system 80, checklist management system 90, one or more administrator terminals 100 operated by the system administrator, one or more user terminals operated by the user 110 is connected.

  Here, the network 1 is a network in a limited area such as an in-house LAN or an in-house intranet in the present embodiment, but may be a wide area network such as the Internet and may be wired or wireless. It does not matter whether it is a wired or wireless mixed type.

  The failure sign detection server 20 is a server that functions as an application server and a database server, is composed of one or more computers, and has a search engine 30, a preparation processing unit 40, a failure sign detection processing unit 50, and a check. List storage means 61, learning result storage means 62, program ID/job ID storage means 63, frequent word storage means 64, design word string storage means 65, word group appearance frequency storage means 66, setting file The storage means 67 and the output target storage means 68 are included.

  The search engine 30 includes a failure information word string generation means 31, a frequent word transmission means 32, a search means 33, a failure list storage means 34, and a word cumulative appearance frequency storage means 35. The search engine 30 may be constructed by modifying existing open source software (hereinafter referred to as OSS), or may be constructed by itself with all the functions. Examples of usable OSS include Apache Solr, which is an existing search engine having a scoring function.

  The preparation processing unit 40 executes a preparation process before performing the process using the design document that is the target of the failure sign detection, and includes a checklist registration unit 41, a failure list registration unit 42, and related word group learning. And means 43.

  The failure sign detection processing unit 50 executes a failure sign detection process using a design document that is a target of failure sign detection, and includes a design document text extraction unit 51, a program ID/job ID acquisition unit 52, and a frequent occurrence. The word acquisition unit 53, the design document word string generation unit 54, the word N-gram unit detection processing unit 55, and the detection result output unit 56 are included.

  Then, the means 31 to 33 included in the search engine 30, the means 41 to 43 included in the preparation processing means 40, and the means 51 to 56 included in the failure sign detection processing means 50 include the failure sign detection server 20. It is realized by a central processing unit (CPU) provided inside the constituting computer body, one or a plurality of programs defining the operation procedure of the CPU, and a working memory such as a main memory or a cache memory. It The details of these means 31 to 33, 41 to 43, 51 to 56 will be described later.

  The failure list storage means 34 and the word cumulative appearance frequency storage means 35, the check list storage means 61, the learning result storage means 62, and the setting file storage means 67 included in the search engine 30 are, for example, a hard disk drive (HDD). , A solid state drive (SSD), etc., but other recording media such as DVD, CD, MO, magnetic tape, etc. are adopted as long as there is no problem in storage capacity, access speed, etc. You may. Further, the data storage format of each storage means 34, 35, 61, 62, 67 is arbitrary, and may be, for example, a database, or a file format such as a flat file or other data files. In the case of a database, each storage means 34, 35, 61, 62, 67 may be a separate database, or may be different tables in one or several databases. When the existing tool, for example, Apache Solr is modified and used as the search engine 30, the failure list storage means 34 and the word cumulative appearance frequency storage means 35 become a file format library. .. The details of these storage means 34, 35, 61, 62, 67 will be described later.

  On the other hand, the program ID/job ID storage unit 63, the frequently-used word storage unit 64, the design word string storage unit 65, the word group appearance frequency storage unit 66, and the output target storage unit 68 are application programs (for example, Java (registered trademark)). The program is written in a language such as)) and is realized by a working memory such as a main memory or a cache memory used in a program. However, each of these storage means 63, 64, 65, 66, 68 may be realized by a non-volatile memory. The details of these storage means 63, 64, 65, 66, 68 will be described later.

  The shared file server 70 is a server that stores various files used in various processes related to failure sign detection, is composed of one or a plurality of computers, and has a processed checklist file storage means 71 and a failure list file storage means. 72, a design document file storage means 73, a design document text file storage means 74, and a failure sign detection result file storage means 75. Each of these storage means 71 to 75 is specifically a folder (directory) for storing each file, and is preferably realized by, for example, a hard disk drive (HDD) or a solid state drive (SSD).

  The fault information management system 80 stores past fault information generated in the system (not the present system, but a large number of systems developed in the past, but the present system may be included therein). This is a management system, which is composed of one or a plurality of computers and is provided with an original failure list storage means 81. The original fault list storage means 81 is preferably realized by, for example, a hard disk drive (HDD), a solid state drive (SSD), or the like, but if the storage capacity, access speed, etc. do not cause a problem, for example, a DVD. , CD, MO, magnetic tape and other recording media may be adopted. Further, the data storage format of the original failure list storage means 81 is arbitrary, and may be, for example, a database or a file format such as a flat file or other data files. The details of the original failure list storage means 81 will be described later with reference to FIG.

  The checklist management system 90 is, for example, a system for managing check information including check items in each work process such as outline design, detailed design, and development. The checklist management system 90 is configured by one or more computers and stores the original checklist. Means 91 are provided. The original checklist storage means 91 is preferably realized by, for example, a hard disk drive (HDD), a solid state drive (SSD), or the like, but may be, for example, a DVD as long as there is no problem in storage capacity, access speed, or the like. , CD, MO, magnetic tape and other recording media may be adopted. The data storage format of the original checklist storage means 91 is arbitrary, and may be, for example, a database or a file format such as a flat file or another data file. The details of the original checklist storage unit 91 will be described later with reference to FIG.

  The administrator terminal 100 and the user terminal 110 are composed of computers, and are provided with input means such as a mouse and a keyboard, and display means such as a liquid crystal display. These terminals 100 and 110 may be mobile devices such as smartphones, tablet terminals, and personal digital assistants (PDAs) as long as they have the necessary functions.

  In addition, the administrator terminal 100 includes a failure-related checklist extraction unit 101. The failure-related check list extracting means 101 can be realized by, for example, a macro function of spreadsheet software. The details of the failure-related check list extraction means 101 will be described later.

  <Detailed configuration of failure sign detection server 20>

  <Structure of failure sign detection server 20/search engine 30/failure information word string generation means 31>

  The failure information word string generation means 31 sets [A]N to 2 or more by using the part including at least the phenomenon and the cause of the failure information stored in the original failure list storage means 81 of the failure information management system 80. The failure information word string for the word N-gram is generated, or [B] a set of failure information word N-grams that are cut out while shifting one word from the failure information word string are generated and generated [A ] In the case of [], the processing for storing the “failure information word string” or in the case of [B] the “set of failure information word N-grams” in the failure list storage means 34 is executed. In the present embodiment, as an example, the word is 3 grams (word Tri-gram). Further, the word 3 grams is, for example, only the phenomenon and the cause in the failure information, but if there is the phenomenon and the cause described for the customer, these are also included in the word 3 gram.

  In the present embodiment, as shown in FIG. 6, when the administrator operates the administrator terminal 100, the failure information stored in the original failure list storage means 81 of the failure information management system 80 becomes a failure list file (for example, CSV file), and is placed in the failure list file storage means 72 of the shared file server 70 via the administrator terminal 100, from which the failure list registration means 42 of the failure sign detection server 20 acquires the failure list file. Since the process of registering the failure list in the search engine 30 is performed, the failure information word string generation means 31 uses the phenomenon and the cause of the failure information in the acquired failure list file, or the failure of the search engine 30. Using the phenomenon and the cause of the failure information registered in the list storage means 34, the process of generating the [A] word string or the [B] word 3-gram set is executed.

  Specifically, the failure information word string generation means 31 performs [A] morpheme analysis on the phenomenon and the cause of the failure information (including the phenomenon and the cause described for the customer, if any). To extract only the noun, generate a word string of only the extracted noun, and store it in the failure list storage means 34, or [B] one word (one noun) from the word string of only the noun. The failure list storage means 34 generates a set of 3 grams of words of the failure information (in this embodiment, the phenomenon and the cause of the failure information) that are cut out while shifting them, and arranges the obtained plurality of 3 grams of words in the order of cutting out. Execute the process to be stored in.

  For example, as shown in FIG. 9, when there is text data “Illegal order for partial repayment designation on the conditional credit repayment order input screen... Credit repayment order input screen Partial repayment designation MAXAS Illegal order …” is generated, or from this word string, “conditional credit”, “additional credit repayment”, “credit repayment order”, “repayment” A permutation set (a set that retains the order) of 3 grams of the words "order input", "order input screen",... Is generated.

  Further, the failure information word string generation means 31 cumulatively appears each word (each noun) extracted from the failure information (in the present embodiment, the phenomenon and the cause of the failure information) stored in the failure list storage means 34. A process of calculating the number of times and storing the calculated cumulative number of appearances of each word in the word cumulative number of appearances storage unit 35 (see FIG. 11) is also executed. As shown in FIG. 11, the cumulative number of appearances is the appearance in the failure information (in this embodiment, the phenomenon and the cause) of all the past failures (all failure management numbers) stored in the failure list storage unit 34. It is the cumulative value of the number of times.

  <Configuration of Failure Prediction Detection Server 20/Search Engine 30/Frequent Word Transmission Means 32>

  In response to a request from the frequently-used word acquisition means 53, the frequently-used word transmission means 32 determines that the frequently-occurred cumulative number of occurrences of each word (each noun) stored in the word-accumulated-number-of-times storage means 35 (see FIG. 11) is frequently used. A process of passing a word (noun) that is equal to or greater than a threshold value (for example, 100 times) specified as a request from the word acquisition unit 53 or exceeds the threshold value to the frequent word acquisition unit 53 is executed. The frequent word transmission unit 32 may execute a process of passing all the words (nouns) and the cumulative number of appearances of each word to the frequent word acquisition unit 53.

  <Configuration of failure sign detection server 20/search engine 30/search means 33>

  As the word N-gram matching search processing, the search means 33 receives the query for the word N-gram matching search request sent from the word N-gram unit detection processing means 55, and regarding the word N-gram of the design document included in this query, [A] The word N of the failure information (phenomenon/cause) that is cut out while shifting one word from the word string of the failure information (in this embodiment, the phenomenon and the cause of the failure information) stored in the failure list storage means 34. It is determined whether or not there is a word N of matching failure information (phenomenon/cause) in the gram, or [B] word N of failure information (phenomenon/cause) stored in the failure list storage means 34. It is determined whether or not there is a matching word N-gram of failure information (phenomenon/cause) in the word N-grams of failure information (phenomenon/cause) acquired in order from the set of grams, and the matching failure information (phenomenon) If there is a word N-gram of (cause), the failure management stored in the failure list storage means 34 in association with the word N-gram of the failure information (phenomenon/cause) or the word string obtained by cutting out the word N-gram. The number or the fault management number and the corresponding fault information are returned to the word N-gram unit detection processing means 55 as a search result (response result) to the query for the word N-gram matching search request. In the present embodiment, the failure information (phenomenon/cause) is set to the word 3 grams (N=3) as an example, and accordingly, the word 3 grams (N =3).

  The difference between the processing of [A] and the processing of [B] is stored in the failure list storage means 34. In the case of [A], it is a word string of failure information (phenomenon/cause). In the case of [B], there is a difference caused by being a set of word N-grams of failure information (phenomenon/cause). That is, it depends on whether the process of cutting out the word N-gram from the word string is executed by the search unit 33 (in the case of [A]) or the failure information word string generation unit 31 (in the case of [B]). is there.

  Further, when the query for the word N-gram matching search request includes a related word group associated with the word N-gram of the design document, the search means 33 performs the word N-gram matching search process, if the related word group is associated with the related word group. It is also determined whether or not there is a word N-gram of failure information (phenomenon/cause) that matches the word group, and if there is a word N-gram of failure information (phenomenon/cause) that matches, the word N of the design document Similar to the case of matching with the gram, the process of returning the search result (response result) to the word N-gram unit detection processing means 55 is executed. In the present embodiment, the word N-gram of the design document is the word 3-gram, and accordingly, the related word group is also a word group in which three words (nouns) are connected in consideration of the arrangement order. For example, “domestic stock buying” and “stock buying domestic” are different related word groups because the order of the three words is different.

  Further, the search means 33 also executes a fuzzy search for the word N-gram of the design document as the word N-gram matching search process. This phrase fuzzy search allows the N words (three words in the present embodiment) forming the word N-gram of the design document to be separated from each other, and allows the word N-gram to be in a state before being cut out. This is a search for determining whether or not N words (three words) are included in the word string of the failure information (phenomenon/cause), and the number of words (for example, consecutive words) indicating the allowable range specified by the query. If N words (3 words) are included in 100 words to be executed, the search is judged to match. At this time, it is not necessary to consider the arrangement order of the N words (3 words), and in any order, the N words (3 words) have the allowable number of words (for example, 100 consecutive words). Words). If they match, the word N-gram matching search request is issued for the failure management number stored in the failure list storage means 34 in association with the matched word string, or the failure information corresponding to this failure management number. Processing for returning to the word N-gram unit detection processing means 55 as a search result (response result) for the query for.

  In addition, the search unit 33 performs a score (matching degree) calculation process using the weight specified in the query, and the calculated score together with the searched failure management number and the like is a query for a word N-gram matching search request. The process of returning to the word N gram unit detection processing means 55 as a search result (result result) is also executed. As the search engine 30, for example, when an existing tool, Apache Solr, is remodeled and used, the search target included in the query (in this embodiment, the word N-gram of the design document and the related word) is used. The shorter the length of the sentence (in this embodiment, the word string of the phenomenon and the cause that is the fault information) to be matched with the group, the higher the score, and the sentence to be matched (in this embodiment, , The higher the number of appearances of the search target (in the present embodiment, the word N-gram of the design document and the related word group) in the failure information, that is, the word string of the cause and the cause, the higher the score. The basic score is adjusted (changed) to be 1 (1 point) regardless of the length of the sentence to be matched and whether the number of appearances of the search target is high or low. Therefore, the magnitude of the score calculated by the search means 33 is adjusted so as to be influenced by the following weights.

  First, there is a weight for each failure impact degree. The failure impact degree is an index indicating the magnitude of the effect of a failure that has occurred in the system, and in the present embodiment, it is, for example, S, A, B, C, D in descending order. As shown in FIG. 12, for example, the word N gram of the design document included in the query (similarly in the case of a related word group or in the case of fuzzy phrase search) corresponds to the failure impact degree S (the present embodiment). Then, in order to increase the score 10 times when the word N-gram of the phenomenon and the cause is matched, the weight of the failure influence degree S is set to WS=10 times, and the failure information (phenomenon/cause) corresponding to the failure influence degree A ), the weight of the obstacle influence degree A is set to WA=5 times, and similarly, the weight of the obstacle influence degree B is set to WB=3 times, and the obstacle influence degree is increased. The weight of the degree C can be set to WC=2 times, and the weight of the failure impact degree D can be set to WD=1 times. Therefore, the score can be set to increase as the failure information (phenomenon/cause) corresponding to the greater failure impact degree is matched. The weights WS, WA, WB, WC, and WD for each failure influence degree are stored in the setting file and stored in the setting file storage unit 67 so that the administrator can change the setting by operating the administrator terminal 100. Has been done. In the example of FIG. 12, “impact” included in the query indicates a failure impact degree, and “^” indicates a weight.

  Next, the weight has a weight according to the type of search target included in the query. The weight WF for the word N-gram in the design document, the weight WR for the related word group, and the weight WV for the fuzzy phrase search. There is. When the phrase fuzzy search is not performed, the setting of the weight WV for the phrase fuzzy search may be omitted. These weights WF, WR, and WV are relative values for different scores depending on whether the word N-gram of the design document matches, the related word group matches, or the phrase fuzzy search matches. Weighting. As shown in FIG. 12, for example, if the word N-gram of the design document matches the word N-gram of the failure information (in this embodiment, the phenomenon and the cause), the score of the design document is doubled in order to double the score. The weight for the word N-gram is set to WF=2 times, and the weight for the related word group is set to be 1 when the related word group matches the word N-gram of the failure information (phenomenon/cause). It is possible to make settings such as setting WR=1. In the example of FIG. 12, one query includes the word N gram of one design document and four related word groups, but the weight WR for each of the four related word groups is the same. .. In the example of FIG. 12, “impact_cause_ngram” included in the query is a phenomenon/cause that is a search target (more accurately, an obstacle that is a target to be matched with the word N-gram of the design document that is the search target and its related word group). It indicates the phenomenon and cause of the information).

  Further, the weight WF for the word N-gram, the weight WR for the related word group, and the weight WV for the phrase fuzzy search in the above-described design document are in a state in which only the reference weights for the relative weighting are considered. Actually, the word N-gram unit detection processing means 55 causes each of the reference weights to have TF (Term Frequency) which is an index indicating the appearance frequency of the word group and IDF (Inverse which is an index indicating the rarity of the word group. Document Frequency, which is the inverse document frequency, and is multiplied by log (number of word groups/TF)) to calculate and determine the respective weights WF, WR, and WV. Therefore, this calculation determination processing will be described later in the description of the configuration of the word N-gram unit detection processing means 55. The reference weights for calculating these WF, WR, and WV are stored in the setting file storage means 67 and stored in the setting file so that the administrator can change the setting by operating the administrator terminal 100. In the example of FIG. 12, a query including WF and WR when TF and IDF are omitted and only reference weights are shown is shown.

  Furthermore, the weights WS, WA, WB, WC, and WD for each failure influence degree described above and the weights WF, WR, and WV for the design document word N-gram/related word group/phrase fuzzy search simultaneously overlap. And can be applied. In the example of FIG. 12 as well, AND is used in the query to simultaneously and redundantly apply it. Therefore, for example, when the word N-gram of the design document included in the query matches the word N-gram of the failure information (phenomenon/cause), the score is doubled by the weight WF of the word N-gram of the design document. Further, when the word N-gram of the matched failure information (phenomenon/cause) corresponds to the failure impact S, the weight WS of the failure impact S increases the score 10 times. WF×WS=2×10=20 times. Further, for example, when the related word group included in the query matches the word N-gram of the failure information (phenomenon/cause), the score is multiplied by 1 due to the weight WR of the related word group, and the matching If the word N-gram of the failure information (phenomenon/cause) that corresponds to the failure impact degree B corresponds to the failure impact degree B, the score is tripled by the weight WB of the failure impact degree B, so 1×3=3 times become.

  Further, the search unit 33 receives the query for the program ID/job ID search request sent by the program ID/job ID acquisition unit 52, and specifies the failure impact degree specified in the query (in this embodiment, an example). , The program ID and/or the job ID corresponding to the failure impact levels S and A are designated as the major failure impact levels. Alternatively, the job ID, the corresponding failure management number, the failure information (including the phenomenon and the cause), the word string of the phenomenon/cause, or the word N-gram (word 3 grams in the present embodiment) cut out from this word string. A process of returning the set to the program ID/job ID acquisition means 52 as a search result (response result) to the query for the program ID/job ID search request is executed.

  <Configuration of failure sign detection server 20/search engine 30/failure list storage means 34>

  As shown in FIG. 3, the failure list storage means 34 includes, for example, a failure management number, a title, a phenomenon, a cause, a word string of the phenomenon/cause, or a word N-gram cut out while shifting one word from this word string (this embodiment). In the form, a set of words (3 grams), a recurrence prevention measure, a failure impact degree, a program ID, a job ID, and the like are stored in association with each other. Although not shown, the failure list storage unit 34 also stores, as failure information, phenomena and causes that are described for customers. Further, there is a record without a program ID or job ID. Here, the record does not necessarily mean a record that constitutes a table of a database, but a record that means a recording unit of information managed in association with one failure management number.

  Among these, a word string of a phenomenon/cause (including a phenomenon/cause for a customer) or a set of word N-grams (words 3 grams in this embodiment) cut out by shifting one word from this word string is the failure information. The other information generated by the word string generation means 31 is the same as the failure information stored in the original failure list storage means 81 of the failure information management system 80, and therefore the failure list of the shared file server 70. This is the same as the contents of the failure list file stored in the file storage means 72.

  <Structure of failure sign detection server 20/search engine 30/word cumulative appearance frequency storage means 35>

  As shown in FIG. 11, the word cumulative appearance count storage unit 35 selects a failure from failure information (in the present embodiment, a phenomenon and a cause of failure information) stored in the failure list storage unit 34 (see FIG. 3). Each word (each noun) extracted by the information word string generating means 31 and the cumulative number of appearances of each word are stored in association with each other.

  <Configuration of failure sign detection server 20/preparation processing means 40/checklist registration means 41>

  The checklist registration means 41 monitors whether or not a processed checklist file is stored in the processed checklist file storage means 71 of the shared file server 70, and if the processed checklist file is stored, the processed checklist file is stored. A process of acquiring a file via the network 1 and storing the content information of the acquired processed checklist file in the checklist storage unit 61 (see FIG. 2) is executed.

  <Configuration of failure sign detection server 20/preparation processing means 40/failure list registration means 42>

  The failure list registration means 42 monitors whether or not the failure list file is stored in the failure list file storage means 72 of the shared file server 70. If the failure list file is stored, the failure list file is transferred via the network 1. And the content information of the acquired failure list file is stored in the failure list storage means 34 (see FIG. 3) of the search engine 34, and a word string of the failure information (phenomenon/cause) or one word from this word string is stored. In order to perform the generation processing of the word N gram (word 3 gram) that is cut out while shifting the contents, the processing of passing the acquired content information of the failure list file to the failure information word string generation means 31 is executed.

  <Configuration of failure sign detection server 20/preparation processing means 40/related word group learning means 43>

  The related word group learning means 43 performs a learning process of the relationship between the word groups (a set of N words in consideration of the arrangement order), and the obtained learning result data is stored in the learning result storage means 62. The processing for storing is executed. In the present embodiment, since the matching process of 3 gram words is performed, the learning process of the relationship between the word groups consisting of 3 words in consideration of the arrangement order is executed.

  Here, the data obtained as the learning result is a weight matrix formed by collecting word vectors as a distributed expression of word groups by the number of unique word groups (vocabulary numbers) that do not overlap. Note that the word group vector may be called a word group vector because it is not a word distributed expression but a word group distributed expression. Therefore, as long as a word vector (word group vector) as a distributed expression of a word group can be obtained, machine learning by any algorithm may be performed. For example, continuous learning that is word-to-beck (word2vec). Elmo with a bag of words (CBOW), skip-gram (Skip-gram), or long-short-term memory (LSTM), which is one of the Grove (GloVe) and recurrent neural networks (RNN). An algorithm such as (ELMo) can be adopted.

  The learning data used for learning is prepared as follows. First, the failure information about a plurality of (many) past failures stored in the original failure list storage unit 81 of the failure information management system 80 (in the present embodiment, the phenomenon and the cause of the failure information are described for the customer. If there are any phenomena and causes that have been caused, these are also included.) is used to generate a word string for each failure information (phenomenon/cause). Next, a word group cut out while shifting one word at a time from the generated word string is formed into a set of word groups arranged in the cutting order, and this is used as learning data. In addition, a word string of phenomena/causes (including for customers) stored in the failure list storage means 34 (see FIG. 3) of the search engine 30 or a set of word groups cut out by shifting one word from this word string. , May be used as learning data.

  Therefore, as shown in FIG. 15, not a word but a word group (a set of three words in consideration of the arrangement order) is made one element, and this word group is used as an input layer and an output layer of the neural network (NN). Machine learning is performed by using as an element corresponding to each node of. Therefore, the number of elements before and after the window size (for example, 10 elements before and after) is also the number of consecutive word groups (for example, 10 word groups before and after). Therefore, the arrangement order of the N (three in the present embodiment) words forming the word group is considered, and the arrangement order of the word group is also considered for the set of word groups. In this state, it is used as learning data.

  <Configuration of failure sign detection server 20/failure sign detection processing means 50/design document text extraction means 51>

  The design document text extracting means 51 executes a process of extracting a document from the data of the design document which is the target of the failure sign detection. Specifically, in the present embodiment, the design document text extracting unit 51 monitors whether or not the design document file is stored in the design document file storage unit 73 of the shared file server 70, and when the design document file is stored, Acquires the design document file via the network 1, performs a process of extracting text data from the content data of the acquired design document file, creates a design document text file storing the extracted text data, and creates The design document text file is transmitted to the shared file server 70 via the network 1 and placed in the design document text file storage means 74.

  <Configuration of failure sign detection server 20/failure sign detection processing means 50/program ID/job ID acquisition means 52>

  The program ID/job ID acquisition means 52 is stored in the failure list storage means 34 in association with a predetermined large failure impact degree (in this embodiment, the failure impact degrees S and A are taken as an example). The failure information including the existing program ID and/or job ID and the corresponding failure management number are acquired and stored in the program ID/job ID storage unit 63.

  Specifically, in the present embodiment, the program ID/job ID acquisition unit 52 sends a query for the program ID/job ID search request (including designation of the failure impact levels S and A) to the search engine 30. A process is executed to receive the failure information including the program ID and the job ID of the failure impact levels S and A and the failure management number as the search result (response result) sent out and returned from the search engine 30.

  Even if the failure information acquired by the program ID/job ID acquisition unit 52 together with the program ID or job ID includes information that can be output by the detection result output unit 56 (for example, a phenomenon or a cause). Well, or, here, only the failure management number is acquired together with the program ID and the job ID, and at the stage of the output processing by the detection result output means 56, the failure management number is used as an output target from the failure list storage means 34. Information (for example, a phenomenon or a cause) may be acquired.

  <Configuration of failure sign detection server 20/failure sign detection processing means 50/frequent word acquisition means 53>

  The frequent word acquisition unit 53 instructs the frequent word transmission unit 32 of the search engine 30 to calculate the cumulative number of appearances of the words stored in the word cumulative number of appearances storage unit 35 (see FIG. 11) as a threshold value (for example, 100 times). ) Is exceeded or a threshold value is exceeded or more than a threshold value (including the designation of the threshold value) is sent, and the frequent word sent from the search engine 30 is received and stored in the frequent word storage means 64. Is to execute.

  The frequent word acquisition unit 53 acquires the cumulative number of appearances of all the words stored in the word cumulative number of appearances storage unit 35 (see FIG. 11) from the frequent word transmission unit 32 of the search engine 30, and calculates the cumulative number of appearances. It is also possible to execute a process in which a word whose value exceeds a threshold value (for example, 100 times or more) or is equal to or more than the threshold value is determined as a frequent word and is stored in the frequent word storage unit 64.

  <Configuration of failure sign detection server 20/failure sign detection processing means 50/design document word string generation means 54>

  The design document word string generation means 54 extracts only nouns by performing [A] morphological analysis using the design document sentences (text data) extracted by the design document sentence extraction means 51, and extracts only the extracted nouns. Generate a word string of the design document for the word N-gram, or [B] generate a set of word N-grams of the design document that are cut out while shifting one word (one noun) from the word string of this design document, A process of storing the generated word string of the design document in the case of [A] or the set of word N-grams of the design document in the case of [B] in the design document word string storage means 65 is executed. Here, N of the word N-gram for the design document is the same number as N of the word N-gram for the failure information (in the present embodiment, the phenomenon and the cause), and in the present embodiment, as an example, N=3. is there.

  At this time, the [A] word string generation process or the [B] word N-gram set generation process for the above design document is performed by the phenomenon/cause (the phenomenon of the fault information This is the same as the [A] word string generation process or the [B] word N-gram set generation process in the case of (cause text data).

  Specifically, in the present embodiment, the design document word string generation unit 54 monitors whether the design document text file is stored in the design document text file storage unit 74 of the shared file server 70, and stores the design document text file. If it is, the design document text file is acquired via the network 1, and the word string of the design document in the case of [A] is generated using the content data (text data) of the acquired design document text file. , Or [B], a set of the word N-gram of the design document is generated and stored in the design document word string storage unit 65.

  Further, the design document word string generation unit 54 stores the design document word string storage unit 65 in the set of the design document word N-grams (or stores it in the design document word string storage unit 65). (Or in the set of word N-grams of the design document cut out by shifting one word from the word string of the design document to be stored), the word group appearance frequency TF indicating the frequency of appearance of the same word group and the rareness of the word group The word group rarity degree IDF indicating the degree is calculated, and a process of storing these TF and IDF in the word group appearance frequency storage means 66 in association with the word group is also executed.

  <Configuration of failure sign detection server 20/failure sign detection processing means 50/word N-gram unit detection processing means 55>

  The word N-gram unit detection processing means 55 executes program ID/job ID reference determination processing, word N-gram matching request processing, and frequent word skip processing.

  Specifically, the word N-gram unit detection processing means 55 sends the word N-gram of the design document to be processed included in the query for the word N-gram matching search request as the program ID/job ID reference determination processing. Before (execution of the word N-gram matching request processing), as shown in FIG. 10, the program ID/job acquired by the program ID/job ID acquisition means 52 in the word N-gram of the design document to be processed is acquired. It is determined whether or not the program ID and/or the job ID stored in the ID storage unit 63 is included. If the program ID and/or the job ID is included, the failure management number corresponding to the program ID and/or the job ID It is adopted as the fault management number of the output target and is stored in the output target storage unit 68.

  At this time, the word N-gram unit detection processing means 55 stores the failure information (for example, phenomenon/cause) other than the program ID or job ID in the program ID/job ID storage means 63 in association with the failure management number. In addition to the failure management number, the failure information is also stored in the output target storage means 68.

  Further, the word N-gram unit detection processing means 55 instructs the search means 33 of the search engine 30 to execute the word N-gram unit detection processing from the word string of the design document stored in the [A] design document word string storage means 65. Each of the word N-grams of the design document extracted by shifting one word at a time, or each of the word N-grams of the design document acquired from the set of the word N-grams of the design document stored in the [B] design document word string storage unit 65, A process of receiving the search result (response result) for this query from the search engine 30 is executed while including and sending out the query for the word N-gram matching search request. FIG. 13 shows a specific example of the response result to the query, but the output content from the search engine 30 can be specified by the query. The output of the score shown in the bottom line of the example of FIG. 13 is also specified by the query.

  The difference between the process [A] and the process [B] is stored in the design word string storage means 65 in the case of [A] is the word string of the design document, In the case of B], the difference arises from the fact that it is a set of word N-grams of the design document. That is, whether the process of cutting out the word N-gram from the word string is executed by the word N-gram unit detection processing means 55 (in the case of [A]) or the design document word string generation means 54 (in [B]). Case).

  Further, the word N-gram unit detection processing means 55 uses the word vector (word group vector) forming the weight matrix stored in the learning result storage means 62 as the word N-gram matching request processing, and selects one of the processing targets. One or a plurality of related word groups associated with the word N-gram of the design document are obtained, and the obtained one or a plurality of related word groups are obtained by the search means 33 of the search engine 30 as the words of the design document to be processed. A process for receiving the search result for this query from the search engine 30 is executed while including the N-gram in the query for the word N-gram matching search request and transmitting the query. The process of obtaining the related word group can be executed by using, for example, word2vec. The setting file storage unit 63 (see FIG. 12) stores the maximum number (for example, four) of the related word groups to be used, so if this maximum number is input to word2vec or the like, for example, Four related word groups having a high degree of similarity (relevance) with the word N-gram of the design document are obtained.

  Further, the word N-gram unit detection processing means 55 may include a phrase fuzzy search request for the word N-gram of the design document in the query as the word N-gram matching request processing. This phrase fuzzy search has been described in detail in the description of the word N-gram matching search process by the search means 33 of the search engine 30, and thus detailed description thereof will be omitted here.

  Then, the word N-gram unit detection processing means 55 performs the following weight designation processing when creating a query for a word N-gram matching search request as the word N-gram matching request processing.

  First, regarding the weights WS, WA, WB, WC, and WD stored in the setting file storage means 67 for each failure impact degree, and the reference weight for calculating the weight WF for the word N-gram of the design document and the related word group. The reference weight for calculating the weight WR and the reference weight for calculating the weight WV for the phrase fuzzy search are acquired. The weight WV may not be used when the phrase fuzzy search is not performed, and the weight WR may not be used when the matching by the related word group is not performed.

  Next, the TF and the IDF for the same word group as the word N-gram of the design document to be processed are acquired from the word group appearance frequency storage means 66. Next, the weight WF=reference weight×TF×IDF for the word N-gram of the design document to be processed is calculated and determined, and the weight WR=related weight for the related word group=reference weight×TF×IDF is calculated and determined, and the phrase fuzzy search is performed. The weight WV=reference weight×TF×IDF is calculated and determined. In the example of FIG. 12, the word 3 grams of the design document to be processed=“conditional order”, and the query has a reference weight of “^2” (indicating double) for this “conditional order”. However, when TF and IDF are taken into consideration, “2×TF×IDF” becomes the weight WF. Also, in the example of FIG. 12, the related word group 1=“stock purchase”, and the query is given a reference weight of “^1” (indicating 1 time) with respect to this “stock purchase”. However, considering TF and IDF, “1×TF×IDF” is the weight WR.

  Then, the word N-gram unit detection processing means 55 causes the word N-gram of the design document to be processed, one or more related word groups, weights WS, WA, WB, WC, WD for each failure impact degree, and processing. Using the weight WF for the word N-gram of the target design document, the weight WR for the related word group, and the weight WV for the phrase fuzzy search, as shown in FIG. 12, a query for a word N-gram matching search request is made. create.

  In addition, the word N-gram unit detection processing unit 55 determines whether or not the query for the word N-gram matching search request including the word N-gram of the same design document has already been transmitted for the word N-gram of the design document to be processed. If it is determined that the query has already been sent, processing for not sending the query is executed. This process is based on calculating and determining the weight using TF and IDF.

  Further, as shown in FIG. 14, the word N-gram unit detection processing means 55 receives the search result (response result) for the query for the word N-gram matching search request from the search engine 30, and stores the received information as an output target storage. When it is stored in the means 68, the following score (matching degree) processing is executed.

  In FIG. 14, the word N-gram unit detection processing unit 55 is linked from the search engine 30 to the failure management number, the score calculated by the search unit 33, and the failure management number as the search result (response result) to the query. If the same failure management number as the received failure management number is already stored in the output target storage means 68 when the failure information (phenomenon, cause, etc.) is received, the score received together with the failure management number is used. If the same failure management number as the received failure management number is not yet stored in the output target storage means 68, the score is added to the score stored in the output target storage means 68 in association with the failure management number. The score received together with the failure management number is stored in the output target storage means 68 as it is.

  In the example of FIG. 14, first, as the search result (response result) by the first word 3 grams from the search engine 30, failure management number=AC24651, score=200 points, failure management number=AC23381, score=250. If the points are received, these failure management numbers are not yet stored in the output target storage means 68, so these scores are stored in the output target storage means 68 as they are.

  Next, as a search result (response result) by the second word 3 gram, failure management number=AC11987, score=100 points, failure management number=AC34765, score=300 points, failure management number=AC64328, score =700 points, since these failure management numbers are not yet stored in the output target storage means 68, these scores are stored in the output target storage means 68 as they are.

  Subsequently, when the failure management number=AC23381, score=500 points, and the failure management number=AC38577, score=400 points are received as the search result (response result) by the third word 3 gram, the former failure occurs. Since the management number=AC23381 is already stored in the output target storage unit 68, the score of the failure management number=AC23381 is set to 250 by adding the received score=500 points to the already stored 250 points. The number of points + 500 points = 750 points. On the other hand, since the latter failure management number=AC38577 is not yet stored in the output target storage means 68, the score=400 points is stored in the output target storage means 68 as it is.

  After that, if the failure management number=AC24651, score=600 points, and failure management number=AC64328, score=250 points are received as the search result (response result) by the fourth word 3 gram, these are output targets. Since the fault management number=AC24651 is already stored in the storage unit 68, the received score=600 points is added to the already stored 200 points to obtain 200 points+600 points=800 points. With respect to the fault management number=AC64328, the received score=250 points is added to the already stored 700 points to obtain 700 points+250 points=950 points. Finally, sort by score, and sort by score.

  Further, the word N-gram unit detection processing means 55 is a target to be included in the query for the word N-gram matching search request and transmitted before executing the word N-gram matching request processing as the frequent word skip processing. Frequent words acquired by the frequent word acquisition unit 53 and stored in the frequent word storage unit 64 among N (three in the present embodiment) words constituting the word N-gram of the design document to be processed. Is included, and if even one is included, a process of not performing the word N-gram matching request process is executed.

  Specifically, as shown in FIG. 11, it is assumed that the frequent words acquired by designating the threshold number of times=100 times are the words A, E, and H, and the word N-gram of the design document to be processed is the first word. It is assumed that the third word 3 gram=“L M Q”, the second word 3 gram=“M Q A”,...

  Since the first word 3 gram=“L M Q” does not include any of the words A, E, and H, the word N gram matching request process (for the word N gram matching search request is not skipped. Is sent to the search engine 30 to perform processing for executing the word N-gram matching search processing by the search means 33). Since the second word 3 gram=“M Q A” includes A of the words A, E, and H, the word N gram matching request process is not performed, that is, the search is skipped. The same applies to the third and subsequent word 3 grams, and if any one of the words A, E, and H is included in the three words forming the processing target word 3 gram, the word N The gram matching request process is not performed, that is, the search is skipped.

  <Configuration of failure sign detection server 20/failure sign detection processing means 50/detection result output means 56>

  The detection result output means 56 performs the program ID/job ID reference determination processing and the failure management number obtained by the word N-gram unit detection processing means 55 in the word N-gram matching request processing and stored in the output target storage means 68. The check information (including the check items) stored in the check list storage unit 61 (see FIG. 2) is acquired using the obtained failure management number stored in the output target storage unit 68 and acquired. The check information is output, or a process of outputting fault information (including a phenomenon, cause, etc.) corresponding to the fault management number in addition to the check information is executed.

  At this time, if the detection result output means 56 outputs the fault information (including the phenomenon, cause, etc.) corresponding to the fault management number in addition to the check information, the fault information (including the phenomenon, cause, etc.) When it is already stored in the output target storage means 68 in association with the failure management number, the failure information is output, and when it is not stored in the output target storage means 68 (the failure management number is stored, , (Causes, etc. are not stored), the failure management number is used to indicate the failure information (phenomenon, phenomenon) stored in the failure list storage means 34 (see FIG. 3) to the search means 33 of the search engine 30. It suffices to send an acquisition request (including the cause and the like) and acquire the failure information (including the phenomenon and the cause) to be output.

  Further, the detection result output means 56, the failure management number obtained by the word N-gram unit detection processing means 55 by the word N-gram matching request processing, and corresponding failure information (including a phenomenon, a cause, etc.) and check information ( (Including check items), the total score value (see FIG. 14) is set to a predetermined number of high-ranking past obstacles, such as within the top 10th and within the top 15th. Outputs the fault management number, etc. Therefore, when outputting these, the check information (including the check items and the like) is acquired from the check list storage means 61 (see FIG. 2) only for past failures such as within the top 10 and within the top 15 Just go.

  Further, as shown in FIG. 4, the detection result output means 56 causes the word N-gram unit detection processing means 55 to perform the word N-gram matching request processing, the failure management number, the score, and the corresponding failure information and check information. A matching result output section for outputting the program ID/job ID reference determination result output section and a failure management number obtained by the program ID/job ID reference determination processing, and corresponding failure information and check information. It is designed to be output separately (display, print, etc.). Since the former matching result output unit has a score, the matching result output unit is arranged and outputted in descending order of the score, and the latter program ID/job ID criterion determination result output unit has no score, and therefore the score is output as zero for convenience. ..

  In the example of FIG. 4, as the check information acquired from the check list storage unit 61 (see FIG. 2 ), description/check of the classification 1, classification 2, check items, and check items associated with the failure management number of the past failure The method and the like are output, and as the failure information acquired from the failure list storage means 34 of the search engine 30 (see FIG. 3), the failure impact degree, the title, and the phenomenon associated with the failure management number of the past failure (described for the customer (Including the phenomenon described above), the cause (including the cause described for the customer), the recurrence prevention measure, and the like, and the score obtained from the search engine 30 is output.

  Note that FIG. 4 is an example of output, and the output mode is not limited to this. For example, the upper matching result output unit and the lower program ID/job ID reference determination result output unit are The output position may be turned upside down. Also, the check information to be output may be, for example, only the check items, or the check information may be output without outputting the failure information (failure impact degree, title, phenomenon, cause, recurrence prevention measure, etc.). Good.

  <Configuration of Failure Prediction Detection Server 20/Storage Units 61 to 68>

  As shown in FIG. 2, the check list storage means 61 stores a failure management number for a past failure, a classification 1, a classification 2, a check item, a description/check method for the check item, etc. in association with each other.

  The learning result storage unit 62 stores, as the learning result data, a weight matrix (model) configured by a plurality (a large number) of word vectors (word group vectors) as a distributed expression of word groups.

  The program ID/job ID storage unit 63 has a large failure impact degree (in this embodiment, the failure impact degree S, A as an example) acquired by the program ID/job ID acquisition unit 52, the program ID and/or the job ID, and the failure. The management number is associated and stored. In addition to the program ID and the job ID, failure information (for example, a phenomenon or a cause) that may be an output target of the detection result output unit 56 may be stored.

  The frequently-used word storage unit 64 stores frequently-used words (words whose cumulative number of appearances is a threshold number of times (for example, 100 times) or more or exceeds the threshold number of times) acquired from the search engine 30.

  The design document word string storage unit 65 is a word string of the design document for the [A] word N gram (in this embodiment, N=3 as an example) created by the design document word string generation unit 54, or [ B] A set of word N-grams of the design document, which are cut out while shifting one word at a time from the word string of the design document, is stored.

  The word group appearance frequency storage means 66 stores the word group appearance frequency TF and the word group rarity degree IDF obtained by the design document word string generation means 54 in association with the word group.

  The setting file storage means 67 calculates weights WS, WA, WB, WC, and WD for each failure influence degree, reference weights for calculating the weight WF for the word N-gram of the design document, and weight WR for the related word group. And a setting file storing the reference weights for calculating the weight WV for the fuzzy phrase search. Further, the setting file may store selection information of use/non-use of the related word group, selection information of use/non-use of the fuzzy phrase search, maximum number of related word groups to be used, and the like.

  The output target storage unit 68 stores the failure management number and the score (total score) obtained by the word N-gram matching request processing by the word N-gram unit detection processing unit 55, or the corresponding failure information (phenomenon, cause, A recurrence prevention measure and the like are stored, and the failure management number obtained by the program ID/job ID reference determination processing by the word N-gram unit detection processing means 55, or the corresponding failure information (phenomenon, cause, recurrence) Preventive measures).

  <Configuration of Fault Information Management System 80/Primary Fault List Storage Means 81>

  As shown in FIG. 3, the original failure list storage means 81 stores, for example, a failure management number, a title, a phenomenon (including a phenomenon described for a customer), a cause (a cause described for a customer) of a past failure. (Including), a recurrence prevention measure, a failure impact degree, a program ID, a job ID, and the like are stored in association with each other. The original failure list storage means 81 stores the source data of the failure information to be stored in the failure list storage means 34 of the search engine 30 and also has a failure management number record to which no program ID or job ID is associated. Note that the record here is not limited to a record that constitutes a table of a database, and is used to mean a set of data (one set of data) associated with one failure management number. The same applies to the following.

  <Construction of Checklist Management System 90/Original Checklist Storage Means 91>

  As shown in FIG. 2, the original check list storage unit 91 associates, for example, a number (No.), a classification 1, a classification 2, a check item, a description/check method for the check item, a failure case/specific example, and the like. It is something to remember. This original checklist storage means 91 stores the source data of the check information to be stored in the checklist storage means 61 of the failure sign detection server 20, and the failure management number is included in the "failure case/specific example" column. There are records and records in which the failure management number is not included in the "failure case/specific example" column.

  For a record in which the failure management number (AC22731 in the example of FIG. 2) is included in the “failure case/specific example” column, the failure management is displayed in each column such as “check item” and “explanation/check method”. The check information associated with the past failure of the number (AC22731) is organized. In addition, each column such as “check item” and “explanation/check method” includes the development rule and information posted in the know-how collection.

  In the checklist files (each file such as the outline design checklist, the detailed design checklist, the development checklist, etc.) output from the original checklist storage means 91, there are records that do not include the failure management number. The processed checklist file processed by the administrator terminal 100 is composed of only records including the failure management number.

  <Configuration of administrator terminal 100/fault-related checklist extraction means 101>

  In the administrator terminal 100, the failure-related checklist extraction unit 101 stores the checklist file (file of spreadsheet software) acquired from the original checklist storage unit 91 of the checklist management system 90 in the failure case/specific example column in the past. Only a record containing a failure management number for a failure is extracted, and a processed checklist file (for example, CSV file) including records containing the failure management number and corresponding check information (check item, description, check method, etc.) File) is created and output is executed.

  <Process flow by the failure sign detection system 10>

  In the present embodiment as described above, the failure sign detection system 10 performs the failure sign detection processing on the design document relating to the design and development of the system (a system different from the failure sign detection system 10) as follows. .. First, before performing the process for the design document of the failure sign detection target, as a preliminary process, the check list registration process shown in FIG. 5 and the failure list registration process shown in FIG. 6 are performed.

  <Checklist registration process flow: See Figure 5>

  In FIG. 5, the system administrator operates the administrator terminal 100 to access the checklist management system 90 via the network 1, and various checklist files such as a general design checklist, a detailed design checklist, and a development checklist. And a checklist file transmitted from the checklist management system 90 via the network 1 is received (step S1). In the checklist management system 90, various checklist files are output from the original checklist storage unit 91 (see FIG. 2) in response to the acquisition request from the administrator terminal 100 and transmitted to the administrator terminal 100 via the network 1. Yes (step S2).

  Subsequently, in the administrator terminal 100, the failure-related checklist extraction unit 101 uses the various checklist files output from the original checklist storage unit 91 (see FIG. 2) to determine the failure management number for the past failure. The included records are extracted, and a processed checklist file (for example, CSV file) composed of records including the fault management number and the check information (check item, description, check method, etc.) corresponding to the fault management number is created and output. (Step S3).

  Then, the created processed checklist file is transmitted from the administrator terminal 100 to the shared file server 70 via the network 1 and placed in the processed checklist file storage means 71 (here, a dedicated directory) (step). S4). The shared file server 70 receives the processed checklist file transmitted from the administrator terminal 100 via the network 1 and stores it in the processed checklist file storage means 71 (step S5).

  On the other hand, in the failure sign detection server 20, the checklist registration means 41 monitors whether or not the processed checklist file is stored in the processed checklist file storage means 71 (step S6). The shared file server 70 receives the inquiry by this monitoring, and when the processed checklist file storage unit 71 stores the processed checklist file, the processed checklist file is damaged via the network 1. It transmits to the sign detection server 20 (step S7).

  In the failure sign detection server 20, the checklist registration means 41 receives the processed checklist file transmitted from the shared file server 70 via the network 1 (step S8), and the content data of the file is checkedlisted. It is registered in the storage means 61 (see FIG. 2) (step S9).

  <Flow of failure list registration processing: See FIG. 6>

  In FIG. 6, the system administrator operates the administrator terminal 100 to access the fault information management system 80 via the network 1 to make a request for obtaining a fault list file, and then from the fault information management system 80 via the network 1. The transmitted failure list file is received (step S21). In the failure information management system 80, in response to an acquisition request from the administrator terminal 100, a failure list file is output from the original failure list storage means 81 (see FIG. 3) and transmitted to the administrator terminal 100 via the network 1 ( Step S22).

  Subsequently, in the administrator terminal 100, the failure list file (eg CSV file) output from the original failure list storage means 81 (see FIG. 3) is transmitted from the administrator terminal 100 to the shared file server 70 via the network 1. Then, it is arranged in the failure list file storage means 72 (here, a dedicated directory) (step S23). The shared file server 70 receives the failure list file transmitted from the administrator terminal 100 via the network 1 and stores it in the failure list file storage means 72 (step S24).

  On the other hand, in the failure sign detection server 20, the failure list registration means 42 monitors whether or not the failure list file is stored in the failure list file storage means 72 (step S25). The shared file server 70 receives the inquiry by this monitoring, and when the failure list file is stored in the failure list file storage means 72, transmits the failure list file to the failure sign detection server 20 via the network 1. Yes (step S26).

  In the failure sign detection server 20, the failure list registration means 42 receives the failure list file transmitted from the shared file server 70 via the network 1, and the content data of the file is stored in the failure list storage means of the search engine 30. 34 (see FIG. 3) (step S27).

  In the search engine 30, the failure information word string generation means 31 causes the failure information stored in the failure list storage means 34 (see FIG. 3) to include a phenomenon (including a phenomenon written for a customer) and a cause. (Including the cause described for the customer) or the phenomenon (including the phenomenon described for the customer) of the failure information included in the failure list file received from the failure list registration means 42. And a cause (including a cause described for the customer) are used to generate a word string of fault information for [A] word N gram (word 3 gram in this embodiment), or [B] A set of failure information words N-grams (words 3 grams) that are cut out one word at a time from this failure information word string is generated, and the generated "failure information word string" or [B] in the case of [A]. In this case, the "set of word N-grams of failure information" is stored in the "phenomenon/cause word string" column of the failure list storage means 34 (see FIG. 3) (step S28). This processing has already been described in detail with reference to FIG. 9 in the description of the configuration of the failure information word string generating means 31, and thus detailed description thereof will be omitted here.

  Further, the failure information word string generation means 31 calculates and calculates the cumulative number of appearances of each word (each noun) extracted from the failure information (phenomenon and cause) stored in the failure list storage means 34 (see FIG. 3). The cumulative number of appearances of each word is stored in the word cumulative number of appearances storage unit 35 (see FIG. 11) (step S29).

  After that, as shown in FIG. 15, the related word group learning means 43 causes a phenomenon (including a phenomenon described for the customer) and a cause of the failure information stored in the failure list storage means 34 (see FIG. 3). A learning process of a relationship between word groups (a set of N (3 in this embodiment) words in consideration of the arrangement order) using all data (including causes described for customers) The learning result data thus obtained is stored in the learning result storage means 62 (step S30). This learning processing has already been described in detail in the description of the configuration of the related word group learning means 43, and therefore a detailed description thereof will be omitted here. With the above, the preparation process in advance is completed.

  <Flow of failure sign detection process for design document: see FIGS. 7 and 8>

  In FIG. 7, a user operates the user terminal 110 to create a design document file (various files such as Word, Excel, text, and Vizio) related to system design and development through the network 1 and the shared file server 70. To the design document file storage unit 73 (here, a dedicated directory) (step S41). The shared file server 70 receives the design document file transmitted from the user terminal 110 via the network 1 and stores it in the design document file storage means 73 (step S42).

  On the other hand, the failure sign detection server 20 monitors whether or not the design document file is stored in the design document file storage means 73 by the design document text extraction 51 (step S43). The shared file server 70 receives the inquiry by this monitoring, and when the design document file is stored in the design document file storage unit 73, transmits the design document file to the failure sign detection server 20 via the network 1. Yes (step S44).

  Then, the failure sign detection server 20 receives the design document file transmitted from the shared file server 70 via the network 1 by the design document text extraction 51, and extracts the text data from the content data of the design document file. The design document text file storing the extracted document data is created (step S45), and the created design document text file is transmitted to the shared file server 70 via the network 1 to store the design document file. It is placed in the means 74 (step S46). The shared file server 70 receives the design document text file transmitted from the failure sign detection server 20 via the network 1 and stores it in the design document text file storage means 74 (step S47).

  On the other hand, in the failure sign detection server 20, the program ID/job ID acquisition unit 52 monitors whether or not the design document text file is stored in the design document text file storage unit 74 (step S48). The shared file server 70 receives the inquiry by this monitoring, and when the design document text file is stored in the design document text file storage means 74, the design document text file is sent to the failure sign detection server via the network 1. 20 (step S49). The monitoring process of whether or not the design document text file is stored in the design document text file storage unit 74 is not related to the contents of the program ID/job ID acquisition process by the program ID/job ID acquisition unit 52. For the convenience of explanation, the program ID/job ID acquisition means 52 is supposed to determine the start timing of the acquisition processing of the program ID and job ID (processing of step S51 described later). Therefore, the acquisition of the design document text file may be performed before the process (the process of step S55 described below) by the design word string generation unit 54.

  Subsequently, in the failure sign detection server 20, the program ID/job ID acquisition means 52 receives the design document text file transmitted from the shared file server 70 (step S50), and then, toward the search engine 30, A program ID/job ID search request query (including designation of failure impact levels S and A) is sent out, and as a search result (response result) returned from the search engine 30, a program ID of failure impact levels S and A The failure information and failure management number including the job ID are received and stored in the program ID/job ID storage means 63 (step S51).

  The search engine 30 receives a program ID/job ID search request query (including designation of failure impact levels S and A) from the program ID/job ID acquisition means 52, and associates the query with the failure impact levels S and A. The program ID, job ID, other failure information (phenomenon, cause, etc.), and failure management number stored in the failure list storage means 34 (see FIG. 3) are returned to the program ID/job ID acquisition means 52 (step S52).

  Then, in the failure sign detection server 20, the frequent word acquisition unit 53 sends a frequent word acquisition request including designation of a threshold value (for example, 100 times) to the frequent word transmission unit 32 of the search engine 30 and performs a search. Frequent words transmitted (replied) from the engine 30 are received and stored in the frequent word storage means 64 (step S53). In the search engine 30, when the frequent word transmission unit 32 receives the acquisition request from the frequent word acquisition unit 53, the cumulative number of occurrences of the words stored in the word cumulative number of occurrences storage unit 35 (see FIG. 11) is a threshold value. Frequent words that exceed (for example, 100 times or more) or are equal to or more than a threshold value are acquired and transmitted to the frequent word acquisition unit 53 (step S54).

  The frequent word acquisition unit 53 acquires the cumulative number of appearances of all the words stored in the word cumulative number of appearances storage unit 35 (see FIG. 11) from the frequent word transmission unit 32 of the search engine 30. A word whose value exceeds a threshold value (for example, 100 times or more) or is equal to or more than the threshold value may be determined as a frequently-used word and stored in the frequently-used word storage unit 64.

  Thereafter, the design document word string generation means 54 monitors whether or not the design document text file is stored in the design document text file storage means 74 of the shared file server 70, and if stored, the design document is stored. A text file is acquired via the network 1, and using the content data of the acquired design document text file, [A] generates a word string of the design document, or [B] cuts out from this word string while shifting it word by word. A set of N words of the design document is generated and stored in the design word string storage means 65 (step S55). This processing has already been described in detail in the description of the configuration of the design document word string generation means 54, and thus detailed description will be omitted here.

  Further, the design-book word string generation unit 54 stores the design-document word string storage unit 65 in the set of the design-document word N-grams (or stores it in the design-book word string storage unit 65). (Or in the set of word N-grams of the design document cut out by shifting one word from the word string of the design document to be stored), the word group appearance frequency TF indicating the frequency of appearance of the same word group and the rareness of the word group The word group rarity degree IDF indicating the degree is obtained, and these TF and IDF are stored in the word group appearance frequency storage means 66 in association with the word group (step S56).

  After that, the process shifts to a loop process (a process of steps S57 to S63 described later) in which the unit of the word N-gram unit detection processing unit 55 is a word N-gram (here, a word 3 gram).

  In FIG. 8, first, the word N-gram unit detection processing means 55 executes the program ID/job ID reference determination processing. That is, as shown in FIG. 10, since the word N-gram of the design document to be processed (loop processing is performed, the word N-gram of the design document to be processed is replaced, one of the processing targets Means the word N-gram of the design document) of the design ID and/or the job ID (and the job ID) acquired by the program ID/job ID acquisition means 52 and stored in the program ID/job ID storage means 63. In the present embodiment, as an example, it is determined whether or not the program IDs and job IDs of the failure impact degrees S and A are included, and if they are included, the program ID and/or the job ID are dealt with. The failure management number is adopted as the failure management number of the output target, and the failure management number is stored in the output target storage means 68, or failure information associated with the failure management number in addition to the failure management number (phenomenon) , Cause, etc.) are stored in the output target storage means 68 (step S57). In this program ID/job ID reference determination process, unlike the case of the word N-gram matching request process, there is no score, so the score is set to zero, or NULL or equivalent to NULL for convenience.

  Next, the word N gram unit detection processing unit 55 executes the frequent word skip processing. That is, before executing the word N-gram matching request processing, the N pieces (the book) that compose the word N-gram of the design document to be included in the query for the word N-gram matching search request and sent out. In the embodiment, it is determined whether or not the frequent words included in the frequent word storage unit 64 acquired by the frequent word acquisition unit 53 are included in the three words (step S58), 1 If any of them is included, the word N-gram matching request process is not performed (search is skipped), and the process proceeds to the word N-gram of the next design target design document (the process proceeds to step S63 described later). ). This frequent word skip processing has already been described in detail with reference to FIG. 11 in the description of the configuration of the word N-gram unit detection processing means 55, and therefore detailed description will be omitted here.

  On the other hand, if it is determined in step S58 that the frequently-used word is not included, the process moves to the following word N-gram matching request process (step S59). In this word N-gram matching request process, if the same word N-gram is the second or subsequent process target among the word N-grams of the design document that is to be replaced by the loop process, the word N-gram and subsequent words Word N-gram matching request processing is not performed for N-grams. This is because weighting using TF and IDF is performed in the query to be sent.

  First, the word N-gram unit detection processing unit 55 stores the weights WS, WA, WB, WC, and WD stored in the setting file storage unit 67 for each failure influence degree, and the weight WF for the word N-gram of the design document. The reference weight for calculation, the reference weight for calculating the weight WR for the related word group, and the reference weight for calculating the weight WV for the fuzzy phrase search are acquired (step S59). The weight WV does not have to be used when the phrase fuzzy search is not performed, and the weight WR does not have to be used when the matching with the related word group is not performed.

  Next, the word N-gram unit detection processing means 55 acquires the TF and IDF for the same word group as the word N-gram of the design document to be processed from the word group appearance frequency storage means 66. Next, the weight WF=reference weight×TF×IDF for the word N-gram of the design document to be processed is calculated and determined, and the weight WR=related weight for the related word group=reference weight×TF×IDF is calculated and determined, and the phrase fuzzy search is performed. Is calculated and determined as the weight WV=reference weight×TF×IDF (step S59).

  Then, the word N-gram unit detection processing unit 55 uses the model stored in the learning result storage unit 62 (example of weighted line by word vector (word group vector)) to process the word N-gram of the design document to be processed (here Then, one or more related word groups (here, a word group in which three words are connected) for 3 words) are obtained (step S59).

  Then, the word N-gram unit detection processing means 55 causes the word N-gram of the design document to be processed, one or more related word groups, weights WS, WA, WB, WC, WD for each failure impact degree, and processing. Using the weight WF for the word N-gram of the target design document, the weight WR for the related word group, and the weight WV for the phrase fuzzy search, as shown in FIG. 12, a query for a word N-gram matching search request is made. The created query is sent to the search means 33 of the search engine 30 (step S59).

  In the search engine 30, the search means 33 receives the query from the word N gram unit detection processing means 55, and based on the query, the word N is used by using the data stored in the failure list storage means 34 (see FIG. 3). The gram matching search process is executed (step S60). That is, the matching process between the word N-gram and the related word group of the design document included in the query and the word N-gram of the fault information (phenomenon and cause) and the phrase fuzzy search are performed to calculate the score (matching degree). Run. The word N-gram matching search process has already been described in detail in the description of the configuration of the search means 33, and therefore a detailed description thereof will be omitted here. The ambiguous phrase search may be omitted, and the matching search with the related word group is not essential, and the use/non-use of these can be selected.

  Then, by the search means 33, as the search result (response result) to the query, the failure management number of the record that matches the search condition based on the query, the calculated score, or the failure information corresponding to the failure management number (phenomenon) And cause) are transmitted (replied) to the word N-gram unit detection processing means 55 (step S61). Note that the failure list storage means 34 (see FIG. 3) is not limited to a database, and may be a data storage format using a file. Therefore, the record here is not necessarily a record that constitutes a table of the database and one failure management number. It means a set of data related to a certain past failure, which is managed by being attached with.

  Subsequently, the word N-gram unit detection processing unit 55 receives the search result (fault management number, score, or, in addition to these, fault information such as a phenomenon or cause) returned from the search unit 33 of the search engine 30. The received search result is stored in the output target storage unit 68 (step S62). At this time, if the same failure management number and the score corresponding thereto are already stored in the output target storage means 68, the score is added. The processing of this score has already been described in detail in the description of the configuration of the word N-gram unit detection processing means 55 with reference to FIG. 14, and thus detailed description will be omitted here.

  Then, whether or not there is a word N-gram that is cut out from the word string of the design document for the [A] word N-gram (word 3 grams in this embodiment) stored in the design-document word string storage unit 65, Alternatively, it is determined whether or not the word N-gram acquired from the set of the word N-grams of the [B] design document stored in the design-word string storage unit 65 still exists (step S63), and if it still exists. In order to shift to the processing of the next word N-gram, the processing returns to step S57. On the other hand, if there is no more, the loop processing by the word N-gram unit detection processing means 55 is ended.

  Thereafter, the detection result output unit 56 outputs the failure management number and the like obtained by the word N-gram unit detection processing unit 55 in the above-described word N-gram matching request process in step S59 and stored in the output target storage unit 68 in step S62. , The total scores are sorted, and the past faults having the highest scores (for example, within the top 10 places, within the top 15 places, etc.) are selected from the fault management numbers and the like (step S64).

  Subsequently, the detection result output means 56 obtains the fault management number for the past fault having the higher score selected in step S64 and the program ID/job ID reference determination process in step S57 described above and stores the output target. The check information (check items, etc.) stored in the checklist storage unit 61 (see FIG. 2) is calculated by using the program IDs of the failure influence degrees S and A stored in the means 68 and the failure management number related to the job ID. Is acquired and the acquired check information is output, or failure information (including a phenomenon, cause, etc.) corresponding to the failure management number is output in addition to this check information (step S65). The output processing such as the display of the failure sign detection result has already been described in detail in the description of the configuration of the detection result output unit 56 with reference to FIG. 4, and thus the detailed description thereof will be omitted here. In the present embodiment, the detection result output means 56 creates and outputs a failure sign detection result file including output check information and the like, and arranges it in the failure sign detection result file storage means 75 of the shared file server 70. The shared file server 70 receives the failure sign detection result file transmitted from the failure sign detection server 20 via the network 1 and stores it in the failure sign detection result file storage means 75 (step S66).

  Then, the user operates the user terminal 110 to send an acquisition request for the failure sign detection result file stored in the failure sign detection result file storage means 75 to the shared file server 70 via the network 1, The failure sign detection result file transmitted from the shared file server 70 is received and output such as screen display and printing is performed (step S67). Upon receiving the acquisition request from the user terminal 110, the shared file server 70 transmits the failure sign detection result file stored in the failure sign detection result file storage means 75 to the user terminal 110 via the network 1. (Step S68).

  <Effect of this embodiment>

  According to this embodiment as described above, the following effects can be obtained. That is, the failure sign detection system 10 automatically analyzes the design document by using the failure information about the past failure and the description content of the design document, and outputs the check information including the check item associated with the past failure, or the past failure. Since the failure information about the check information and the check information including the check items associated therewith are output, it is possible to automate the review of the design document.

  In addition, by automating the review of design documents, it is possible to move away from the quality assurance system that relies on people, so ordinary reviewers such as line officers under the existing system and third party reviews can be performed. The reviewer who performs it can reduce the burden on the reviewer because the time constraint is reduced or eliminated accordingly. Further, since it is possible to reduce or eliminate the necessity of considering the reviewer's convenience, it is possible to perform an immediate review.

  Further, by realizing the automation of the review of the design document, it is possible to secure a certain level of review quality and standardize the review, as compared with the case of relying on the experience and knowledge of the reviewer.

  Then, since the check information including the check items associated with the past failure is output, it is possible to prevent the failure from occurring.

  Further, since the matching is performed by the word N-gram having N equal to or more than 2 (N consecutive words, which is three words in the present embodiment as an example), only N-grams (not consecutive words but continuous N-grams). It is possible to improve the accuracy of automatic analysis of the description content of the design document, that is, the accuracy of failure sign detection, as compared with the case of performing matching by each character) and matching by a word N-gram (word unigram) where N is 1. it can.

  Further, since the word N-gram unit detection processing means 55 is configured to execute the frequent word skip processing (see step S58 in FIG. 8), it is possible to realize the processing in which the frequent words are regarded as noise, which causes a trouble. The accuracy of sign detection can be improved. Therefore, when viewed in terms of a word N-gram as a word group (a word string in which N consecutive words are arranged in that order), a word group that has a large number of appearances is automatically identified by the contents described in the design document. While we consider it to be an important factor in our analysis (since the score when matching is added, we consider it to be an important factor.) Frequent words in case can be treated as noise. Therefore, the accuracy of the automatic analysis of the contents described in the design document is improved by adopting the matching by the word N-gram (N consecutive words, 3 words as an example in the present embodiment) in which N is 2 or more. The effect, that is, the effect of improving the accuracy of the failure sign detection can be made more remarkable.

  Further, the word N-gram unit detection processing means 55 executes the program ID/job ID reference determination processing (see step S57 in FIG. 8) in addition to the word N-gram matching request processing (see steps S59 to S62 in FIG. 8). Since it is configured, the program ID and/or the job ID can be used to associate the description content of the design document with the past failure. Therefore, the automatic analysis of the design document created in the design and development of the computer system can be performed more accurately, and the accuracy of failure sign detection can be further improved.

  Further, the word N-gram unit detection processing means 55, as the word N-gram matching request processing (see steps S59 to S62 in FIG. 8), assigns the weights for each of the failure impact degrees S, A, B, C, D to the words of the design document. It is included in the query together with the N-gram and its related word group, and is sent out. The search means 33 of the search engine 30 is configured to calculate the score by using the weight for each of the failure impact degrees. When the word N-gram of the design document and the related word group thereof match the word N-gram of the fault information (in this embodiment, the phenomenon and the cause), the word N-gram of the fault information has a large fault impact degree. If it corresponds, the score becomes relatively large, and if it corresponds to a small obstacle influence degree, it can be adjusted so that the score becomes relatively small. Therefore, the accuracy of failure sign detection can be further improved.

  Since the word N-gram unit detection processing unit 55 is configured to perform matching using the related word group, it is possible to match the word N-gram in the design document with the word N-gram in the failure information (phenomenon and cause). In addition, since it is possible to match the related word group with the word N-gram of the failure information, it is possible to further improve the accuracy of failure sign detection.

  Further, the word N-gram unit detection processing means 55 performs, as the word N-gram matching request processing, the reference weight for calculating the weight WF for the word N-gram of the design document, and the weight WR for the related word group smaller than the reference weight. Since the reference weight for calculation is included in the query together with the word N-gram of the design document and the related word group and transmitted, the word N-gram of the design document included in the query is included in the failure information (phenomenon and cause). ), the score becomes relatively large when it matches the word N-gram, and the related word group included in the query becomes relatively small when it matches the word N-gram in the failure information. Therefore, the accuracy of failure sign detection can be further improved.

  Furthermore, since the word N-gram unit detection processing means 55 can also perform a fuzzy search for a phrase, it is possible to perform various searches depending on the situation.

  Further, the word N-gram unit detection processing unit 55 uses the word group appearance frequency TF and the word group rarity degree IDF as the word N-gram matching request processing, and uses the weight WF for the word N-gram of the design document and the related word group. By calculating and determining the weight WR of the query, it is possible to avoid repeating the sending of the query including the word N-gram of the same design document, and thus it is possible to avoid repeating the same matching process by the search unit 33. The calculation speed can be improved.

  [Deformation form]

  It should be noted that the present invention is not limited to the above embodiment, and modifications and the like within the range in which the object of the present invention can be achieved are included in the present invention.

  For example, the failure sign detection system 10 of the above-described embodiment has a configuration in which the failure sign detection server 20 and other devices/systems/terminals are connected via the network 1. However, the present invention has such a network. The configuration is not limited, and it may be configured standalone.

  Further, the failure sign detection server 20 and the other devices/systems/terminals of the above-described embodiments each have one or more functions, but the arrangement of each function (each device/system/ Whether it belongs to the terminal) is not necessarily limited to the state of the failure sign detection system 10 of the above-described embodiment, and the mode of distribution or combination of each function is arbitrary.

  For example, the search engine 30 in the failure sign detection server 20 may be configured by an independent server, and the check list storage unit 61 in the failure sign detection server 20 may be installed in the independent server. Further, the failure information management system 80 and the checklist management system 90 may be one system. Further, the various data necessary for detecting the failure sign may be directly transmitted/received instead of transmitting/receiving the various files via the shared file server 70.

  Further, in the above-described embodiment, the design document targeted for failure sign detection is the design document for the system development design, but the design specification targeted for failure sign detection of the present invention is not limited to this. Instead of the above, various design documents such as design documents for products such as home electric appliances and merchandise, design documents for vehicles such as automobiles and aircraft, and design documents for buildings may be used. However, if it is not the design document of the system development design, the program ID/job ID reference determination processing is not performed. That is, the word N-gram unit detection processing means 55 of the above-described embodiment performs the program ID/job ID reference determination processing (see step S57 in FIG. 8) in addition to the word N-gram matching request processing (see steps S59 to S62 in FIG. 8). ) Was performed, but the latter process will not be performed.

  Further, the word N-gram unit detection processing means 55 of the above-described embodiment uses the program ID/job ID reference determination processing (see step S57 in FIG. 8), the frequent word skip processing (see step S58 in FIG. 8), and the word N-gram matching. Although the request processing (see steps S59 to S62 in FIG. 8) is configured to be executed, the word N-gram unit detection processing means of the present invention, as shown in FIG. It may be configured to execute the extended matching request process. Since the functions are increased as compared with the case of the above embodiment, the same reference numerals as those in the above embodiment will be used in FIG. 16 and the following description.

  In FIG. 16, the program ID/job ID storage means 63 has a large failure impact degree (for example, failure impact degree S, A) acquired from the failure list storage means 34 (see FIG. 3) by the program ID/job ID acquisition means 52. Program ID and/or job ID and other failure information (phenomenon, cause, recurrence prevention measure, etc.), failure management number, and word string of [A] failure information (for example, phenomenon/cause) or [B] failure A set of word N-grams of information (for example, phenomenon/cause) is associated and stored.

  In the example of FIG. 16, the word N-gram unit detection processing unit 55 uses the failure management number of the output target adopted in the program ID/job ID reference determination processing (see step S57 in FIG. 8) as the word N-gram extended matching request processing. The word N-gram of the fault information, which is cut out while shifting one word at a time from the word string of the [A] fault information (for example, phenomenon/cause) stored in the program ID/job ID storage unit 63 in association with, or [B ] Each of the word N-grams of the failure information acquired from the set of the word N-grams of the failure information (for example, the phenomenon/cause) is included in the query for the word N-gram extended matching search request, and is transmitted. A process for receiving a search result (response result) for a query for a gram extended matching search request is executed.

  Then, in the word N-gram extended matching request processing, the word N-gram unit detection processing means 55 responds to the query for the word N-gram extended matching search request from the search engine 30 as in the case of the word N-gram extended matching request processing. As a search result (response result), the fault management number for the matched past fault or this fault management number and corresponding fault information (phenomenon, cause, etc.) is received together with the score (degree of match), and the received information is output. The data is stored in the target storage unit 68, and at this time, the score processing as shown in FIG. 14 is executed. That is, when the received failure management number is already stored in the output target storage unit 68, a process of adding a score to the failure management number is performed.

  In the example of FIG. 16, the search means 33 of the search engine 30 receives the query for the word N-gram expanded matching search request sent from the word N-gram unit detection processing means 55, and the failure information included in the received query. A word N gram of (for example, a phenomenon/cause) is cut out from the word string of the [A] failure information (for example, a phenomenon/cause) stored in the failure list storage means 34 (see FIG. 3) while shifting one word at a time. Whether or not there is a matching word N-gram of failure information among the word N-grams of failure information or [B] the word N-gram of failure information acquired from the set of word N-grams of failure information (for example, a phenomenon/cause). If there is a matching word N-gram of failure information, the failure stored in the failure list storage means 34 in association with the word N-gram of the failure information or the word string cut out from the word N-gram. The management number or this failure management number and corresponding failure information (phenomenon, cause, recurrence prevention measure, etc.) together with the score is used as a search result (response result) for the query for the word N-gram extended matching search request. The processing to be returned to the unit detection processing means 55 is executed.

  Further, in the example of FIG. 16, the detection result output means 56 causes the failure management number obtained by the word N-gram matching request processing, the failure management number obtained by the program ID/job ID reference determination processing, and the word N-gram. The check information stored in the check list storage unit 61 is acquired using the failure management number obtained by the extended matching search process, and the check information acquired using each failure management number is output side by side. Alternatively, in addition to these pieces of check information, failure information (phenomenon, cause, etc.) corresponding to each failure management number is arranged and output.

  At this time, the detection result output means 56, similarly to the case of the word N-gram matching request processing, has the failure management number and the corresponding failure obtained by the word N-gram extended matching request processing and stored in the output target storage means 68. The information (phenomenon, cause, etc.) is sorted by the score (total value), and the total value of the scores (see FIG. 14) is a predetermined number such as within the top 10 or within the top 15 Outputs the failure management number, etc. for the higher-ranking past failures. Further, when outputting these, the check information (including the check items etc.) is acquired from the check list storage means 61 (see FIG. 2) only for past failures such as within the top 10 and within the top 15 Just go. The number of higher-ranking past failures to be output (how many higher ranks are to be output) may be different between the case of the word N-gram matching request process and the case of the word N-gram extended matching request process.

  Further, the detection result output means 56 outputs the failure management number and the like obtained by the word N-gram matching request processing and the failure management number and the like obtained by the word N-gram extended matching search processing separately to different areas. Alternatively, they may be mixed and output. When mixed and output, the score (total value) corresponding to the fault management number obtained by the word N-gram matching request process and the score corresponding to the fault management number obtained by the word N-gram extended matching search process Since the scores are ranked in a state where (total value) is mixed, the failure management obtained by the word N-gram extended matching search process is performed in order to adjust the relative size of these scores. It is preferable to multiply the score (total value) corresponding to the number by K times (K>1).

  When the word N-gram extended matching request process is executed in this way, the [A] failure corresponding to the failure management number obtained in the program ID/job ID reference determination processing (see step S57 in FIG. 8). Using a word string of information (for example, phenomenon/cause) or a set of word N-grams of [B] fault information (for example, phenomenon/cause), a word N-gram extension including a word N-gram of fault information (for example, phenomenon/cause) A query for a matching search request is created, and the word N-gram of the failure information (for example, phenomenon/cause) included in the query and the failure information (refer to FIG. 3) stored in the failure list storage means 34 of the search engine 30 (see FIG. 3). For example, the word N-gram of (phenomenon/cause) is matched. Therefore, the word N-gram extended matching request process (matching between the word N-grams in the fault information) includes the program ID/job ID reference determination process and the word N-gram matching request process (the word N-gram in the design document and the fault information). It can be said that the processing is a combination of (matching with a word N gram), and therefore the word N gram unit detection processing means 55 performs two different processings and a processing that fuses them, so that three processings related to each other are performed. By the processing of, the failure management number of the output target can be obtained.

  As described above, the failure sign detection system and program of the present invention are suitable for use in, for example, performing automatic analysis of a design document created in system design development.

10 Failure Prediction Detection System 31 Failure Information Word Sequence Generation Means 33 Searching Means 34 Failure List Storage Means 35 Word Cumulative Occurrence Count Storage Means 43 Related Word Group Learning Means 51 Design Document Text Extraction Means 52 Program ID/Job ID Acquisition Means 53 Frequent Words Acquisition unit 54 Design document word string generation unit 55 Word N gram unit detection processing unit 56 Detection result output unit 61 Checklist storage unit 62 Learning result storage unit 63 Program ID/Job ID storage unit 64 Frequent word storage unit 65 Design document word string Storage means 66 Word group appearance frequency storage means 81 Primary failure list storage means

Claims (9)

A failure sign detection system configured by a computer that executes a process of detecting a sign of failure occurrence from the contents of a design document,
Original failure list storage means for storing failure information including a phenomenon and a cause of past failures in association with a failure management number,
Using the part including at least the phenomenon and the cause of the failure information stored in the original failure list storage means, a word string of failure information for a word N-gram having N of 2 or more is generated, or Failure information word string generation means for executing a process of generating a set of failure information word N-grams, which are cut out one by one from the failure information word string,
Failure list storage means for storing the failure information word string or the set of failure information word N-grams generated by the failure information word string generation means in association with the failure information in association with the failure management number;
Check information including check items related to past failures, check list storage means for storing the check information in association with the failure management number,
Design document text extraction means for executing a process of extracting a text from the data of the design document which is the target of the failure sign detection,
Using the design document text extracted by the design document text extraction means, a word string of the design document for the word N-gram with N as many as N of the word N-gram of the fault information is generated, or this design document Design document word string generation means for executing a process of generating a set of design word N-grams, which are cut out from the word string while shifting one word at a time;
Design document word string storage means for storing a word string of the design document or a set of word N-grams of the design document generated by the design document word string generation means;
A word N gram of the design document cut out by shifting one word from the word string of the design document stored in the design document word string storage means, or a word N of the design document acquired from a set of the word N grams of the design document. Word N gram unit detection processing means for executing each word N gram matching request processing for sending each gram included in a query for a word N gram matching search request and receiving a search result for the query;
The query sent from the word N-gram unit detection processing means is received, and the word N-gram of the design document included in the query is word by word from the word string of the failure information stored in the failure list storage means. It is determined whether or not there is a word N-gram of the fault information that is cut out while shifting, or a word N-gram of the fault information that matches the word N-gram of the fault information acquired from the set of the word N-grams of the fault information. When there is a matching word N-gram of failure information, the failure management number or the failure management number stored in the failure list storage means in association with the word N-gram of the failure information or the word string obtained by cutting out the word N-gram. Search means for executing the word N-gram matching search processing for returning the corresponding failure information in addition to the failure management number to the word N-gram unit detection processing means as a search result for the query;
Using the fault management number received by the word N-gram unit detection processing means as the search result by the search means, the check information stored in the check list storage means is acquired, and the acquired check information is output. Or a detection result output means for executing a process of outputting the failure information corresponding to the failure management number in addition to the check information, and a failure sign detection system. The fault information word string generating means,
A configuration is also executed in which a cumulative appearance count of each word extracted from the fault information stored in the fault list storage means is obtained, and the obtained cumulative appearance count of each word is stored in the word cumulative appearance count storage means. ,
Of the words stored in the word cumulative appearance count storage means, a frequent word acquisition unit that executes a process of acquiring a frequent word in which the cumulative occurrence count exceeds a threshold value or is a threshold value or more,
The word N-gram unit detection processing means,
Before executing the word N-gram matching request process, the frequent word acquisition unit selects N words constituting the word N-gram of the design document to be included in the query and transmitted. It is configured to determine whether or not the acquired frequent word is included, and if it is included, a frequent word skip process that does not perform the word N-gram matching request process is executed. The failure sign detection system according to claim 1. The failure list storage means is
In addition to the word string of the fault information or the set of the word N-grams of the fault information, the fault management includes the fault impact degree indicating the magnitude of the impact of the fault, the program ID and/or the job ID as the fault information. It is configured to store in association with the number,
The failure information including the program ID and/or the job ID stored in the failure list storage unit in association with a predetermined large failure impact degree, and the corresponding failure management number are acquired to obtain the program ID A program ID/job ID acquisition means for executing the processing to be stored in the job ID storage means,
The word N-gram unit detection processing means,
In addition to the word N-gram matching request processing,
Before sending the word N gram of the design document in the query, the program ID and/or job ID stored in the program ID/job ID storage means is included in the word N gram of the design document. It is determined whether or not it is included, and if it is included, a program ID/job ID criterion that employs the fault management number corresponding to the program ID and/or the job ID as the fault management number of the output target. It is configured to execute judgment processing,
The detection result output means,
The failure management number obtained by the word N-gram matching request processing and the failure management number obtained by the program ID/job ID reference determination processing are used to store the check list storage means. Check information is acquired and the check information acquired using the respective failure management numbers is arranged and output, or the failure information corresponding to the respective failure management numbers is arranged and output in addition to these check information. The failure sign detection system according to claim 1, wherein the failure sign detection system is configured to execute processing. The failure list storage means is
In addition to the word string of the failure information or the set of the word N-grams of the failure information, the failure information indicating the magnitude of the effect of the failure is stored as the failure information in association with the failure management number,
The word N-gram unit detection processing means,
The word N-gram matching request process is configured to execute a process of including a weight for each failure influence degree in the query together with the word N-gram of the design document and transmitting the weight.
The search means is
As the word N-gram matching search process, when there is a word N-gram of the failure information that matches the word N-gram of the design document included in the query, the word N-gram of the failure information or this word N-gram is cut out. The score indicating the degree of matching is calculated using the weight of the failure influence degree stored in the failure list storage means in association with the word string, and the calculated score is the failure management number or this failure management number. In addition to the corresponding failure information, a process of returning to the word N-gram unit detection processing means as a search result for the query is executed.
The detection result output means,
It is configured to perform a process of outputting the check information by using the score or outputting the fault information corresponding to the fault management number in addition to the check information. The failure sign detection system according to any one of 1 to 3. A word group in which the word groups cut out while shifting one word from the word string generated by using the failure information including the phenomena and causes about the past failures stored in the original failure list storage means are arranged in the cut order. Using the set of as a learning data, a related word group learning means for executing a learning process of the relationship between the word groups,
A learning result storing means for storing a word vector as a distributed expression of a word group obtained as a learning result by the related word group learning means,
The word N-gram unit detection processing means,
As the word N-gram matching request processing, using the word vector stored in the learning result storage means, a related word group associated with the word N-gram of the design document is obtained, and the obtained related word group is It is configured to execute a process of including the word N-gram of the design document in the query and transmitting the query.
The search means is
As the word N-gram matching search process, if there is a word N-gram of the failure information that matches the word N-gram of the design document included in the query, the failure information that matches the related word group included in the query. Or there is a word N-gram of the fault information that matches the word N-gram of the design document included in the query, and the word N-gram that matches the related word group included in the query When there is a word N-gram of failure information, the failure management number or the failure management stored in the failure list storage means in association with the word N-gram of the failure information or the word string obtained by cutting out the word N-gram. It is configured to execute a process of returning the corresponding failure information in addition to a number to the word N-gram unit detection processing unit as a search result for the query. The failure sign detection system described. The word N-gram unit detection processing means,
As the word N-gram matching request process, the weight for the word N-gram of the design document and the weight for the related word group smaller than this weight are set to the query together with the word N-gram of the design document and the related word group. It is configured to execute the process of sending in
The search means is
As the word N-gram matching search process, when there is a word N-gram of the failure information that matches the word N-gram of the design document included in the query, the weight for the word N-gram of the design document is used. A score indicating the degree of matching is calculated, and if there is a word N-gram of the failure information that matches the related word group included in the query, a score indicating the degree of matching is calculated using the weight for the related word group. There is a word N-gram of the failure information that is calculated and matches the word N-gram of the design document included in the query, and a word N-gram of the failure information that matches the related word group included in the query is In some cases, a score indicating the degree of matching is calculated using both weights of the word N-gram of the design document and the related word group, and the calculated score is added to the fault management number or this fault management number. Together with the corresponding failure information, a process of returning to the word N-gram unit detection processing means as a search result for the query is executed.
The detection result output means,
It is configured to perform a process of outputting the check information by using the score or outputting the fault information corresponding to the fault management number in addition to the check information. 5. The failure sign detection system according to 5. The design word string generation means,
In the set of word N-grams in the design document, a word group appearance frequency indicating the frequency of occurrence of the same word group is obtained, and a word group rarity is obtained using this word group appearance frequency, and these word group appearances are obtained. The frequency and the word group rarity are configured to execute a process of storing the word group appearance frequency storage means,
The word N-gram unit detection processing means,
As the word N-gram matching request process, the weight for the word N-gram of the design document is set as a reference weight for relative weighting with the related word group, and the same word group as the word N-gram of the design document is used. While determining and calculating by multiplying the word group appearance frequency and the word group rarity stored in the word group appearance frequency storage means,
The weight for the related word group is stored in the word group appearance frequency storage means as a reference weight for relative weighting with the word N-gram of the design document, for the same word group as the related word group. Calculated by multiplying word group appearance frequency and word group rarity,
It is judged whether or not the query including the word N-gram of the same design document has already been transmitted, and if it has not been transmitted yet, the calculated weight of the word N-gram of the design document and the related word A weight for a group is sent together with the word N-gram of the design document and the related word group in the query, and if already sent, a process of not sending the query is executed. The failure sign detection system according to claim 6, wherein: The program ID/job ID storage means is
The failure information including the program ID and/or job ID acquired from the failure list storage means by the program ID/job ID acquisition means, the failure management number, and the word string of the failure information or the word N of the failure information. It is configured to associate and store a set of grams,
The word N-gram unit detection processing means,
In addition to the word N-gram matching request processing and the program ID/job ID reference determination processing,
The word strings of the fault information stored in the program ID/job ID storage means in association with the fault management number of the output target adopted in the program ID/job ID criterion determination processing are cut out while shifting one word at a time. The word N-gram of the fault information or each of the word N-grams of the fault information acquired from the set of the word N-grams of the fault information is included in the query for the word N-gram extended matching search request, and is transmitted. A configuration for executing a word N-gram extended matching request process that receives a search result for a query for an N-gram extended matching search request,
The search means is
The query for the word N-gram extended matching search request sent from the word N-gram unit detection processing means is received, and the word N-gram of the failure information included in the received query is stored in the failure list storage means. The fault information word N-gram cut out by shifting one word from the fault information word string, or the fault information word N matching in the fault information word N-gram obtained from the set of fault information word N-grams. It is determined whether or not there is a gram, and when there is a matching word N gram of the fault information, the word N gram of the fault information or a word string obtained by cutting out the word N gram is associated with the fault list storage means. The stored fault management number or the corresponding fault information in addition to this fault management number is returned to the word N gram unit detection processing means as a search result for the query for the word N gram extended matching search request. It is configured to execute gram extended matching search processing,
The detection result output means,
The failure management number obtained by the word N-gram matching request processing, the failure management number obtained by the program ID/job ID reference determination processing, and the failure obtained by the word N-gram extended matching search processing A management number is used to acquire the check information stored in the check list storage means, and the check information acquired using each failure management number is output side by side, or the check information is stored in the check information. In addition, the failure sign detection system according to claim 3, wherein the failure sign detection system is configured to execute a process of outputting the failure information corresponding to each failure management number side by side.   A program for causing a computer to function as the failure sign detection system according to claim 1.

Images