JP6692565B1 - FT diagram generation device, troubleshooting flow diagram generation device and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an FT diagram generation device and a program for directly generating an FT diagram from a trouble case. An FT diagram generation device A1 generates an FT diagram related to a work machine from a plurality of trouble cases 300b related to the work machine. The FT diagram generation device A1 is extracted by an event extraction unit 3 that extracts an event from a plurality of trouble cases 300b, and a first causal relationship extraction unit 4 that extracts a first causal relationship between events for each trouble case 300b. A connecting unit 5 that connects events corresponding to the first causal relationship in a top-down manner to generate an event group, and an integration that forms similar event groups among a plurality of event groups into a tree shape. And a section 8. When the trouble case 300b is input, the first causal relationship extracting unit 4 is a learned first neural network machine-learned using teacher data so as to extract the first causal relationship between events in each trouble case 300b. (First NN) 40 is included. [Selection diagram] Fig. 3

Description

  An FT diagram generation device that generates an FT diagram based on a trouble case of a work machine, a troubleshooting flow diagram generation device including the FT diagram, and a computer function as the FT diagram generation device and / or the troubleshooting flow diagram generation device. Regarding programs that let you.

  FTA (Fault Tree Analysis) is known as a method for analyzing accidents and failures that occur in work machines and the like. FTA is also called failure tree analysis. FTA uses an FT diagram that expresses events related to top events in a top-down tree structure. The top event corresponds to, for example, an accident or a failure that has occurred in the work machine. In the FT diagram, the lowest event among the events related to the top event is the basic event, and the event that is neither the top event nor the basic event is the intermediate event. In the FT diagram, when there are a plurality of events directly related to one event, that is, there may be a plurality of primary events, secondary events, ... N-order events in parallel.

  As the material for creating the FT diagram, the material created by the service person of the work machine is used. When an accident or failure occurs in the work machine, the service person solves it by taking some action on the work machine. The service person usually records such trouble cases in a report or the like, and the records (hereinafter referred to as "trouble cases") are accumulated.

  By the way, it is time-consuming and troublesome to create an FT diagram based on these plural trouble cases. Therefore, for example, programs that support FT diagram creation have been developed, such as the programs disclosed in Patent Documents 1 and 2.

  In the program disclosed in Patent Document 1, term groups are stored in advance for each technical field, and the FT diagram creator selects the corresponding phenomenon from the list of term groups one after another by top-down selecting the FT diagram. Can be created.

  However, in the programs disclosed in Patent Document 1 and Patent Document 2, data of a term group for creating an FT diagram must be created in advance by an operator from a trouble case. That is, the operator must manually extract the top event, the intermediate event, the basic event, and their causal relationship from the trouble case by using a spreadsheet software or the like, and make the data in advance. These operations are very troublesome and time-consuming, which is a problem.

  By the way, an inexperienced service person usually cannot easily determine what kind of action should be taken with respect to an accident or failure that occurs in the work machine. In addition, it takes time and labor for a skilled serviceman to provide guidance to such an inexperienced serviceman. Therefore, it is preferable to develop the above FT diagram to create a troubleshooting flow diagram that can refer to the action to be taken for the event that occurred in the work machine. However, creating such a troubleshooting flow chart is very troublesome and time-consuming, which is a problem.

JP, 2001-195433, A JP, 2013-073384, A

  Therefore, the problem to be solved by the present invention is to provide an FT diagram generation device and a troubleshooting flow diagram generation device and a program that can directly generate an FT diagram from a trouble case.

In order to solve the above problems, an FT diagram generation device according to the present invention is
An FT diagram generation device for generating an FT diagram relating to the working machine from a plurality of trouble cases relating to the working machine,
An event extraction unit that extracts an event from the plurality of trouble cases,
A learned first neural network machine-learned using first teacher data so as to extract a first causal relationship between the events for each of the trouble cases when the trouble case is input; A first causal relationship extraction unit that extracts one causal relationship;
A connecting unit that connects the extracted events corresponding to the first causal relationship in a top-down manner to generate an event group;
An integrating unit that forms the plurality of event groups in a tree shape by integrating the same events among the plurality of event groups ,
A morphological analysis unit that divides the plurality of trouble cases for each word,
Learning in which unsupervised learning is performed by using a neural network with the plurality of trouble cases divided for each word as learning data, and when a word is input, learning that outputs a distributed expression vector corresponding to the meaning of the input word Completed second neural network,
A word vectorization unit that converts the trouble case divided into words into a distributed representation vector by the learned second neural network;
The extracted event, based on the distributed expression vector for each word contained in the event, a sentence vectorization unit that converts the distributed expression vector for each event,
A similarity calculation unit that calculates the similarity between the events based on the cosine similarity between the events that are distributed representation vectorized by the sentence vectorization unit,
The integration unit may integrate the events for which a degree of similarity equal to or higher than a predetermined value is calculated .

The FT diagram generation device is preferably
The distributed expression vector corresponding to the meaning of the input word is the sum or average of the distributed expression vectors of the subwords extracted by the N-gram division of the input word.

The FT diagram generation device is, for example,
When the trouble case is input, the event extracting unit creates a learned third neural network machine-learned using the second teacher data so as to extract the event included in the trouble case. Have.

In order to solve the above-mentioned problems, a troubleshooting flow diagram generation device according to the present invention,
A flow chart generation device for troubleshooting a work machine,
An FT diagram generation device for generating an FT diagram relating to the working machine from a plurality of trouble cases relating to the working machine;
The FT diagram generator is
An event extraction unit that extracts an event from the plurality of trouble cases,
A learned first neural network machine-learned using first teacher data so as to extract a first causal relationship between the events for each of the trouble cases when the trouble case is input; A first causal relationship extraction unit that extracts one causal relationship;
A connecting unit that connects the extracted events corresponding to the first causal relationship in a top-down manner to generate an event group;
An integrating unit that forms the plurality of event groups in a tree shape by integrating the same events among the plurality of event groups,
The flow diagram generation device further includes
A treatment extraction unit that extracts a treatment performed on the event included in the plurality of trouble cases,
When the trouble case is input, the learned fourth machine-learned using the third teacher data so as to extract the second causal relationship between the action for each trouble case and the event corresponding to the action. A second causal relationship extracting unit that has a neural network and extracts the second causal relationship;
The connection unit further connects the treatment to the events in the event group in a top-down manner based on the second causal relationship,
It is characterized in that the integration unit further integrates similar treatments among the plurality of event groups.

The flow diagram generation device, for example,
When the trouble case is input, the treatment extraction unit has a learned fifth neural network machine-learned using the fourth teacher data so as to extract the treatment included in the trouble case.

In order to solve the above problems, the program according to the present invention is
The computer is caused to function as any one of the above generation devices.

  The FT diagram generation device according to the present invention can directly generate an FT diagram from a trouble case. Further, the flow diagram generation device including the FT diagram generation device according to the present invention can directly generate a troubleshooting flow diagram from a trouble case.

It is a block diagram showing a schematic structure of one embodiment of an FT figure generation device of the present invention. It is a figure which shows an example of a report. FIG. 2 is a functional block diagram of the FT diagram generation device of FIG. 1. A and B show an example of the generated event group, and C shows an example of the event group in which the events are integrated into a tree shape. It is a flowchart which shows the processing operation of the FT figure production | generation apparatus of FIG. 2 is an example of an FT diagram generated by the FT diagram generation device of FIG. 1. It is a block diagram which shows schematic structure of one Embodiment of the flowchart generation apparatus of this invention. FIG. 8 is a functional block diagram of the flow chart generation device in FIG. 7. It is a flowchart which shows the process operation of the flowchart generator of FIG. 8 is an example of a flow chart generated by the flow chart generation device of FIG. 7.

<First Embodiment>
<FT diagram generator>
Hereinafter, an embodiment of an FT diagram generation device A1 of the present invention will be described with reference to the drawings. Although the working machine in the present embodiment is a forklift, it is merely an example and the working machine is not limited to this.

  FIG. 1 is an overall view showing the outline of this embodiment. The computer C is configured to function as an FT diagram generation device that generates an FT diagram when a plurality of trouble cases 300b (see FIG. 2) described in the report 300 are input by the FT diagram generation program P1. ing.

  The computer C includes an input unit C1, a control unit C2, a memory C3, a storage unit C4, and an output unit C5. The storage unit C4 includes a hard disk or the like, and stores the OS and the FT diagram generation program P1. The input unit C1 includes a keyboard, a scanner, and the like, and inputs the trouble case 300b described in the plurality of reports 300 to the computer C. The control unit C2 operates the FT diagram generation program P1. The output unit C5 is composed of a monitor, a printer, and the like, and outputs the contents of the input trouble case 300b, the operation result of the FT diagram generation device, and the like.

  FIG. 2 shows an example of the report 300. In the report 300, the contents are described along with items such as date, subject, customer name, person in charge, model / model, and contents of failure. In the failure content column, the actions taken by the person in charge are recorded along with each event that occurred in the forklift. The trouble case 300b in the present specification refers to a sentence described in the trouble content column. In the trouble case 300b, each person may write each event in a bulleted list, or connect each case with a conjunction or a punctuation mark.

The respective events and actions in the trouble case 300b described in the report 300 of FIG. 2 are as follows.
・ Top event: The engine did not start.
-Intermediate event 1 (primary event): There was an abnormality in the intake system.
-Intermediate event 2 (secondary event): The intake hose was crushed.
・ Basic event: The air cleaner was clogged.
・ Remedy: Air cleaner was cleaned / filter was replaced.

  FIG. 3 is a block diagram of the FT diagram generation apparatus A1. The FT diagram generation device A1 includes a morphological analysis unit 1, a word vectorization unit 2, an event extraction unit 3, a first causal relationship extraction unit 4, a connection unit 5, a sentence vectorization unit 6, and similarity calculation. The unit 7 and the integration unit 8 are provided.

  When the trouble case 300b is input, the morpheme analysis unit 1 performs morpheme analysis. As a result, the trouble case 300b is divided with a space for each word, and is divided into words. For example, the sentence "the engine was not started" is divided into "engine", "ga", "kakekara", "not", and "ta". As a tool for morphological analysis, for example, "Mecab" may be used. Alternatively, a corpus for a forklift created in advance may be used as a morphological analysis tool. This is preferable because each noun in the trouble case 300b is appropriately extracted.

  The word vectorization unit 2 has a learned second neural network (hereinafter, simply referred to as “second NN”) 20. The second NN 20 is unsupervised in advance by using the neural network with the plurality of trouble cases 300b divided for each word as learning data, and when the plurality of trouble cases 300b divided for each word are input. , A distributed expression vector corresponding to the meaning of each word is output based on the relationship between the words included in the plurality of trouble cases 300b. The second NN20, when a highly relevant word such as a word that is likely to appear in the same event (co-occurrence) or a word that is likely to appear in a close position in the same event is input, It is pre-learned so that similar distributed expression vectors are output. This is based on the distribution hypothesis (Harrs, 1954, Rubenstein & Goodenough, 1965) that words that appear in similar contexts are also semantically similar. As the word vectorization unit 2, for example, a language model such as “Word2Vec”, “Glove”, “Fasttext” based on neural network technology may be used. For the distributed representation and Word2Vec, "MIKOLOV, Tomas, et al. Distributed representations of words and phrases and their compositionality. In: Advances in neural information processing systems. 2013. p. 3111-3119." Can be referred to. Regarding Grove, "PENNINGTON, Jeffrey; SOCHER, Richard; MANNING, Christopher D. Glove: Global Vectors for Word Representation. In: EMNLP. 2014. p. 1532-1543." Can be referred to. For Fasttext, "JOULIN, Armand, et al. Bag of tricks for efficient text classification. ArXiv preprint arXiv: 1607.01759, 2016." and the like can be referred to.

  As a method of outputting the distributed expression vector of each word, the learned second NN 20 preferably generates a distributed expression vector in consideration of subwords included in each word. Specifically, the word vectorization unit 2 divides the input word into N-grams to extract subwords, and determines the distributed expression vector corresponding to the meaning of the input word as the sum of the distributed expression vectors of the subwords or It is preferable to use an average. For example, in the case of the word "intake hose", the subword corresponds to "intake", "intake", "ki-ho", "ho", "su", and "su". As a result, the second NN 20 outputs, for example, similar distributed expression vectors to the “intake hose”, the “intake hose”, and the “hose”. Can also respond. For example, “Fasttext” may be used as a tool for generating a distributed expression vector in consideration of subwords.

  The event extracting unit 3 extracts an event from the trouble case 300b. For example, when the trouble case 300b is input, the event extracting unit 3 uses a learned third neural network that has been machine-learned using teacher data so as to extract the event included in the trouble case 300b (hereinafter, (Hereinafter simply referred to as “third NN”) 30 may be included. In this case, the teacher data is an event in which the input data is a plurality of trouble cases 300b and the correct answer data is included in each trouble case 300b. This teacher data corresponds to the "second teacher data" of the present invention. Further, in this case, it is preferable that the teacher data used for machine learning be converted into a distributed expression vector in advance by the word vectorization unit 2. This is because the words included in the teacher data have the context information because they are distributed representation vectorization, so that the features can be easily learned in machine learning. The third NN 30 may use deep learning and may use a convolutional neural network, a recurrent neural network, or both. Further, the third NN 30 may be composed of a plurality of neural networks and subjected to ensemble learning.

  The first causal relationship extracting unit 4 extracts a causal relationship between events in each trouble case 300b (hereinafter, simply referred to as "first causal relationship"). When the trouble case 300b is input, the first causal relationship extraction unit 4 performs a machine-learned first neural network (hereinafter simply referred to as “first NN”) that has been machine-learned using the teacher data to extract the first causal relationship. 40). The teacher data is a causal relationship between events in which the input data is each trouble case 300b and the correct answer data is included in each trouble case 300b. This teacher data corresponds to the "first teacher data" of the present invention. Note that the first causal relationship also includes a subordinate relationship such as which of the events having the first causal relationship is a higher event or a lower event. In addition, it is preferable that the used teacher data is distributed representation vectorized in advance by the word vectorization unit 2 from the viewpoint of accuracy and efficiency of machine learning. The first NN 40 may use deep learning and may use a convolutional neural network, a recurrent neural network, or both. The first NN 40 may be composed of a plurality of neural networks and have been ensemble-learned.

  The connection unit 5 connects the events corresponding to the first causal relationship extracted by the event extraction unit 3 in a top-down manner to generate an event group E.

  As shown in FIG. 4A, for example, in the case of the trouble case 300b related to the report 300 in FIG. 2, the connection unit 5 connects each event with a connection line to generate an event group Ea.

  The sentence vectorization unit 6 defines the event extracted by the event extraction unit 3 as a distributed expression vector for each word included in the event so that the similarity calculation unit 7 can calculate the similarity for each event. Based on the above, it is converted into a distributed expression vector for each event. The sentence vectorization unit 6 may use the average of the distributed expression vectors of the words included in the event as the feature vector of the event itself, as a method of converting the event into the distributed expression vector. Alternatively, the text vectorization unit 6 may use feature vector averaging using SCDV (Sparse Composite Document Vectors) as a method of converting the event into a distributed representation vector with higher accuracy. Regarding SCDV, “Dheeraj Mekala et al .: SCDV: Sparse Composite Document Vectors using soft clustering over distributional representations, Proc of EMNLP, 2017” can be referred to.

  The similarity calculation unit 7 calculates the similarity between the events based on the cosine similarity between the events that is distributed representation vectorized by the sentence vectorization unit 6. The cosine similarity is a similarity calculation method used when comparing documents in a vector space model. Since the cosine similarity expresses the closeness of the angle formed by the vectors, the events are more similar when the cosine similarity is close to 1, and are not so similar when the cosine similarity is close to 0.

  The integration unit 8 integrates similar events among the plurality of event groups E to form the event group E in a tree shape. For example, as shown in FIGS. 4A and 4B, when the event groups Ea and Eb are generated by the connection unit 5, the integration unit 8 includes the same “not to start engine” included in the event groups Ea and Eb. 4C, the event group Ea and the event group Eb are formed into a tree-shaped event group Ec by integrating the events of FIG.

  When the similarity between certain events is equal to or higher than a predetermined level, the integration unit 8 determines that the events are similar and integrates them.

  Next, the processing operation of the FT diagram generation apparatus A1 will be described with reference to the flowchart of FIG.

  First, the morpheme analysis unit 1 morphologically analyzes a plurality of trouble cases 300b and divides them into words (S500).

  Next, the word vectorization unit 2 converts the words included in the trouble case 300b into a distributed expression vector (S501).

  Next, the event extraction unit 3 extracts an event from the trouble case 300b that is distributed expression vectorized for each word (S502).

  Next, the first causal relationship extracting unit 4 extracts the first causal relationship from the trouble case 300b in which the distributed expression vector is converted for each word (S503).

  Next, the connection unit 5 connects the extracted events to each other based on the first causal relationship to generate the event group E (S504).

  Next, the sentence vectorization unit 6 converts the extracted event into a distributed expression vector for each event based on the included distributed expression vector for each word (S505).

  Next, the similarity calculation unit 7 calculates the similarity between the events based on the cosine similarity between the distributed expression vectors for each event (S506).

  Next, the integration unit 8 determines that the events for which the degree of similarity is equal to or greater than a predetermined value are calculated as similar events, integrates the events, and forms the related event group E in a tree shape (S507).

  FIG. 6 shows an example of the FT diagram 400 generated by the FT diagram generation device A1. As described above, the FT diagram generation device A1 can directly generate the FT diagram 400 from a plurality of trouble cases.

<Second Embodiment>
<Troubleshooting flow diagram generator>
Next, with reference to FIGS. 7 to 10, one embodiment of a troubleshooting flow diagram generation device (hereinafter, simply referred to as “flow diagram generation device”) A2 including the FT diagram generation device A1 according to the present invention will be described. explain. It should be noted that the respective configurations of the FT diagram generation device A1 included in the computer C and the flow diagram generation device A2 have already been described, and thus detailed description thereof will be omitted.

  As illustrated in FIG. 7, the computer C according to the present embodiment functions as a troubleshooting flow diagram generation device that generates a troubleshooting flow diagram when the trouble case 300b is input by the flow diagram generation program P2. Is configured.

  As shown in FIG. 8, the flow chart generation device A2 includes an FT diagram generation device A1, a treatment extraction unit 9, and a second causal relationship extraction unit 10.

  First, in the same manner as in the first embodiment, the morpheme analysis unit 1 morphologically analyzes a plurality of trouble cases 300b and divides each of the words into words (S800), and the word vectorization unit 2 expresses each of those words in a distributed expression. Vectorize (S801). Then, by the same method as in the first embodiment, the event extracting unit 3 extracts an event from the trouble case 300b in which the distributed expression vector is converted for each word (S802), and the first causal relationship extracting unit 4 causes the first causal relationship extracting unit 4 to extract the event. The relationship is extracted (S803).

  Next, the treatment extraction unit 9 extracts the treatment from the trouble case 300b that is distributed expression vectorized for each word (S804). For example, when the trouble case 300b is input, the treatment extracting unit 9 uses the learned fifth neural network that has been machine-learned using the teacher data to extract the treatment included in the trouble case 300b (hereinafter, (Hereinafter simply referred to as “fifth NN”) 90. In this case, the teacher data is a procedure in which the input data is a plurality of trouble cases 300b and the correct answer data is included in each trouble case 300b. This teacher data corresponds to the "fourth teacher data" of the present invention. Further, in this case, it is preferable that the teacher data used for machine learning be converted into a distributed expression vector in advance by the word vectorization unit 2. The fifth NN 90 may use deep learning and may use a convolutional neural network, a recurrent neural network, or both. The fifth NN 90 may be composed of a plurality of neural networks and have been ensemble-learned.

  Then, the second causal relationship extraction unit 10 determines, from the trouble case 300b that is distributed expression vectorized for each word, the causal relationship between the action for each trouble case 300b and the event corresponding to the action (hereinafter, referred to as "second causal relation" Is called) (S805). When the trouble case 300b is input, the second causal relationship extracting unit 10 performs a machine-learned fourth neural network (hereinafter simply referred to as “fourth NN”) that has been machine-learned so as to extract the second causal relationship. Say) 100. The teacher data is a causal relationship between the action in which the input data is each trouble case 300b and the correct answer data is included in each trouble case 300b and the event corresponding to the action. This teacher data corresponds to the "third teacher data" of the present invention. In addition, it is preferable that the used teacher data is distributed representation vectorized in advance by the word vectorization unit 2 from the viewpoint of accuracy and efficiency of machine learning. The fourth NN 100 may use deep learning and may use a convolutional neural network, a recurrent neural network, or both. Further, the fourth NN 100 may be composed of a plurality of neural networks and subjected to ensemble learning.

  Next, the connection unit 5 connects the extracted events to each other based on the first causal relationship (S806), and connects the treatment to the events in the event group E in a top-down manner based on the second causal relationship. Then (S807), the event group E in which the treatments are linked is generated.

  Next, the sentence vectorization unit 6 converts the extracted event and procedure into a distributed expression vector for each event or procedure based on the included distributed expression vector for each word (S808).

  Next, the similarity calculation unit 7 calculates the similarity between the events based on the cosine similarity between the distributed expression vectors for each event, and the treatment based on the cosine similarity between the distributed expression vectors for each procedure. The degree of similarity between them is calculated (S809).

  The integration unit 8 determines that the events and the treatments for which the degree of similarity is equal to or higher than a predetermined value are calculated as the same event or the similar treatment, integrates the similar events and the treatments, and creates a related event group E as a tree. It is formed into a shape (S810).

  FIG. 10 shows an example of a troubleshooting flow chart 500 generated by the flow chart generation apparatus A2. As described above, the flow diagram generation apparatus A2 can directly generate the troubleshooting flow diagram 500 from the trouble case 300b.

  Although the embodiments of the FT diagram generation apparatus A1 and the flow chart generation apparatus A2 according to the present invention have been described above, the present invention is not limited to the above embodiments.

  The FT diagram generation device A1 and the flow diagram generation device A2 may each include an editing unit that edits the generated FT diagram and flow diagram.

1 Morphological Analysis Section 2 Word Vectorization Section 20 Learned Second Neural Network 3 Event Extraction Section 30 Learned Third Neural Network 4 First Causal Relationship Extraction Section 40 Learned First Neural Network 5 Connecting Section 6 Text Vectorization Section 7 Similarity calculation section 8 Integration section 9 Treatment extraction section 90 Learned fifth neural network 10 Second causal relationship extraction section 100 Learned fourth neural network 300 Report 300b Trouble case 400 FT diagram 500 Troubleshooting flowchart C Computer A1 FT diagram generator A2 Troubleshooting flow diagram generator

Claims (6)

An FT diagram generation device for generating an FT diagram relating to the working machine from a plurality of trouble cases relating to the working machine,
An event extraction unit that extracts an event from the plurality of trouble cases,
A learned first neural network machine-learned using first teacher data so as to extract a first causal relationship between the events for each of the trouble cases when the trouble case is input; A first causal relationship extraction unit that extracts one causal relationship;
A connecting unit that connects the extracted events corresponding to the first causal relationship in a top-down manner to generate an event group;
An integrating unit that forms the plurality of event groups in a tree shape by integrating the same events among the plurality of event groups ,
A morphological analysis unit that divides the plurality of trouble cases for each word,
Learning in which unsupervised learning is performed by using a neural network with the plurality of trouble cases divided for each word as learning data, and when a word is input, learning that outputs a distributed expression vector corresponding to the meaning of the input word Completed second neural network,
A word vectorization unit that converts the trouble case divided into words into a distributed representation vector by the learned second neural network;
The extracted event, based on the distributed expression vector for each word contained in the event, a sentence vectorization unit that converts the distributed expression vector for each event,
A similarity calculation unit that calculates the similarity between the events based on the cosine similarity between the events that are distributed representation vectorized by the sentence vectorization unit,
The FT diagram generation device is characterized in that the integration unit integrates the events for which a degree of similarity equal to or greater than a predetermined value is calculated . Distributed representation vector corresponding to the meaning of the word that was the input to claim 1, wherein the words that are the input is a sum or average of the variance representation vectors subword extracted are N-gram division The described FT diagram generation device. When the trouble case is input, the event extraction unit creates a learned third neural network machine-learned using the second teacher data so as to extract the event included in the trouble case. The FT diagram generation device according to claim 1 or 2 , further comprising: A flow chart generation device for troubleshooting a work machine,
An FT diagram generation device for generating an FT diagram relating to the working machine from a plurality of trouble cases relating to the working machine;
The FT diagram generator is
An event extraction unit that extracts an event from the plurality of trouble cases,
A learned first neural network machine-learned using first teacher data so as to extract a first causal relationship between the events for each of the trouble cases when the trouble case is input; A first causal relationship extraction unit that extracts one causal relationship;
A connecting unit that connects the extracted events corresponding to the first causal relationship in a top-down manner to generate an event group;
An integrating unit that forms the plurality of event groups in a tree shape by integrating the same events among the plurality of event groups,
The flow diagram generation device further includes
A treatment extraction unit that extracts a treatment performed on the event included in the plurality of trouble cases,
When the trouble case is input, the learned fourth machine-learned using the third teacher data so as to extract the second causal relationship between the action for each trouble case and the event corresponding to the action. A second causal relationship extracting unit that has a neural network and extracts the second causal relationship;
The connection unit further connects the treatment to the events in the event group in a top-down manner based on the second causal relationship,
The flow diagram generation apparatus, wherein the integration unit further integrates the same treatments among a plurality of the event groups. The treatment extraction unit has a learned fifth neural network machine-learned using the fourth teacher data so as to extract the treatment included in the trouble case when the trouble case is input. The flow diagram generation device according to claim 4 . Program for functioning as a generator according to the computer in any one of claims 1-5.

Images