Learning device, management form creation assistance device, program, learning method, and management form creation assistance method

Abstract

An FMEA form creation assistance device (100) is provided with: a past case form storage unit (121) that stores past case forms that have been created in the past as management forms that include a plurality of rows; each of the plurality of rows includes at least work process information indicating one work process included in the plurality of work processes, and a risk statement indicating information related to a risk in the one work process; a learning data generation unit (111) that generates correspondence relationship learning data in which a combination of work process information included in one row among a plurality of rows in a past case form and risk sentences included in the row is used as a positive example and a combination of work process information included in the other row is used as a negative example; taking the combination of the operation process information contained in the row and the risk statement contained in the row different from the row as a negative example; and a correspondence relationship learning unit (113) that generates a correspondence relationship model using the correspondence relationship learning data.

Description

Technical Field

[0001] The present disclosure relates to a learning device, a management sheet production assisting device, a program, a learning method, and a management sheet production assisting method. Background Art

[0002] Conventionally, a quality management method called FMEA (Failure Mode Effect Analysis) is used, which uses FMEA sheets to predict possible failures that may occur during the design phase or process execution phase and to identify necessary countermeasures.

[0003] FMEA forms are often prepared based on the operator's own knowledge or experience, or past examples of adverse events, but if the operator's knowledge or experience is relied upon too much, the contents of the FMEA form will vary depending on the operator, and adverse events that the operator has not experienced may be omitted. Even when referring to past examples, it is not easy to determine the appropriate document for making the form from a large number of documents, and the operation time and labor become huge.

[0004] Patent document 1 discloses an auxiliary system for assisting the creation of the FMEA form. When the user specifies a location of a document related to the FMEA form to be read, the auxiliary system creates reference text data based on text data recorded in the specified location, and creates reference feature data that associates word relationships in the text with the strength of the degree of association. In addition, the auxiliary system also creates the same feature data for the defect case document that is the search target, calculates the similarity between these feature data, and outputs a defect case document with a high similarity.

[0005] Prior art literature

[0006] Patent Literature

[0007] Patent Document 1: Japanese Patent Application Publication No. 2011-8355 Summary of the invention

[0008] Problems to be solved by the invention

[0009] However, conventional support systems only create feature quantities for text retrieval based on text information of FMEA forms, without considering the structural characteristics of FMEA forms and FMEA forms created in the past, and sometimes fail to provide users with effective information.

[0010] Therefore, one or more aspects of the present disclosure are intended to provide more effective information when a user creates a management form.

[0011] Means for solving problems

[0012] A learning device according to one embodiment of the present invention comprises: a past case form storage unit, which stores a past case form that was created in the past as a management form, wherein the management form includes a plurality of rows, each of the plurality of rows at least including operation process information representing one operation process included in a plurality of operation processes, and a risk statement representing information related to a risk in the one operation process; a learning data generation unit, which generates correspondence learning data, wherein the correspondence learning data is learning data that uses a combination of the operation process information included in one row among the plurality of rows in the past case form and the risk statement included in the row as a positive example, and a combination of the operation process information included in the row and the risk statement included in a row different from the row as a negative example; and a correspondence learning unit, which uses the correspondence learning data to learn the correspondence between the operation process information and the risk statement, thereby generating a correspondence model.

[0013] A management form preparation assistance device according to one embodiment of the present disclosure comprises: a correspondence model storage unit, which stores a correspondence model generated by learning the correspondence between operation process information and risk statements using correspondence learning data, wherein the correspondence learning data is learning data that uses a combination of the operation process information contained in one of a plurality of rows in a past case form prepared in the past as a management form and the risk statement contained in the row as a positive example, and a combination of the operation process information contained in the row and the risk statement contained in a row different from the row as a negative example, wherein the management form comprises the plurality of rows, each of the plurality of rows comprising at least the operation process information representing one of the plurality of operation processes. process information, and the risk statement representing information related to the risk in the one work process; a document storage unit that stores a plurality of documents; an information acquisition unit that acquires work process information for retrieval, i.e., retrieval work process information; a correspondence estimation unit that adds each of the plurality of sentences contained in the plurality of documents to the retrieval work process information, thereby generating a plurality of retrieval sequence data, and in each of the plurality of documents containing each of the plurality of sentences, summing up the plurality of scores obtained by inputting the plurality of retrieval sequence data into the correspondence model, thereby determining the document with the highest summed score as reference information; and a display processing unit that generates a screen image to display the reference information.

[0014] A program of one embodiment of the present disclosure enables a computer to function as the following parts: a past case form storage unit, which stores a past case form that was created in the past as a management form, the management form including multiple rows, each of the multiple rows at least including operation process information representing one operation process included in multiple operation processes, and a risk statement representing information related to the risk in the one operation process; a learning data generation unit, which generates correspondence learning data, the correspondence learning data is learning data that uses the combination of the operation process information included in one row among the multiple rows in the past case form and the risk statement included in the row as a positive example, and uses the combination of the operation process information included in the row and the risk statement included in a row different from the row as a negative example; and a correspondence learning unit, which uses the correspondence learning data to learn the correspondence between the operation process information and the risk statement, thereby generating a correspondence model.

[0015] A program in one aspect of the present disclosure causes a computer to function as: a correspondence model storage unit that stores a correspondence model generated by learning the correspondence between operation process information and risk statements using correspondence learning data, wherein the correspondence learning data is learning data that uses a combination of the operation process information contained in one row among a plurality of rows in a past case form that was previously created as a management form and the risk statement contained in the row as a positive example, and a combination of the operation process information contained in the row and the risk statement contained in a row different from the row as a negative example, wherein the management form includes the plurality of rows, each of the plurality of rows includes at least the one representing one operation process included in the plurality of operation processes. Operation process information, and the risk statement representing information related to the risk in the one operation process; a document storage unit that stores a plurality of documents; an information acquisition unit that acquires operation process information for retrieval, i.e., retrieval operation process information; a correspondence estimation unit that adds each of the plurality of sentences contained in the plurality of documents to the retrieval operation process information, thereby generating a plurality of retrieval sequence data, and in each of the plurality of documents containing each of the plurality of sentences, summing up the plurality of scores obtained by inputting the plurality of retrieval sequence data into the correspondence model, thereby determining the document with the highest summed score as reference information; and a display processing unit that generates a screen image to display the reference information.

[0016] In a learning method of one embodiment of the present invention, correspondence learning data is generated, wherein the correspondence learning data is learning data that uses a combination of operation process information contained in one row among multiple rows in a past case form that was previously created as a management form and a risk statement contained in the row as a positive example, and a combination of the operation process information contained in the row and the risk statement contained in a row different from the row as a negative example, wherein the management form includes the multiple rows, each of the multiple rows at least includes the operation process information representing one operation process included in multiple operation processes, and the risk statement representing information related to the risk in the one operation process, and the correspondence learning data is used to learn the correspondence between the operation process information and the risk statement, thereby generating a correspondence model.

[0017] In a management form preparation assisting method of one embodiment of the present invention, work process information for retrieval, namely, retrieval work process information, is obtained, each of a plurality of sentences contained in a plurality of documents is added to the retrieval work process information, thereby generating a plurality of retrieval sequence data, and in each of the plurality of documents containing each of the plurality of sentences, a plurality of scores obtained by inputting the plurality of retrieval sequence data into a correspondence model are summed up, thereby determining a document having the highest summed score as reference information, wherein the correspondence model is generated by learning the correspondence between the work process information and the risk statements using correspondence learning data The correspondence learning data is learning data that uses a combination of the operation process information contained in one of multiple rows in a past case form that was previously created as a management form and the risk statement contained in the row as a positive example, and a combination of the operation process information contained in the row and the risk statement contained in a row different from the row as a negative example, the management form includes the multiple rows, each of the multiple rows includes at least the operation process information representing one operation process included in multiple operation processes, and the risk statement representing information related to the risk in the one operation process, and generates a screen image to display the reference information.

[0018] Effects of the Invention

[0019] According to one or more aspects of the present disclosure, more effective information can be provided when a user creates a management form. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a block diagram schematically showing the structure of the FMEA sheet preparation support device according to the first embodiment.

[0021] Figure 2 This is a schematic diagram showing the FEMA form.

[0022] Figure 3 (A) and (B) are schematic diagrams showing an example of concatenated sequence data.

[0023] Figure 4 (A) and (B) are schematic diagrams showing an example of permutation-linked sequence data.

[0024] Figure 5 This is a schematic diagram for explaining machine learning in the integrated feature learning unit 112.

[0025] Figure 6 This is a first schematic diagram for explaining machine learning in the correspondence relationship learning unit.

[0026] Figure 7 This is a second schematic diagram for explaining machine learning in the correspondence relationship learning unit.

[0027] Figure 8 (A) and (B) are schematic diagrams for explaining the processing in the correspondence relationship estimation unit.

[0028] Fig. 9 This is a block diagram showing an example of a computer.

[0029] Fig.10 This is a block diagram schematically showing the structure of the FMEA sheet preparation support device according to the second embodiment.

[0030] Fig.11 (A) and (B) are schematic diagrams showing examples of expanding the integrated feature learning data.

[0031] Fig.12 This is a block diagram schematically showing the structure of the FMEA sheet preparation support device according to the third embodiment.

[0032] Fig.13 This is a schematic diagram showing an example of additional connection sequence data.

[0033] Fig.14 This is a block diagram schematically showing the structure of the FMEA sheet preparation support device according to the fourth embodiment. DETAILED DESCRIPTION

[0034] Implementation Method 1

[0035] Figure 1 This is a block diagram schematically showing the structure of the FMEA sheet preparation support device 100 according to the first embodiment.

[0036] The FMEA form preparation support device 100 includes a pre-processing unit 110 , a storage unit 120 , a search processing unit 130 , an input unit 140 , and a display unit 150 .

[0037] The preprocessing unit 110 functions as a learning unit that learns a learning model used by the search processing unit 130 .

[0038] The preprocessing unit 110 includes a learning data generating unit 111 , an integrated feature learning unit 112 , and a correspondence relationship learning unit 113 .

[0039] The learning data generation unit 111 generates learning data used for learning. Here, the learning data generation unit 111 generates integrated feature learning data used for learning by the integrated feature learning unit 112 and correspondence relationship learning data used for learning by the correspondence relationship learning unit 113 .

[0040] Here, we first explain the FMEA form.

[0041] In FEMA, a tabular management form called a FEMA form is created to manage quality. In this FEMA form, the contents of various failures are recorded separately in multiple items. Furthermore, one or more items are set for specific solutions to the failures, and necessary matters are recorded.

[0042] Figure 2 This is a schematic diagram showing the FEMA form.

[0043] The FEMA form 101 shown in the figure is a table data having a product column 101a, a function column 101b, a process sequence 101c, a risk column 101k, an impact column 101l, a generation column 101m, a detection column 101n, and an importance column 101o. Items are stored in each of these columns.

[0044] The product column 101 a stores product identification information such as product names for identifying products manufactured through work processes.

[0045] The function column 101 b stores function identification information such as function names for identifying the functions of the product.

[0046] As described above, the product column 101 a and the function column 101 b store product information for specifying a product.

[0047] The work sequence 101c stores work process information indicating work processes for manufacturing the product.

[0048] Here, the worker sequence 101c is divided into a senior worker sequence 101d, a middle worker sequence 101e and a junior worker sequence 101f, and the junior worker sequence 101f is further divided into a "who" column 101g, a "where" column 101h, a "what to do" column 101i and a "how to do" column 101j.

[0049] In other words, the work process is classified into a large process, a medium process, and a small process. In the small process, a work process in which a certain person performs a certain thing in a certain place is managed.

[0050] The risk column 101k stores risk words which are words indicating information related to the risk in one work process.

[0051] The impact degree column 1011 stores the impact degree of the risk.

[0052] The occurrence degree column 101m stores the occurrence degree of the risk.

[0053] The detection degree column 101n stores the detection degree of the risk.

[0054] The importance column 101o stores the importance of risks.

[0055] As described above, the influence column 1011, occurrence column 101m, detection column 101n, and importance column 101o store evaluation values ​​obtained by evaluating risks. In the following, the information stored in the influence column 1011, occurrence column 101m, detection column 101n, and importance column 101o is also referred to as evaluation information.

[0056] The FMEA sheet has the above-described structure, and as described later, the FMEA sheet created in the past is stored in the storage unit 120 as a past case sheet.

[0057] Furthermore, it is assumed that the empty columns of the FMEA form contain the same content as the information originally stored above in the same column.

[0058] In addition, it is assumed that the plurality of rows included in the FMEA sheet are arranged in the order in which the work processes are performed.

[0059] As described above, the FMEA sheet functions as a management sheet including a plurality of rows, each of which includes at least work process information indicating one work process included in a plurality of work processes and a risk statement indicating information related to risks in the one work process.

[0060] return Figure 1 The learning data generating unit 111 generates learning data based on the FMEA sheet created in the past and stored in the storage unit 120. The FMEA sheet created in the past and stored in the storage unit 120 is also referred to as a past case sheet.

[0061] Here, the learning data generation unit 111 creates a pair of input data and output data as learning data for learning the content of the past example table. Here, the output data is also called supervisory data.

[0062] In the first embodiment, the learning data generating unit 111 generates integrated feature learning data for learning the sequence task as the first task and the word task as the second task, and correspondence relationship learning data for learning the correspondence relationship.

[0063] First, integrated feature learning data will be described.

[0064] The learning data generating unit 111 extracts at least the information stored in the work sequence 101c, i.e., the work process information, and the information stored in the risk column 101k, i.e., the risk phrase, as text from the stored information in units of two consecutive lines from the top of the past example table. In addition, in the case of a blank column, the corresponding information is added and extracted as text. Here, the product information stored in the product column 101a and the function column 101b is also extracted.

[0065] For example, in Figure 2 In the FEMA form 101 shown, the learning data generating unit 111 extracts the product information, work process information, and risk statements stored in the row 102 a and the product information, work process information, and risk statements stored in the row 102 b as one unit.

[0066] Then, the learning data generating unit 111 performs morphological analysis on the text extracted from one line, thereby dividing the text into each token, which is the smallest unit having meaning. The learning data generating unit 111 arranges the divided tokens in the order in which they appear in the corresponding text as a character string as sequence data.

[0067] The learning data generating unit 111 assigns a positive example label to the connected sequence data, which is data obtained by connecting the sequence data contained in one unit in the order from top to bottom of the past example table, and assigns a negative example label to the connected sequence data, which is data obtained by connecting the sequence data contained in one unit in the order from bottom to top of the past example table.

[0068] For example, Figure 3 (A) and (B) are schematic diagrams showing an example of concatenated sequence data.

[0069] Figure 3 (A) and (B) will be from Figure 2 An example of concatenating the extracted sequence data from rows 102a and 102b of FEMA form 101 is shown.

[0070] Figure 3(A) is concatenated sequence data obtained by concatenating the sequence data SDa extracted from the row 102a and the sequence data SDb extracted from the row 102b of the FEMA form 101 in the order of the row 102a and the row 102b. In this case, the concatenated sequence data is labeled as a positive example.

[0071] on the other hand, Figure 3 (B) is concatenated sequence data obtained by concatenating the sequence data SDa extracted from the row 102a and the sequence data SDb extracted from the row 102b of the FEMA form 101 in the order of the row 102b and the row 102a. In this case, a negative label is given to the concatenated sequence data.

[0072] The concatenated sequence data to which the labels are given as described above is also referred to as labeled concatenated sequence data.

[0073] Next, the learning data generating unit 111 replaces each of the multiple word-grams in the connection sequence data with a mask token, which is a special word-gram for learning, with a certain probability, thereby generating replaced connection sequence data.

[0074] Figure 4 (A) and (B) are schematic diagrams showing an example of permutation-linked sequence data.

[0075] Figure 4 (A) and (B) will also be from Figure 2 An example of concatenating the extracted sequence data from rows 102a and 102b of FEMA form 101 is shown.

[0076] like Figure 4 As shown in (A), in the concatenated sequence data obtained by concatenating the sequence data SDa extracted from the row 102a of the FEMA form 101 and the sequence data SDb extracted from the row 102b in the order of the rows 102a and 102b, a plurality of word units are replaced with the special word unit [MASK].

[0077] In addition, Figure 4 In (A), in the sequence data SDa, data in which one or more words are replaced by a special word [MASK] is set to SDa#1, and in the sequence data SDb, data in which one or more words are replaced by a special word [MASK] is set to SDb#1, but the replacement with special words is not limited to this example and can be performed randomly.

[0078] On the other hand, Figure 4As shown in (B), in the concatenated sequence data obtained by concatenating the sequence data SDa extracted from the row 102a of the FEMA form 101 and the sequence data SDb extracted from the row 102b in the order of the row 102b and the row 102a, a plurality of word units are replaced with the special word unit [MASK].

[0079] In addition, Figure 4 In (B), in the sequence data SDa, data in which one or more words are replaced by a special word [MASK] is set to SDa#2, and in the sequence data SDb, data in which one or more words are replaced by a special word [MASK] is set to SDb#2, but the replacement with special words is not limited to this example and can be performed randomly.

[0080] The above pair of permuted connected sequence data as input data and labeled connected sequence data as output data constitutes integrated feature learning data.

[0081] In other words, the learning data generating unit 111 extracts two consecutive rows, namely, the first row and the second row following the first row, from the plurality of rows of the FMEA form. Then, the learning data generating unit 111 determines a plurality of first word-grams as a plurality of word-grams by performing morphological analysis on the operation process information and the risk statement contained in the first row, and generates first sequence data in which the plurality of first word-grams are arranged. Furthermore, the learning data generating unit 111 determines a plurality of second word-grams as a plurality of word-grams by performing morphological analysis on the operation process information and the risk statement contained in the second row, and generates second sequence data in which the plurality of second word-grams are arranged. Furthermore, the learning data generating unit 111 generates first connected sequence data in which the first sequence data and the second sequence data are connected in the order of the first sequence data and the second sequence data, and generates second connected sequence data in which the first sequence data and the second sequence data are connected in the order of the second sequence data and the first sequence data. The learning data generation unit 111 changes one or more randomly selected words from a plurality of first words and a plurality of second words included in the first concatenated sequence data into masked words for making the meaning of the words unknown as first input data, and changes one or more randomly selected words from a plurality of first words and a plurality of second words included in the second concatenated sequence data into masked words as second input data. Furthermore, the learning data generation unit 111 uses the concatenated sequence data with the first label generated by adding a label of a positive example to the first concatenated sequence data as first output data of the first input data, and uses the concatenated sequence data with the second label generated by adding a label of a negative example to the second concatenated sequence data as second output data of the second input data. Furthermore, the learning data generation unit 111 generates learning data consisting of the first input data and the first output data, and the second input data and the second output data, namely, integrated feature learning data.

[0082] Next, the correspondence relationship learning data will be described.

[0083] The learning data generating unit 111 extracts at least the information stored in the work sequence 101c, i.e., the work process information, and the information stored in the risk column 101k, i.e., the risk sentence, as text from the stored information in units of one row of the past example table. In addition, in the case of a blank column, the corresponding information is added and extracted as text. Here, the product information stored in the product column 101a and the function column 101b is also extracted.

[0084] Then, the learning data generating unit 111 performs morphological analysis on the text extracted from one line, thereby dividing the text into each token, which is the smallest unit having meaning. The learning data generating unit 111 arranges the divided tokens in the order in which they appear in the corresponding text as a character string as sequence data.

[0085] The learning data generating unit 111 uses the portion other than the risk words in the sequence data as form structure information.

[0086] Then, the learning data generating unit 111 combines the risk words of another sequence data with the form structure information in one sequence data to obtain combined sequence data. Here, it is assumed that the risk words of all other sequence data are combined with the form structure information in one sequence data.

[0087] The learning data generating unit 111 uses the above sequence data and combined sequence data as input sequence data which is input data for learning the correspondence relationship.

[0088] Furthermore, the learning data generating unit 111 adds a label of a positive example to the sequence data, and adds a label of a negative example to the combined sequence data.

[0089] Then, the learning data generation unit 111 uses the sequence data to which the labels are added and the combined sequence data as output sequence data which is output data for learning the correspondence relationship.

[0090] The above input sequence data and output sequence data constitute the correspondence relationship learning data.

[0091] As described above, the learning data generating unit 111 generates correspondence learning data, which is learning data that uses the combination of the operation process information contained in one row among multiple rows in the past case form and the risk statement contained in the row as a positive example, and uses the combination of the operation process information contained in the row and the risk statement contained in a row different from the row as a negative example.

[0092] The integrated feature learning unit 112 generates an integrated feature model as a machine learning model by performing learning using the integrated feature learning data generated by the learning data generating unit 111. The generated integrated feature model is stored in the storage unit 120.

[0093] For example, the integrated feature learning unit 112 learns word units before being replaced with masked word units based on the integrated feature learning data and learns the arrangement order of the first sequence data and the second sequence data, thereby generating an integrated feature model.

[0094] For the machine learning in the integrated feature learning unit 112 , known methods described in the following documents can be used.

[0095] Literature: Jacob Devlin, Ming-WeiChang, Kenton Lee, Kristina Toutanova, "Bert: Pre-training ofdeep bidirectional transformers for language understanding", arXiv preprint arXiv:1810.04805, 2018

[0096] Figure 5 This is a schematic diagram for explaining machine learning in the integrated feature learning unit 112.

[0097] like Figure 5 As shown, by using the permuted connection sequence data InD#1 as input data and the labeled connection sequence data OuD#1 as output data, the word task of estimating the original word is learned for the special word in the permuted connection sequence data InD#1, and the learning of the row-direction features of the FMEA sheet can be performed. In addition, if the arrangement order of the sequence data included in the permuted connection sequence data InD#1 is the same as the order of the operation process, the positive example learning is performed, and if it is different, the negative example learning is performed, thereby, the column-direction features of the FMEA sheet can be learned.

[0098] In the integrated feature learning unit 112, by performing machine learning of these two tasks through multitasking, it is possible to obtain a feature quantity integrating the structural features of the FMEA sheet and the language features in the FMEA sheet as a parameter of the neural network.

[0099] return Figure 1 The correspondence relationship learning unit 113 performs learning using the correspondence relationship learning data generated by the learning data generating unit 111 , thereby generating a correspondence relationship model as a machine learning model. The generated correspondence relationship model is stored in the storage unit 120 .

[0100] For example, the correspondence relationship learning unit 113 learns at least the correspondence relationship between the work process information and the risk words using the correspondence relationship learning data, thereby generating a correspondence relationship model.

[0101] Figure 6 and Figure 7 This is a schematic diagram for explaining machine learning in the correspondence relationship learning unit 113 .

[0102] like Figure 6 As shown in FIG. 1 , when the input sequence data InD#2 is consistent with the output sequence data OuD#2 to which the positive example label is added, the correspondence relationship learning unit 113 determines it as the positive example "1". Figure 7 As shown, when the input sequence data InD#3 matches the output sequence data OuD#3 to which the label of the negative example is added, the correspondence relationship learning unit 113 determines that it is a negative example "0".

[0103] By learning the above, it is possible to learn the correspondence between the structure of the FMEA form and the risk statements. In particular, it is possible to learn the correspondence between the work process information and the risk statements.

[0104] The correspondence relationship learning unit 113 can improve the learning accuracy of the correspondence relationship by using the parameters of the integrated feature model as initial value parameters of the neural network used in machine learning.

[0105] return Figure 1 The storage unit 120 stores data and programs required for processing in the FMEA form preparation supporting device 100.

[0106] The storage unit 120 includes a past example list storage unit 121 , an integrated feature model storage unit 122 , a document storage unit 123 , and a correspondence relationship model storage unit 124 .

[0107] The past example sheet storage unit 121 stores a past example sheet which is an FMEA sheet created in the past.

[0108] The integrated feature model storage unit 122 stores the integrated feature model generated by the integrated feature learning unit 112 .

[0109] The document storage unit 123 stores a plurality of documents that are searched when creating the FMEA form.

[0110] The correspondence relationship model storage unit 124 stores the correspondence relationship model generated by the correspondence relationship learning unit 113 .

[0111] The search processing unit 130 performs a process of searching for information required when creating an FMEA sheet.

[0112] The search processing unit 130 includes an information acquisition unit 131 , a correspondence relationship estimation unit 132 , and a display processing unit 133 .

[0113] The information acquisition unit 131 acquires search information which is information for search.

[0114] Here, the information acquisition unit 131 receives input from the user via the input unit 140 to acquire search information.

[0115] In the first embodiment, the search information is assumed to include at least the work process information, and here, the description is made assuming that the search information includes the product information and the work process information. Therefore, the search information is also referred to as the work process information for search, that is, the search work process information.

[0116] The correspondence relationship estimation unit 132 estimates information required for creating an FMEA sheet using the search information, the document stored in the document storage unit 123 , and the correspondence relationship model stored in the correspondence relationship model storage unit 124 .

[0117] For example, the correspondence estimation unit 132 generates a plurality of search sequence data by linking a plurality of sentences included in a document stored in the document storage unit 123 with the search information.

[0118] Next, the correspondence relationship estimation unit 132 inputs each of the plurality of search sequence data into the correspondence relationship model, thereby acquiring the scores of each of the plurality of search sequence data.

[0119] Then, the correspondence estimation unit 132 adds the acquired scores for each document, and specifies the document having the highest added value as reference information.

[0120] Figure 8 (A) and (B) are schematic diagrams for explaining the processing in the correspondence relationship estimation unit 132.

[0121] like Figure 8 As shown in (A), the correspondence estimation unit 132 generates search sequence data SID1, SID2, and SID3 by linking a plurality of sentences SE1, SE2, and SE3 included in the document stored in the document storage unit 123 to the search information SI, respectively.

[0122] Next, the correspondence estimation unit 132 performs Figure 8 As shown in (B), by inputting the search sequence data SID1, SID2, and SID3 into the correspondence model, scores of each of the plurality of search sequence data are obtained.

[0123] Then, the correspondence estimation unit 132 adds the acquired scores for each document, and specifies the document having the highest added value as reference information.

[0124] In other words, the correspondence estimation unit 132 adds each of the multiple sentences contained in the multiple documents to the retrieval operation process information, thereby generating a plurality of retrieval sequence data, and in each of the multiple documents containing each of the multiple sentences, sums up the multiple scores obtained by inputting the multiple retrieval sequence data into the correspondence model, thereby determining the document with the highest summed score as reference information.

[0125] The display processing unit 133 generates a reference screen image indicating reference information, and causes the display unit 150 to display the reference screen image.

[0126] The input unit 140 receives input from the user.

[0127] The display unit 150 displays various screen images.

[0128] The FMEA form preparation support device 100 described above can be Fig. 9 The computer 15 shown is implemented.

[0129] The computer 15 includes a memory 10 , a processor 11 such as a CPU (Central Processing Unit), an auxiliary storage device 12 such as a HDD (Hard Disk Drive) or an SSD (Solid State Drive), a display device 13 such as a display, and an input device 14 such as a mouse or a keyboard.

[0130] For example, part or all of the pre-processing unit 110 and the search processing unit 130 can be constituted by, for example, the memory 10 and a processor 11 such as a CPU (Central Processing Unit) that executes a program stored in the memory 10 .

[0131] Such a program may be provided via a network, or may be provided by being recorded in a recording medium. In other words, such a program may be provided as a program product, for example.

[0132] The storage unit 120 can be realized by the auxiliary storage device 12 .

[0133] The display unit 150 can be realized by the display device 13 .

[0134] The input unit 140 can be realized by the input device 14 .

[0135] As described above, according to the first embodiment, when the user creates an FMEA form, information of a document in which valid sentences are described can be provided.

[0136] Implementation Method 2

[0137] Fig.10 This is a block diagram schematically showing the structure of the FMEA sheet preparation support device 200 according to the second embodiment.

[0138] The FMEA sheet preparation support device 200 includes a pre-processing unit 210 , a storage unit 120 , a search processing unit 130 , an input unit 140 , and a display unit 150 .

[0139] The storage unit 120 , search processing unit 130 , input unit 140 , and display unit 150 of the FMEA form preparation supporting device 200 of the second embodiment are the same as those of the FMEA form preparation supporting device 100 of the first embodiment.

[0140] The preprocessing unit 210 functions as a learning unit that learns a learning model used by the search processing unit 130 .

[0141] The preprocessing unit 210 includes a learning data generating unit 111 , an integrated feature learning unit 112 , a correspondence relationship learning unit 113 , and a learning data expanding unit 214 .

[0142] The learning data generating unit 111 , the integrated feature learning unit 112 , and the correspondence learning unit 113 of the preprocessing unit 210 in the second embodiment are the same as the learning data generating unit 111 , the integrated feature learning unit 112 , and the correspondence learning unit 113 of the preprocessing unit 110 in the first embodiment.

[0143] However, the learning data generating unit 111 supplies the generated integrated feature learning data to the learning data expanding unit 214 .

[0144] Furthermore, the integrated feature learning unit 112 performs learning using the integrated feature learning data expanded by the learning data expansion unit 214 .

[0145] The learning data expansion unit 214 uses at least one word element contained in the form structure information as a search query in the labeled connection sequence data contained in the integrated feature learning data generated by the learning data generation unit 111, retrieves the documents stored in the document storage unit 123, and thereby determines the sentence containing such a word element.

[0146] Then, the learning data expansion unit 214 generates new integrated feature learning data from the integrated feature learning data by replacing the determined sentence with a risk sentence containing the form structure information of the search query. Furthermore, the integrated feature learning data generated by the learning data generation unit 111 is expanded by appending the newly generated integrated feature learning data to the integrated feature learning data generated by the learning data generation unit 111.

[0147] Fig.11 This is a schematic diagram showing an example of expanding the integrated feature learning data.

[0148] Fig.11 Show according to Figure 5 The example shown is that the replaced concatenated sequence data InD#1 and the labeled concatenated sequence data OuD#1 are replaced to generate new replaced concatenated sequence data InD#3 and the labeled concatenated sequence data OuD#3.

[0149] like Fig.11 As shown, in the new replacement connection sequence data InD#3 and the labeled connection sequence data OuD#3, at least any one of "A", "A1", "A11", "#1", "$1", "%1" and "&1" is used as a search query to retrieve the document stored in the document storage unit 123, thereby Figure 5 The risky sentence L1 included in the replaced connected sequence data InD#1 and the labeled connected sequence data OuD#1 is replaced with a determined sentence L1T.

[0150] exist Fig.11 In the example, sentence L1T is determined by at least one search query among “$1”, “%1”, and “&1” included only in the sequence data configured in the front, and therefore, risk sentence L1 is replaced with sentence L1T. However, in the case where a sentence is determined by at least one search query among “A”, “A1”, “A11”, and “#1” included in the sequence data of both parties, both risk sentence L1 and risk sentence L2 are replaced by the determined sentence.

[0151] Furthermore, when a sentence is identified by at least one search query included only in “$2”, “%2”, and “&2” in the sequence data arranged at the rear, only the risk word L2 may be replaced by the identified sentence.

[0152] In other words, in embodiment 2, the learning data expansion unit 214 generates extended integrated feature learning data based on the integrated feature learning data by at least one of "replacing the risk sentences included in the first sequence data with sentences detected by searching multiple documents using the work process information included in the first sequence data" and "replacing the risk sentences included in the second sequence data with sentences detected by searching multiple documents using the work process information included in the second sequence data".

[0153] Then, the integrated feature learning unit 112 further performs learning on the expanded integrated feature learning data to generate an integrated feature model.

[0154] As described above, according to the second embodiment, in addition to the combination of form structure information and risky words, the combination of form structure information and sentences included in the associated document can be newly used as learning data, thereby expanding the learning data.

[0155] Implementation 3

[0156] Fig.12 This is a block diagram schematically showing the structure of the FMEA sheet preparation support device 300 according to the third embodiment.

[0157] The FMEA sheet preparation support device 300 includes a pre-processing unit 310 , a storage unit 120 , a search processing unit 130 , an input unit 140 , and a display unit 150 .

[0158] The storage unit 120 , search processing unit 130 , input unit 140 , and display unit 150 of the FMEA form preparation supporting device 300 of the third embodiment are the same as those of the FMEA form preparation supporting device 100 of the first embodiment.

[0159] The preprocessing unit 310 functions as a learning unit that learns a learning model used by the search processing unit 130 .

[0160] The preprocessing unit 310 includes a learning data generating unit 111 , an integrated feature learning unit 112 , a correspondence relationship learning unit 113 , and a sequence adding unit 315 .

[0161] The learning data generating unit 111 , the integrated feature learning unit 112 , and the correspondence learning unit 113 of the preprocessing unit 310 in the third embodiment are the same as the learning data generating unit 111 , the integrated feature learning unit 112 , and the correspondence learning unit 113 of the preprocessing unit 110 in the first embodiment.

[0162] However, the integrated feature learning unit 112 also performs learning using the connection sequence data added by the sequence adding unit 315 as input data.

[0163] The sequence adding unit 315 adds additional connected sequence data storing information indicating the content of each item as input data of the integrated feature learning data generated by the learning data generating unit 111 .

[0164] Fig.13 This is a schematic diagram showing an example of additional connection sequence data.

[0165] The additional connection sequence data stores information indicating what contents each information included in the permuted connection sequence data used as input data of the integrated feature learning data indicates.

[0166] For example, Fig.13 The "R1" shown is information indicating "product identification information", "R2" is information indicating "function identification information", "R3" is information indicating "large process" as a classification of operation processes, "R4" is information indicating "medium process" as a classification of operation processes, "R5" is information indicating "small process" as a classification of operation processes, and "R6" is information indicating "risk statement".

[0167] In other words, in the third embodiment, the sequence adding unit 315 adds additional sequence data indicating the contents of the plurality of first word units and the plurality of second word units before being changed to masked word units to the first input data and the second input data, respectively.

[0168] According to the third embodiment, by adding the additional connection sequence data as input data to the learning in the integrated feature learning unit 112, the structural information of the FMEA sheet can be explicitly provided, thereby improving the accuracy of machine learning.

[0169] Implementation 4

[0170] Fig.14 This is a block diagram schematically showing the structure of the FMEA sheet preparation support device 400 according to the fourth embodiment.

[0171] The FMEA sheet preparation support device 400 includes a pre-processing unit 410 , a storage unit 420 , a search processing unit 430 , an input unit 140 , and a display unit 150 .

[0172] The input unit 140 and the display unit 150 of the FMEA form preparation supporting device 400 of the fourth embodiment are the same as the input unit 140 and the display unit 150 of the FMEA form preparation supporting device 100 of the first embodiment.

[0173] The preprocessing unit 410 functions as a learning unit that learns a learning model used by the search processing unit 130 .

[0174] The preprocessing unit 410 includes a learning data generating unit 411 , an integrated feature learning unit 112 , a correspondence relationship learning unit 113 , and an evaluation learning unit 416 .

[0175] The integrated feature learning unit 112 and the correspondence relationship learning unit 113 of the preprocessing unit 410 in the fourth embodiment are the same as the integrated feature learning unit 112 and the correspondence relationship learning unit 113 of the preprocessing unit 110 in the first embodiment.

[0176] The learning data generating unit 411 generates the integrated feature learning data and the correspondence relationship learning data, as well as the evaluation learning data used for learning by the evaluation learning unit 416 , similarly to the first embodiment.

[0177] For example, the learning data generation unit 411 obtains product information, operation process information, risk statements, and evaluation information from each row of the past case form stored in the past case form storage unit 121, and generates evaluation learning data, which uses the product information, operation process information, and risk statements as input data and the product information, operation process information, risk statements, and evaluation information as output data.

[0178] The generated evaluation learning data is provided to the evaluation learning unit 416 .

[0179] The evaluation learning unit 416 generates an evaluation model as a machine learning model by performing learning using the evaluation learning data generated by the learning data generating unit 411. The generated evaluation model is stored in the storage unit 420.

[0180] Furthermore, the evaluation learning unit 416 can improve the learning accuracy of the correspondence relationship by using the parameters of the integrated feature model as initial value parameters of the neural network used in machine learning.

[0181] The storage unit 420 stores data and programs necessary for processing in the FMEA form preparation supporting device 400 .

[0182] The storage unit 420 includes a past example list storage unit 121 , an integrated feature model storage unit 122 , a document storage unit 123 , a correspondence model storage unit 124 , and an evaluation model storage unit 425 .

[0183] The past example form storage unit 121, the integrated feature model storage unit 122, the document storage unit 123, and the correspondence model storage unit 124 of the storage unit 420 in Implementation Example 4 are the same as the past example form storage unit 121, the integrated feature model storage unit 122, the document storage unit 123, and the correspondence model storage unit 124 of the storage unit 120 in Implementation Example 1.

[0184] The evaluation model storage unit 425 stores the evaluation model generated by the evaluation learning unit 416 .

[0185] The search processing unit 430 performs a process of searching for information required when creating an FMEA sheet.

[0186] The search processing unit 430 includes the information acquisition unit 131 , a correspondence relationship estimation unit 432 , and a display processing unit 433 .

[0187] The information acquisition unit 131 of the search processing unit 430 in the fourth embodiment is the same as the information acquisition unit 131 of the search processing unit 130 in the first embodiment.

[0188] Similar to the first embodiment, the correspondence relationship estimation unit 432 uses the search information, the documents stored in the document storage unit 123, and the correspondence model stored in the correspondence relationship model storage unit 124 to identify the document that best matches the search information as reference information.

[0189] Furthermore, the correspondence estimation unit 432 generates evaluation search sequence data by linking one or more sentences included in the document determined as reference information with the search information.

[0190] The correspondence estimation unit 432 may link all sentences included in the document identified as reference information with the search information, but here, the evaluation search sequence data is generated by linking a predetermined number of sentences starting from the sentence with the highest score with the search information.

[0191] Then, the correspondence relationship estimation unit 432 inputs the evaluation search sequence data into the evaluation model storage unit 425 , thereby estimating the evaluation information in the evaluation search sequence data. The estimated evaluation information is provided to the display processing unit 433 .

[0192] The display processing unit 433 generates a reference screen image showing the reference information and the evaluation information, and causes the display unit 150 to display the reference screen image.

[0193] In other words, in Embodiment 4, the learning data generation unit 411 generates evaluation learning data, which is learning data that uses the work process information included in one row among the multiple rows of the past example table as input data and the work process information and risk statements included in the row as output data. Then, the evaluation learning unit 416 uses the evaluation learning data to learn risk statements based on the work process information, thereby generating an evaluation model.

[0194] Furthermore, the correspondence estimation unit 432 adds one or more sentences selected from a plurality of sentences included in the document determined as reference information to the search operation process information, thereby generating evaluation estimation sequence data, and inputs the evaluation estimation sequence data into the evaluation model, thereby estimating the evaluation corresponding to the evaluation estimation sequence data. Then, the display processing unit 433 also displays the estimated evaluation on the screen image.

[0195] As described above, according to the fourth embodiment, it is also possible to estimate evaluation information when a document specified as reference information is used.

[0196] In the above-described embodiments 1 to 4, the parameters of the integrated feature model learned by the integrated feature learning unit 112 are set as initial parameters to learn the correspondence model or the evaluation model, but the embodiments 1 to 4 are not limited to such examples. For example, if it is possible to prepare enough learning data for learning the correspondence model or the evaluation model, it is also possible to learn the correspondence model or the evaluation model instead of the integrated feature model.

[0197] Furthermore, although the FMEA sheet preparation support devices 100 to 400 described above have both the learning function and the inference function, the first to fourth embodiments are not limited to such an example.

[0198] For example, a learning device (not shown) may be formed by a part that performs a learning function, such as a preprocessing unit 110 to 410, storage units 120, 420, an input unit 140, and a display unit 150, and an inference device (not shown) or a management form creation auxiliary device may be formed by a part that performs an inference function, such as a storage unit 120, 420, a retrieval processing unit 130, 430, an input unit 140, and a display unit 150.

[0199] In addition, the storage units 120 and 420 may be provided in an external device.

[0200] Description of symbols

[0201] 100, 200, 300, 400: FMEA form preparation auxiliary device; 110, 210, 310, 410: pre-processing unit; 111, 411: learning data generation unit; 112: integrated feature learning unit; 113: correspondence learning unit; 214: learning data expansion unit; 315: sequence appending unit; 416: evaluation learning unit; 120, 420: storage unit; 121: past case form storage unit; 122: integrated feature model storage unit; 123: document storage unit; 124: correspondence model storage unit; 425: evaluation model storage unit; 130, 430: retrieval processing unit; 131: information acquisition unit; 132, 432: correspondence estimation unit; 133, 433: display processing unit; 140: input unit; 150: display unit.

Claims

1. A learning device, characterized in that: The learning device has: a past case form storage unit storing a past case form created in the past as a management form, the management form including a plurality of rows, each of the plurality of rows including at least work process information indicating one work process included in a plurality of work processes, and a risk statement indicating information related to a risk in the one work process; a learning data generating unit that generates correspondence learning data, wherein the correspondence learning data is learning data that uses a combination of the operation process information included in one of the multiple rows in the past example form and the risk statement included in the row as a positive example, and uses a combination of the operation process information included in the row and the risk statement included in a row different from the row as a negative example; as well as The correspondence relationship learning unit learns the correspondence relationship between the work process information and the risk words using the correspondence relationship learning data, thereby generating a correspondence relationship model.

2. The learning device according to claim 1, characterized in that The plurality of rows are arranged in the order in which the plurality of operation steps are performed, The learning data generating unit extracts two consecutive rows, namely, a first row and a second row sequentially following the first row, from the plurality of rows. The learning data generating unit performs morphological analysis on the work process information and the risk sentence included in the first line to thereby determine a plurality of first word units as a plurality of word units, and generates first sequence data in which the plurality of first word units are arranged. The learning data generating unit performs morphological analysis on the work process information and the risk sentence included in the second line to thereby determine a plurality of second word units as a plurality of word units, and generates second sequence data in which the plurality of second word units are arranged. The learning data generating unit generates first connected sequence data in which the first sequence data and the second sequence data are connected in the order of the first sequence data and the second sequence data. The learning data generating unit generates second connected sequence data by connecting the first sequence data and the second sequence data in the order of the second sequence data and the first sequence data. The learning data generating unit changes one or more word-grams randomly selected from the plurality of first word-grams and the plurality of second word-grams included in the first connection sequence data into masked word-grams for making the meaning of the word-gram unknown, thereby providing the first input data, The learning data generating unit replaces one or more word units randomly selected from the plurality of first word units and the plurality of second word units included in the second connection sequence data with the masked word units as the second input data, The learning data generation unit generates the first labeled connected sequence data as first output data by adding a label of a positive example to the first connected sequence data, wherein the first output data is output data of the first input data. The learning data generation unit generates the second labeled concatenated sequence data by adding a negative example label to the second concatenated sequence data as second output data, wherein the second output data is output data of the second input data. The learning data generating unit generates integrated feature learning data, wherein the integrated feature learning data is learning data composed of the first input data and the first output data, and the second input data and the second output data. The learning device further includes an integrated feature learning unit, which learns the word unit before being replaced by the masked word unit based on the integrated feature learning data, and learns the arrangement order of the first sequence data and the second sequence data, thereby generating an integrated feature model. The correspondence relationship learning unit sets the parameters of the integrated feature model as initial parameters and performs learning of the correspondence relationship model.

3. The learning device according to claim 2, characterized in that: The learning device also has: a document storage unit that stores a plurality of documents; and a learning data expansion unit that generates expanded integrated feature learning data based on the integrated feature learning data by at least one of "replacing the risk sentence included in the first sequence data with a sentence detected by searching the plurality of documents using the work process information included in the first sequence data" and "replacing the risk sentence included in the second sequence data with a sentence detected by searching the plurality of documents using the work process information included in the second sequence data", The integrated feature learning unit further performs learning on the extended integrated feature learning data to generate the integrated feature model.

4. The learning device according to claim 2, characterized in that: The learning device further includes a sequence adding unit that adds additional sequence data indicating the contents of the plurality of first word units and the plurality of second word units before being changed to the masked word units, to the first input data and the second input data, respectively.

5. The learning device according to any one of claims 1 to 4, characterized in that: The learning data generating unit generates evaluation learning data, wherein the evaluation learning data is learning data that uses the operation process information included in one of the plurality of rows in the past example form as input data and uses the operation process information and the risk statement included in the one row as output data, The learning device further includes an evaluation learning unit configured to generate an evaluation model by learning the risk sentences based on the work process information using the evaluation learning data.

6. A management form preparation auxiliary device, characterized in that: The management form making auxiliary device has: a correspondence model storage unit storing a correspondence model generated by learning the correspondence between operation process information and risk statements using correspondence learning data, the correspondence learning data being learning data that uses a combination of the operation process information included in one row among a plurality of rows in a past case form that was previously created as a management form and the risk statement included in the row as a positive example, and a combination of the operation process information included in the row and the risk statement included in a row different from the row as a negative example, the management form including the plurality of rows, each of the plurality of rows including at least the operation process information indicating one operation process included in a plurality of operation processes and the risk statement indicating information related to the risk in the one operation process; A document storage unit storing a plurality of documents; An information acquisition unit that acquires operation process information for retrieval, that is, retrieves the operation process information; a correspondence estimation unit that adds each of the plurality of sentences included in the plurality of documents to the search operation process information, thereby generating a plurality of search sequence data, and summing up a plurality of scores obtained by inputting the plurality of search sequence data into the correspondence model for each of the plurality of documents including each of the plurality of sentences, thereby determining a document having the highest summed score as reference information; as well as A display processing unit generates a screen image to display the reference information.

7. The management form preparation auxiliary device according to claim 6, characterized in that: The management form preparation auxiliary device further includes an evaluation model storage unit for storing an evaluation model, wherein the evaluation model is a learning model generated by learning the risk statement based on the operation process information using evaluation learning data, wherein the evaluation learning data is learning data that uses the operation process information included in the row as input data and uses the operation process information and the risk statement included in the row as output data, The correspondence estimation unit adds one or more sentences selected from a plurality of sentences included in the document determined as the reference information to the search operation process information, thereby generating sequence data for evaluation estimation, and inputs the sequence data for evaluation estimation into the evaluation model, thereby estimating the evaluation corresponding to the sequence data for evaluation estimation. The display processing unit displays the estimated evaluation also on the screen image.

8. A program, characterized in that The program enables the computer to function as: a past case form storage unit storing a past case form created in the past as a management form, the management form including a plurality of rows, each of the plurality of rows including at least work process information indicating one work process included in a plurality of work processes, and a risk statement indicating information related to a risk in the one work process; a learning data generating unit that generates correspondence learning data, wherein the correspondence learning data is learning data that uses a combination of the operation process information included in one of the multiple rows in the past example form and the risk statement included in the row as a positive example, and uses a combination of the operation process information included in the row and the risk statement included in a row different from the row as a negative example; as well as The correspondence relationship learning unit learns the correspondence relationship between the work process information and the risk words using the correspondence relationship learning data, thereby generating a correspondence relationship model.

9. A program, characterized in that The program enables the computer to function as: a correspondence model storage unit storing a correspondence model generated by learning the correspondence between operation process information and risk statements using correspondence learning data, the correspondence learning data being learning data that uses a combination of the operation process information included in one row among a plurality of rows in a past case form that was previously created as a management form and the risk statement included in the row as a positive example, and a combination of the operation process information included in the row and the risk statement included in a row different from the row as a negative example, the management form including the plurality of rows, each of the plurality of rows including at least the operation process information indicating one operation process included in a plurality of operation processes and the risk statement indicating information related to the risk in the one operation process; A document storage unit storing a plurality of documents; An information acquisition unit that acquires operation process information for retrieval, that is, retrieves the operation process information; a correspondence estimation unit that adds each of the plurality of sentences included in the plurality of documents to the search operation process information, thereby generating a plurality of search sequence data, and summing up a plurality of scores obtained by inputting the plurality of search sequence data into the correspondence model for each of the plurality of documents including each of the plurality of sentences, thereby determining a document having the highest summed score as reference information; as well as A display processing unit generates a screen image to display the reference information.

10. A learning method, characterized in that: Generate correspondence learning data, the correspondence learning data is learning data that uses a combination of operation process information contained in one of a plurality of rows in a past case form that was previously created as a management form and a risk statement contained in the row as a positive example, and uses a combination of the operation process information contained in the row and the risk statement contained in a row different from the row as a negative example, the management form includes the plurality of rows, each of the plurality of rows at least including the operation process information representing one operation process included in the plurality of operation processes, and the risk statement representing information related to the risk in the one operation process, The correspondence relationship model is generated by learning the correspondence relationship between the work process information and the risk words using the correspondence relationship learning data.

11. A management form production auxiliary method, characterized in that: Acquiring the work process information for retrieval means retrieving the work process information. Each of a plurality of sentences included in a plurality of documents is added to the retrieval operation process information, thereby generating a plurality of retrieval sequence data, and in each of the plurality of documents including each of the plurality of sentences, a plurality of scores obtained by inputting the plurality of retrieval sequence data into a correspondence model are summed up, thereby determining a document having the highest summed score as reference information, wherein the correspondence model is generated by learning the correspondence between operation process information and risk statements using correspondence learning data, wherein the correspondence learning data is learning data that uses a combination of the operation process information included in one of a plurality of rows in a past case form that has been prepared in the past as a management form and the risk statement included in the row as a positive example, and a combination of the operation process information included in the row and the risk statement included in a row different from the row as a negative example, wherein the management form includes the plurality of rows, each of the plurality of rows includes at least the operation process information indicating one operation process included in a plurality of operation processes and the risk statement indicating information related to the risk in the one operation process, A screen image is generated to display the reference information.

Images