Information processing device, information processing method, and cause diagnosis system

Abstract

This can reduce the development effort required for root cause diagnosis models. [Solution] The cause diagnosis model building tool 120 acquires a flow file 100 and a table file 105 that show maintenance information including symptoms, causes, and countermeasures when the target equipment fails. From the flow file 100 and table file 105, it extracts failure modes that indicate the state of equipment components that cause the failure of the target equipment, and inspection items that indicate the state of the target equipment in order to identify the failure modes. The failure modes and inspection items are linked and stored in the temporary storage unit 150, and a Bayesian network 183 is generated based on the failure modes and inspection items.

Description

Technical Field

[0001] The present invention relates to an information processing apparatus, an information processing method, and a cause diagnosis system.

Background Art

[0002] In order to always operate industrial equipment such as air compressors and cranes, maintenance work of the equipment is essential. In particular, when the equipment fails and stops, it is required to quickly investigate the details of the failure, take measures, and resume the operation of the equipment. In the investigation, it is important to inspect each part of the equipment and identify a failure mode indicating the state of parts that cause the failure.

[0003] In order to realize such a technology, after defining a failure mode to be diagnosed and inspection items indicating the state of the equipment for identifying the failure mode, a technology of incorporating them into a rule-based diagnosis system or a model of machine learning (for example, a Bayesian network) is well-known.

[0004] However, the failure modes and inspection items differ for each piece of equipment. Therefore, it has taken a lot of man-hours to define these items manually by interviewing equipment experts or referring to manuals, and to develop a model.

[0005] As a technology for solving this, for example, in Patent Document 1, a method of generating a learned model for performing failure diagnosis by machine learning using, as input data, a failure mechanism database having a plurality of failure report sentences including a failure cause, a failure location, and a failure event, which are failure information about a failure that has occurred in a target device, is disclosed.

Prior Art Documents

Patent Documents

[0006]

Patent Document 1

Summary of the Invention

[0007] Patent Document 1 requires the preparation of a table data called a failure mechanism database to generate a trained model for fault diagnosis, which increases development time. Therefore, there is a need to generate a cause diagnosis model by extracting failure modes and inspection items from a manual that contains maintenance information for the equipment.

[0008] In equipment manuals, troubleshooting sections describing failure modes and inspection items are often presented in tables or flowcharts. In the case of tables, unlike the failure mechanism database in Patent Document 1, the names of the columns included in the table vary from manual to manual, making it unclear which column contains the failure mode or inspection item. In the case of flowcharts, it is necessary to recognize the steps included in the flowchart and their connection information. Therefore, it is difficult to extract failure modes and inspection items from equipment manuals using the technology in Patent Document 1.

[0009] The present invention aims to provide a technology that can reduce the development time required for developing a cause diagnosis model. [Means for solving the problem]

[0010] To solve the above problems, one representative information processing device of the present invention is an information processing device that generates a cause diagnosis model for diagnosing the cause of failure of a target device, comprising a control unit and a storage unit, wherein the control unit acquires a chart showing maintenance information including symptoms, causes, and countermeasures when the target device fails, extracts from the chart a failure mode indicating the state of the device's components that cause the failure of the target device, and inspection items indicating the state of the target device in order to identify the failure mode, stores the failure mode and inspection items in the storage unit in association with each other, and generates a cause diagnosis model based on the failure mode and inspection items. [Effects of the Invention]

[0011] According to the present invention, the man-hours for developing a cause diagnosis model can be reduced.

[0012] Problems, configurations, and effects other than those described above will be clarified by the description of the following embodiments.

Brief Description of the Drawings

[0013] [Figure 1A] It is a diagram showing the configuration of the cause diagnosis system of this embodiment. [Figure 1B] It is a block diagram showing an example of the hardware configuration of a computer to which the cause diagnosis system of this embodiment is applied. [Figure 2] It is a diagram showing an example of a flowchart image existing on each page of a flow file. [Figure 3] It is a diagram showing an example of a table image existing on each page of a table file. [Figure 4] It is a diagram showing an example of a table of steps and edges of a flowchart. [Figure 5] It is a diagram showing an example of the extraction result of failure modes and inspection items from a flowchart. [Figure 6] It is a diagram showing an example of the extraction result of failure modes from a table. [Figure 7] It is a diagram showing an example of the extraction result of inspection items from a table. [Figure 8] It is a diagram showing an example of a table for creating a Bayesian network before probability entry. [Figure 9] It is a diagram showing an example of a table for creating a Bayesian network. [Figure 10] It is a flowchart showing an example of the process performed by the cause diagnosis system of this embodiment. [Figure 11] It is a flowchart showing an example of the process of extracting failure modes and inspection items from the flow file of SUB01. [Figure 12] It is a flowchart showing an example of the process of extracting failure modes and inspection items from the table file of SUB02. [Figure 13]It is a flowchart showing an example of a process for creating a Bayesian network of SUB03. [Figure 14] It is a flowchart showing an example of a process for performing cause diagnosis of SUB04. [Figure 15] It is a diagram showing an example of a screen for inputting conditional probability and probability. [Figure 16] It is a diagram showing an example of a screen of the cause diagnosis system at the time of inputting inspection results. [Figure 17] It is a diagram showing an example of a screen of the cause diagnosis system at the time of displaying cause diagnosis results. [Figure 18] It is a diagram showing an example of the network structure of a Bayesian network.

Embodiments for Carrying Out the Invention

[0014] In this embodiment, an example of diagnosing the cause of a failure of an air compressor will be described from a table and a flowchart of troubleshooting related to the electrical system and the pressure system of the air compressor. In this embodiment, the air compressor is taken as an example of the target device for which cause diagnosis is performed, but it is not limited thereto.

Examples

[0015] <> First, the users and inputs / outputs of the examples will be described using FIGS. 1A and 1B showing the overall cause diagnosis system, and FIGS. 2, 3, 9, and 18. Thereafter, the processes and screen examples performed in the examples will be described. [Users of this embodiment]

[0016] FIG. 1A is a diagram showing the configuration of the cause diagnosis system of this embodiment.

[0017] The cause diagnosis system includes a cause diagnosis model construction tool 120 and a cause diagnosis processing unit 180.

[0018] In FIG. 1A, the model developer 107 and the maintenance staff 190 are users of the cause diagnosis system.

[0019] Model developer 107 is a user whose job is to create a Bayesian network 183, which is a model for root cause diagnosis. Maintenance worker 190 is a user who performs root cause diagnosis using a root cause diagnosis processing unit 180 that incorporates the created Bayesian network 183.

[0020] In this embodiment, the failure mode refers to the condition of a component that causes a failure. For example, "the cable is broken" or "the exhaust vent is clogged." The inspection items indicate the condition of the equipment in order to identify the failure mode. For example, "is the power on?" or "is the alert information displayed?" [How to use the root cause diagnosis system]

[0021] As shown in Figure 1A, the model developer 107 inputs a flow file 100 containing an image of a flowchart and a table file 105 containing an image of a table into the cause diagnosis model construction tool 120. The cause diagnosis model construction tool 120 outputs a Bayesian network 183 for diagnosing the cause of failure of the target equipment from the flow file 100 and the table file 105 to the cause diagnosis processing unit 180. This completes the preparation for cause diagnosis.

[0022] The maintenance worker 190 inputs the inspection results of the air compressor into the cause diagnosis processing unit 180. The cause diagnosis processing unit 180 uses a Bayesian network 183 to calculate the probability of failure modes from the inspection results and presents it to the maintenance worker 190.

[0023] Flow file 100 and table file 105 are files that describe maintenance information for the air compressor in flowchart and table formats, respectively. The maintenance information includes information such as symptoms, causes of failure, and countermeasures when the air compressor malfunctions.

[0024] The cause diagnosis model construction tool 120 includes a branching step detection unit 123, a treatment step detection unit 126, a column string detection unit 130, an extracted column estimation unit 133, a failure mode extraction unit 140, an inspection item extraction unit 145, a temporary storage unit 150, a multimodal generation AI 155, a probability parameter input unit 160, and a Bayesian network creation unit 165.

[0025] The branch step detection unit 123 detects branch steps that branch off from the flowchart of the flow file 100 based on conditions and decisions. The treatment step detection unit 126 detects normal steps from the flowchart of the flow file 100.

[0026] The column string detection unit 130 detects strings from each column of the table in the table file 105. The extraction column estimation unit 133 selects the columns from which to extract the failure mode and inspection items.

[0027] The failure mode extraction unit 140 extracts failure modes from normal steps detected by the treatment step detection unit 126. The failure mode extraction unit 140 also detects failure modes from columns selected by the extraction column estimation unit 133, based on the strings detected by the column string detection unit 130.

[0028] The inspection item extraction unit 145 extracts inspection items from the branching steps detected by the branching step detection unit 123. The inspection item extraction unit 145 also detects inspection items from the columns selected by the extraction column estimation unit 133, based on the strings detected by the column string detection unit 130.

[0029] The temporary storage unit 150 stores the data acquired by the branch step detection unit 123, the treatment step detection unit 126, the column string detection unit 130, the extracted column estimation unit 133, the failure mode extraction unit 140, and the inspection item extraction unit 145.

[0030] The multimodal generation AI155 uses a failure mode extraction unit 140 and an inspection item extraction unit 145 to extract failure modes and inspection items.

[0031] The probability parameter input unit 160 receives input of failure modes and inspection item probabilities from the model developer 107. The Bayesian network creation unit 165 creates a Bayesian network 183 from the failure modes and inspection items extracted by the failure mode extraction unit 140 and the inspection item extraction unit 145, and the probabilities input by the probability parameter input unit 160, and outputs it to the cause diagnosis processing unit 180.

[0032] The cause diagnosis processing unit 180 includes a failure mode probability calculation unit 185 and an input reception / presentation unit 187.

[0033] The input reception / display unit 187 receives input from the maintenance staff 190 regarding the inspection results of the air compressor (e.g., power does not turn on) when the air compressor malfunctions.

[0034] The failure mode probability calculation unit 185 uses a Bayesian network 183 to calculate the probability of the failure mode from the inspection results and presents it to the maintenance staff 190 via the input reception / presentation unit 187. [Hardware for implementing a root cause diagnosis system]

[0035] The cause diagnosis system of this embodiment is implemented using a computer such as the one shown in Figure 1B. Figure 1B is a block diagram showing an example of the hardware configuration of a computer to which the cause diagnosis system of this embodiment is applied.

[0036] Figure 1 shows an example of the hardware configuration of the computer 1000 on which the cause diagnosis model construction tool 120 and the cause diagnosis processing unit 180 shown in Figure 1 operate. The computer 1000 is a server or computer configured by interconnecting a CPU (Central Processing Unit) 1001, memory 1002, storage device 1003, communication unit 1004, input unit 1005, and output unit 1006 via a bus 1007.

[0037] The CPU 1001 is the central processing unit, and it implements the necessary functions by executing programs stored in memory 1002 (or storage device 1003). The programs include programs that implement each block shown in Figure 1A.

[0038] Memory 1002 is the main memory used by the CPU 1001 when executing processing, and is composed of volatile memory elements such as RAM (Random Access Memory).

[0039] The storage device 1003 is an auxiliary storage device for storing input data provided to the CPU 1001 and output data output from the CPU 1001, and is composed of non-volatile storage elements such as an HDD (Hard Disk Drive) or SSD (Solid State Drive).

[0040] The communication unit 1004 is an interface used by the computer 1000 to communicate with external devices, and is composed of a NIC (Network Interface Card) or the like. The communication unit 1004 is connected to a network (e.g., the Internet) and communicates with external devices via the network.

[0041] The input unit 1005 is an interface that receives input from the user (operator) and consists of a keyboard, touch panel, card reader, or voice input device, etc.

[0042] The output unit 1006 is an interface for outputting data to the operator and consists of a display, speaker, or printer, etc.

[0043] Bus 1007 is the internal communication channel of computer 1000.

[0044] In this embodiment, the cause diagnosis model construction tool 120 and the cause diagnosis processing unit 180 are executed on one or more computers 1000 having the hardware configuration illustrated in Figure 1B, thereby enabling the processing described later. [Input flow file 100]

[0045] Flow file 100 consists of multiple pages, with one flowchart present as an image on each page.

[0046] Figure 2 shows an example of a flowchart image present on each page of flow file 100.

[0047] The flowchart image 200 has, for example, a start step 210, an end step 250, multiple branch steps 220, 230, 240, and multiple normal steps 260, 270, 280.

[0048] The initial step 210 describes a major symptom: "The power won't turn on." To pinpoint the cause of this problem, the test items described in branching steps 220, 230, and 240 are performed.

[0049] For example, if the answer to branch step 220, "Is the power supply switch ON?", is "No", then the system can be restored by performing the action in normal step 260, "Turn the power supply ON". Here, the failure mode resolved by the action "Turn the power supply ON" is considered to be "The power supply is OFF". Therefore, the relationship between the test item "Is the power supply switch ON?" and the failure mode "The power supply switch is OFF" can be extracted. Similarly, the extraction of failure modes and test items can also be performed between branch step 230 and normal step 270, and between branch step 240 and normal step 280.

[0050] Furthermore, the "power does not turn on" item written in the starting step 210 can be considered an inspection item that is related to all the procedures written in the usual steps 260, 270, and 280. Therefore, the inspection item in the starting step 210 can be linked to the failure modes extracted from the usual steps 260, 270, and 280. The specific processing flow using the above principles will be described later. [Input table file 105]

[0051] Table file 105 consists of multiple pages, with one table existing as an image on each page.

[0052] Figure 3 shows an example of a table image present on each page of table file 105.

[0053] Table image 300 has, for example, a display column 310, a cause column 320, a response column 330, and a reference column 340. The display column 310 stores the error name displayed on the panel of the malfunctioning air compressor. The cause column 320 stores the cause of that error. The response column 330 stores the corrective action, error, and corrective action for restoring the air compressor. The reference column 340 stores the page number of the manual that describes the details of the cause and corrective action.

[0054] In this invention, in order to extract failure modes and inspection items from each of the columns 310, 320, 330, and 340, the extraction column estimation unit 133 estimates which column should be used to extract the failure modes and inspection items. The estimation method involves selecting the column from which the most failure modes / inspection items were extracted as the column containing the failure modes / inspection items.

[0055] The method for extracting failure modes and inspection items involves using the multimodal generation AI155 to determine and extract failure modes / inspection items from column names and text using natural language processing. Specifically, for example, from the column name "Display" and text such as "Phase Anomaly" in display column 310, natural language processing determines that it is an inspection item, and "Phase Anomaly" is extracted as the inspection item. The specific processing flow using the above principles will be described later. [Output Bayesian network 183]

[0056] Bayesian network 183 is a Bayesian network for estimating failure modes, and its conceptual structure is represented by the network structure shown in Figure 18.

[0057] Figure 18 shows an example of the network structure of Bayesian network 183.

[0058] The Bayesian network 183 has nodes 1810-1855 representing inspection items for a faulty air compressor, and node 1870 representing the failure mode. This structure is called Naive Bayes because of its simplicity, but other more complex structures can also be employed.

[0059] This Bayesian network 183 incorporates a causal relationship model in which the failure mode causing the failure of node 1870 triggers the phenomena inspected for the inspection items of nodes 1810 to 1855. The Bayesian network 183 can be used by inputting Yes / No results for the inspection items of nodes 1810 to 1855, thereby calculating the probability of the failure mode occurring in node 1870.

[0060] The probabilities of the failure modes in Figure 18 can be calculated using tables 900 and 950 shown in Figure 9. In other words, the implementation of Bayesian network 183 is the table data in Figure 9.

[0061] Figure 9 shows an example of a table for creating a Bayesian network 183.

[0062] Table 900 includes failure modes 910, test items 920, and conditional probabilities 930. Table 950 includes failure modes 960 and probabilities 970.

[0063] The inspection items for nodes 1810-1855 correspond to inspection item 920, and the failure modes for node 1870 correspond to failure modes 910 and 960. Tracing it back further, the names of the inspection items for nodes 1810-1855 and the failure modes for node 1870 were extracted from flow file 100 and table file 105.

[0064] Table 900 in Figure 9 records the conditional probability 930 of inspection item 920 occurring when failure mode 910 occurs. This probability can be manually entered by the model developer 107 or automatically entered as an equal value. Table 950 records the probability of failure mode 960 occurring. This probability can also be manually entered by the model developer 107 or automatically entered as an equal value.

[0065] Then, by using Bayes' formula with the probabilities of conditional probability 930 and probability 970 as input, the probability of failure mode occurring when the test item is Yes / No can be estimated. When test results C1, C2, C3,... corresponding to test item 920 are obtained, the i-th failure mode F among the failure modes 960 is... i Probability of occurrence P(F) i (C1, C2, C3, ...) can be calculated using the following formula 1.

[0066]

number

[0067] P(C) needed to calculate the right-hand side of equation 1 j │F i ) is failure mode F i When this occurs, the j-th test result C j This is the probability of getting [the specified value], which is the conditional probability of 930. P(F) i ) is failure mode F i This is the probability of occurrence, which is 970. [Processing of this process] [Main routine]

[0068] The main routine performed in the cause diagnosis system of this embodiment will be explained with reference to Figure 10.

[0069] Figure 10 is a flowchart showing an example of the process performed by the cause diagnosis system in this embodiment.

[0070] The main routine in Figure 10 consists of four subroutines, S1000 through S1300.

[0071] In the first half, SUB01 (S1000) is executed to extract failure modes and inspection items from flow file 100, and SUB02 (S1100) is executed to extract failure modes and inspection items from table file 105.

[0072] In the second half, SUB03 (S1200) is executed to create a Bayesian network 183 using the results from the first half, and SUB04 (S1300) is executed to perform a root cause diagnosis using the inspection results from maintenance staff 190 as input, before the process ends.

[0073] SUB01 to SUB04 are explained below. [Extract failure mode and test items from flow file (SUB01)]

[0074] The processing of this subroutine is explained using Figure 11, the input and output results of the intermediate steps are explained using Figures 4 and 5, and the final result is output as part of Table 800 shown in Figure 8.

[0075] Figure 11 is a flowchart showing an example of the process of extracting failure modes and inspection items from the SUB01 flow file.

[0076] In S1110 of Figure 11, a variable N representing the page number of the flow file 100 is defined and initialized to a value of 1. This process confirms that the extraction of failure modes and inspection items from the flowchart images 200 of all pages of the flow file 100 has been completed.

[0077] In S1120, the flowchart image 200 on page N of the flow file 100 is read into the temporary storage unit 150.

[0078] In S1130, a table 400 relating to the steps in the flowchart and a table 450 relating to the edges (arrows) connecting the steps are output to the temporary storage unit 150 from the loaded flowchart image 200, as shown in Figure 4.

[0079] Here, an ID is assigned to each step, its type is determined, and the text written in that step is read. Additionally, the edges connecting the steps are identified, and the text written on those edges is read.

[0080] This process is performed by providing the multimodal generation AI155 with prompts for image recognition. However, other image recognition techniques can also be used.

[0081] Now, let's explain Figure 4, which shows the output of S1130.

[0082] Figure 4 shows an example of a flowchart step and edge table.

[0083] Table 400 of the flowchart steps has an ID 410, a type 420, and text 430.

[0084] ID410 stores the ID of a step in the flowchart image 200 contained on page N of the flow file 100. Type420 stores the type of the step. Step types include, for example, a start step, a branch step, a normal step, and an end step. Text430 stores the text written in the step.

[0085] For example, the step where ID 410 is "1" is the start step 210 in Figure 2, so the type 420 is "Start" and the text 430 is "Power not turned on" as written in the start step 210.

[0086] Furthermore, the table 450 at the edge of the flowchart has a source ID 460, a destination ID 470, and text 480.

[0087] The source ID 460 stores the ID of the step from which the edge connection originates. The destination ID 470 stores the ID of the destination step. The text 480 stores the text, such as "yes / no," if the edge contains such text.

[0088] For example, in Figure 2, there is an edge between branch step 220 and normal step 260, and according to table 400, the IDs of branch step 220 and normal step 260 are 2 and 5. Also, this edge has the word "No". Therefore, in the 5th row of table 450, the source ID 460 is "2", the destination ID 470 is "5", and the text 480 is "No".

[0089] Returning to Figure 11, in S1140, the step where type 420 is "branch" is searched from table 400, text 430 is copied to create the inspection item 520 of the extraction result 500 in Figure 5, and stored in the primary storage unit 150.

[0090] Figure 5 shows an example of the results of extracting failure modes and inspection items from a flowchart.

[0091] Extraction result 500 includes inspection items 520 extracted from the branching step of the flowchart and failure modes 510 extracted from the normal step of the flowchart. Extraction result 550 includes inspection items 570 extracted from the branching step and the start step, and failure modes 560 extracted from the normal step of the flowchart.

[0092] In S1150, the system searches for steps where an edge exists between them and the branch step obtained in S1140, and where type 420 is "normal". The existence of an edge can be determined by searching for the ID of the branch step obtained in S1140 from the source ID 460 in table 400, and then searching for steps with destination ID 470 corresponding to the source ID 460 where type 420 is "normal".

[0093] The text 430 of the step found in table 400 describes the corrective action for the failure mode, so we extract the failure mode information from this text 430. For example, if text 430 is "Turn on the power supply," the failure mode is "The power supply is OFF." Alternatively, if text 430 is "Replace the cable," the specific condition of the cable is unknown, so we assume it is "Cable failure."

[0094] Extracting information about such failure modes can be achieved by prompting the multimodal generation AI155.

[0095] The failure modes extracted through the above process are copied and arranged in the same row as the corresponding inspection items to create failure mode 510 in the extracted result 500 of Figure 5.

[0096] In S1170, the text for the start step is added below the inspection item 520 to create inspection item 570. Furthermore, as mentioned above, the text for the start step is linked to all failure modes of failure mode 510, so the corresponding failure mode 510 is linked to inspection item 570 to create failure mode 560.

[0097] The resulting extraction result 550 is copied to rows 1-6 of failure mode 810 and inspection item 820 in Figure 8, as the final output of the failure mode and inspection item extracted from page N of flow file 100. Rows 7 and beyond in Figure 8 are created by extracting the failure mode and inspection item from table file 105 using SUB02, which will be explained next.

[0098] Figure 8 shows an example of a table used to create Bayesian network 183 before entering probabilities.

[0099] In table 800, failure modes 810 and inspection items 820 are created, and the model developer 107 inputs their conditional probabilities 830. In table 850, failure modes 860 are created, and the model developer 107 inputs their probabilities 870.

[0100] Finally, in S1180, it is checked whether the failure modes and inspection items have been extracted from the flowchart images 200 of all pages of the flow file 100. If they have not been extracted, in S1190 the variable N is set to N+1, and the process returns to S1120 to repeat.

[0101] Once the failure modes and inspection items have been extracted from the flowchart images 200 on all pages of flow file 100, the process of SUB01 is terminated. [Extract failure modes and test items from a table file (SUB02)]

[0102] The processing of this subroutine is explained using Figure 12, the input and output results of the intermediate steps are explained using Figures 6 and 7, and the final result is output as part of the table shown in Figure 8.

[0103] Figure 12 is a flowchart showing an example of the process of extracting failure modes and inspection items from the SUB02 table file.

[0104] In S1210, a variable N representing the page number of table file 105 is defined and initialized to the value 1. This process confirms that the extraction of failure modes and inspection items from table images 300 for all pages of table file 105 has been completed.

[0105] In S1215, the table image 300 on page N of the table file 105 is read into the temporary storage unit 150.

[0106] In S1230, the table image 300 is converted into searchable table data while retaining its original structure. This process is performed by providing prompts for image recognition to the multimodal generation AI 155. However, other image recognition techniques can also be used.

[0107] In S1235, the failure mode is extracted from the text of all cells in the converted table data and stored in the primary storage unit 150 as the extraction result 600 shown in Figure 6.

[0108] Figure 6 shows an example of the results of extracting failure modes from the table.

[0109] As shown in Figure 6, the extracted failure mode is stored in one of the display column 610, cause column 620, corresponding column 630, or reference column 640. These correspond to the display column 310, cause column 320, corresponding column 330, and reference column 340 of the table image 300. This can be achieved by prompting the multimodal generation AI 155, as with S1150 mentioned above. However, if a failure mode could not be extracted, as in the first row of column 610, "-" is stored.

[0110] Next, the number of failure modes extracted from each column from 610 to 640 is tallied. This tallied result is stored in the primary storage unit 150 as the number of extracted items, 650. For example, in the display column 610, only 1 out of 4 rows were extracted, so "1" is entered in column 660. In the cause column 620, 3 items were extracted, so "3" is stored in the cause column 670. The same applies to the corresponding columns 630 and 680, and the reference columns 640 and 690.

[0111] In S1240, the column with the highest extraction count of 650 is the one with the highest extraction count of 1. In this case, the corresponding column 680 has the highest extraction count of "4", so the extraction result of the corresponding column 630 is output as the failure mode and copied to rows 7-10 of failure mode 810 in Figure 8.

[0112] Here, we have chosen to output the extraction result of the column with the highest extraction count of 650 as the failure mode. However, it is also possible to use columns with extraction counts above a predetermined rank, such as outputting the extraction result of columns up to the Kth rank as the failure mode. In this case, a string can be created by concatenating the text in the cells of the same row up to the Kth rank, and the multimodal generation AI155 can be used to extract the failure mode. For example, if K=2, the extraction counts of the cause column 670 and the corresponding column 680 will be "3" and "4", respectively, making them columns up to the 2nd rank. Therefore, the text written in the cause column 620 and the corresponding column 630, for example, the first line, will contain the sentence "Power connection is reversed. Wiring is incorrect," and the failure mode can be extracted from it. The same applies to the second line and beyond.

[0113] In S1245 and S1250, the extraction target is changed from failure modes to inspection items, and the processing performed in S1235 and S1240 is repeated. The extracted inspection items are stored in the primary storage unit 150 as extraction result 700 in Figure 7, and the aggregated result of the extracted inspection items is stored as extraction count 750. In addition, the extraction results of the inspection items are copied to rows 7 to 10 of inspection items 820 in Figure 8. This links and stores the failure modes and inspection items extracted from the same row in table image 300.

[0114] Figure 7 shows an example of the results of extracting inspection items from the table.

[0115] As shown in Figure 7, the extracted failure modes are stored in one of the following columns: display column 710, cause column 720, corresponding column 730, or reference column 740. The number of inspection items extracted from each of columns 710-740 is also tallied. This tallied result, 750, is stored in the primary storage unit 150.

[0116] Finally, in S1255, it is checked whether the failure modes and inspection items have been extracted from the table images 300 of all pages of table file 105. If the extraction is not complete, in S1260 the variable N is set to N+1, and the process returns to S1215 to repeat.

[0117] Once the failure modes and inspection items have been extracted from all table images 300 of all pages in table file 105, the SUB02 process will terminate.

[0118] The processing by SUB01 and SUB02 results in the final output being as shown in Figure 8 for failure mode 810 and inspection item 820. [Creating a Bayesian Network (SUB03)]

[0119] This subroutine starts from a state where the failure modes 810 and inspection items 820 of table 800 have already been created by SUB01 and SUB02. This subroutine inputs the conditional probabilities 830 that were not entered in table 800 and creates the failure mode table 850. The process of SUB03 will be explained below using Figure 13.

[0120] Figure 13 is a flowchart showing an example of the process for creating a Bayesian network for SUB03.

[0121] In S1310 of Figure 13, first, a failure mode table 850 is created. This can be done by taking the column of failure mode 810 and removing duplicates to create failure mode 860. Note that probability 870 does not need to be entered at this point.

[0122] In S1320, the model developer 107 is presented with an input screen (Figure 15) to input conditional probability 830 and probability 870.

[0123] Figure 15 shows an example of a screen for inputting conditional probabilities 830 and 870.

[0124] In Figure 15, screen 1500 displays the text for failure mode 1510 and inspection item 1515 from failure mode 810 and inspection item 820 from Figure 8, while failure mode 1530 displays the content of failure mode 860. Screen 1500 also provides input fields for inputting conditional probability 1520, which corresponds to conditional probability 830, and probability 1535, which corresponds to probability 870.

[0125] Returning to Figure 13, in S1330, the model developer 107 enters probability values ​​into input fields 1527 and 1545 on screen 1500 in Figure 15. These probability values ​​are obtained by interviewing caregivers and others and entered into input fields 1527 and 1545. Alternatively, by pressing the auto-input buttons 1525 and 1540, the results of learning using existing statistical information / machine learning technologies, or simply equal values ​​(1 / 6 ≈ 0.166 for 6 failure modes), can be automatically entered into input fields 1527 and 1545. Once all probability values ​​have been entered, tables 900 and 950 in Figure 9 are completed, and the process proceeds to S1340.

[0126] In S1340, the probabilities entered in S1330 are placed into tables 900 and 950, and these two tables are combined to form a Bayesian network 183 which is then stored in the cause diagnosis processing unit 180. After that, this subroutine terminates. [Perform root cause diagnosis (SUB04)]

[0127] This subroutine uses the Bayesian network 183 created in SUB03 to diagnose the cause of a faulty air compressor. Specifically, maintenance worker 190 inputs inspection results from screen 1600 in Figure 16, estimates the probability of the failure mode, and displays it as shown on screen 1700 in Figure 17. The process of SUB04 will be explained below using Figure 14.

[0128] Figure 16 shows an example screen of the cause diagnosis system when inputting inspection results. Screen 1600 has an inspection result input section 1610 and a cause diagnosis result display section 1650, where the inspection item 1615 and failure mode 1655 are displayed, respectively. The probability of the failure mode 1660 is left blank. The maintenance staff 190 inputs the inspection result in the input field 1620.

[0129] Figure 14 is a flowchart showing an example of the process for performing a root cause diagnosis for SUB04.

[0130] In S1410 of Figure 14, the test items 920 and failure modes 960 from tables 900 and 950 of the Bayesian network 183 are read and displayed on screen 1600 as test items 1615 and failure modes 1655, respectively. Screen 1600 has a field for inputting test items 1615 and a field for inputting test results 1620.

[0131] In the test result input field 1620, you can select and enter a value of Yes / No / Unknown from the pull-down menu that appears when you press button 1625.

[0132] In S1420, maintenance worker 190 enters Yes / No / Unknown values ​​into the inspection result input field 1620. This should be done by simply entering the inspection results of the faulty air compressor. Once the input in the inspection result input field 1620 is complete, maintenance worker 190 presses the diagnostic button 1665 to proceed to S1430.

[0133] In S1430, the probability of each failure mode occurring is calculated from the value entered in the inspection result input field 1620, as well as the conditional probability 930 and the probability 970. The calculation result is displayed in probability 1760 in Figure 17.

[0134] Figure 17 shows an example of the screen of the cause diagnosis system when the cause diagnosis results are displayed. Screen 1700 differs from screen 1600 in that a probability of 1760 is displayed.

[0135] The 190 maintenance personnel can expect to quickly restore the air compressor by addressing the issues in order of the highest probability displayed in probability 1760. This completes the processing of subroutine SUB04, and the main routine in Figure 10 also ends.

[0136] As explained above, according to the example, the development time required for a root cause diagnosis model can be reduced by extracting failure modes and inspection items from diagrams and tables included in the manual of the target equipment.

[0137] It should be noted that the present invention is not limited to the embodiments described above, and various modifications are included. For example, the embodiments described above are explained in detail to make the present invention easier to understand, and are not necessarily limited to those having all the configurations described. [Explanation of Symbols]

[0138] 100: Flow file 105: Table file 107: Model Developer 120: Root Cause Diagnosis Model Building Tool 123: Branch step detection unit 126: Treatment step detection unit 130: Column string detection unit 133: Extraction Column Estimation Unit 140: Failure Mode Extraction Unit 145: Inspection Item Extraction Unit 150: Temporary storage 160: Probability parameter input section 165: Bayesian Network Creation Department 180: Cause Diagnosis Processing Unit 183: Bayesian Networks 185: Failure Mode Probability Calculation Unit 187: Input / Reception / Display Section 190: Maintenance staff

Claims

1. In an information processing device that generates a cause diagnosis model for diagnosing the cause of failure of a target device, It comprises a control unit and a memory unit, The control unit, Obtain a chart or graph showing maintenance information, including the symptoms, causes, and countermeasures in the event of a failure of the aforementioned target equipment. From the aforementioned diagram, a failure mode indicating the state of the components of the target equipment that causes the failure of the target equipment, and inspection items indicating the state of the target equipment in order to identify the failure mode are extracted. The failure mode and the inspection item are linked and stored in the memory unit. An information processing device that generates a cause diagnosis model based on the failure mode and the inspection items.

2. In the information processing apparatus according to claim 1, The aforementioned diagram is a flowchart having multiple steps, The control unit, Read the flowchart image mentioned above, Extract the steps and edges included in the flowchart, Read the text described in the aforementioned step and edge, An information processing device that extracts the failure mode and the inspection item based on the type of step and the text.

3. In the information processing apparatus according to claim 2, The types of steps include branching steps and normal steps. The control unit, The inspection items are extracted from the aforementioned branch step, and the failure mode is extracted from the aforementioned normal step connected to the said branch step. An information processing device that stores the relevant inspection item and the relevant failure mode in the memory unit, linking them together.

4. In the information processing apparatus described in claim 3, The type of step further includes a start step, The control unit, The inspection items are extracted from the aforementioned start step, and the failure modes are extracted from all of the aforementioned normal steps in the flowchart that include the said start step. An information processing device that stores the relevant inspection item and the relevant failure mode in the memory unit, linking them together.

5. In the information processing apparatus according to claim 1, The aforementioned figure is a table having multiple columns, The control unit, Load the image from the table above, Convert the aforementioned image into table data, The failure mode and the inspection item are extracted from the text of the table data, For each of the aforementioned columns, the number of extracted failure modes and inspection items is calculated, Based on the number of extracted items, select columns for extracting the failure mode and the inspection item used to generate the cause diagnosis model, An information processing device that associates the failure mode and the inspection item for the same row of the selected column and stores them in the storage unit.

6. In the information processing apparatus according to claim 5, The control unit is an information processing device that extracts the failure mode and the inspection item, respectively, from the column names and words contained in the text using natural language processing.

7. In the information processing apparatus according to claim 5, The control unit, An information processing device that generates the cause diagnosis model using the failure mode and the inspection item extracted from the column with the largest number of extracted items.

8. In the information processing apparatus according to claim 5, The control unit, An information processing device that generates a cause diagnosis model using the failure modes and inspection items extracted from columns whose number of extracted items is equal to or greater than a predetermined rank.

9. In the information processing apparatus according to claim 8, The control unit is an information processing device that extracts the failure mode and the inspection item by combining the text of the same row in a column whose number of extracted items is equal to or greater than a predetermined rank.

10. In the information processing apparatus according to claim 1, The control unit, The system accepts input of the failure mode and the probability of the inspection item. An information processing device that generates a Bayesian network based on the failure mode, the inspection items, and the probabilities as the cause diagnosis model.

11. In an information processing method for an information processing device that generates a cause diagnosis model for diagnosing the cause of failure of a target device, The information processing device comprises a control unit and a storage unit, The control unit, Obtain a chart or graph showing maintenance information, including the symptoms, causes, and countermeasures in the event of a failure of the aforementioned target equipment. From the aforementioned diagram, a failure mode indicating the state of the components of the target equipment that causes the failure of the target equipment, and inspection items indicating the state of the target equipment in order to identify the failure mode are extracted. The failure mode and the inspection item are linked and stored in the memory unit. An information processing method for generating a cause diagnosis model based on the failure mode and the inspection items.

12. A cause diagnosis system comprising an information processing device that generates a cause diagnosis model for diagnosing the cause of failure of a target device, and a cause diagnosis device that performs cause diagnosis using the cause diagnosis model, The aforementioned information processing device is Obtain a chart or graph showing maintenance information, including the symptoms, causes, and countermeasures in the event of a failure of the aforementioned target equipment. From the aforementioned diagram, a failure mode indicating the state of the components of the target equipment that causes the failure of the target equipment, and inspection items indicating the state of the target equipment in order to identify the failure mode are extracted. The failure mode and the inspection item are linked and stored in memory. A cause diagnosis model is generated based on the failure mode and the inspection items. The aforementioned cause diagnosis device is The inspection results for the aforementioned target equipment have been received. Using the aforementioned cause diagnosis model, the failure mode is diagnosed from the inspection results. A causal diagnosis system that presents the aforementioned diagnostic results.

Images