Failure cause analysis device and failure cause analysis method

Abstract

The present invention addresses the problem in that when AI training is performed on the basis of the work record of a low-skilled worker, erroneous inference results are obtained for a failure cause and a countermeasure plan.　A representative failure cause analysis device according to the present invention comprises: a work record storage unit that stores work records including a failure event that has occurred, a failure cause, and information about a countermeasure work for the failure event; a work record selection unit that selects, from the work records stored in the work record storage unit, a work record to be used for training a generative AI on the basis of a predetermined condition; and an inference unit that trains the generative AI for a newly occurred failure event on the basis of the work record selected by the work record selection unit, and infers a failure cause and a countermeasure work for the newly occurred failure event using the trained generative AI.

Description

Failure Cause Analysis Device and Failure Cause Analysis Method

[0001] The present invention relates to an input device for work records and a cause analysis device for failure events in maintenance work of various plants such as manufacturing plants, power plants, water supply facilities, and various devices such as elevators, railways, and power conversion devices.

[0002] Conventionally, when detecting an abnormality in a plant, a technique for assisting work by showing work records in past events in operation has been disclosed (for example, Patent Document 1). In the technique described in Patent Document 1, when receiving abnormality information, an artificial intelligence platform searches a event database for past events highly relevant to the abnormal event and provides the search results.

[0003] Japanese Patent Application Laid-Open No. 2020-201764

[0004] When utilizing artificial intelligence technology represented by the rapidly developing generative AI technology in recent years for work support as described above, it is desirable to use high-quality work records as work records for artificial intelligence learning. AI learning based on work records has the problem that while it is relatively easy to collect learning data, incorrect inferences are likely to occur when learning from work records made by workers with low proficiency. In addition, in high-reliability devices such as manufacturing plant facilities, since the number of occurrences of failure events is small, there is a problem that learning data is insufficient.

[0005] The present invention has been made in consideration of such circumstances, and aims to provide a failure cause analysis device that performs AI learning based on work records selected by maintenance workers to analyze the cause of failure, and allows the maintenance worker to evaluate the failure cause analysis result by AI and use it for AI learning.

[0006] To solve the above problems, one representative fault cause analysis device of the present invention includes: a work record storage unit that stores work records including an occurrence of a fault event, the cause of the fault, and information on countermeasures taken for the fault event; a work record selection unit that selects work records to be used for training a generating AI from the work records stored in the work record storage unit based on predetermined conditions; and an estimation unit that, in response to a newly occurring fault event, trains the generating AI based on the work records selected by the work record selection unit, and estimates the cause of the fault and countermeasures for the newly occurring fault event using the trained generating AI.

[0007] According to the present invention, when a user inputs a failure event, they can select a work record to be used for learning, which has the effect of enabling an analysis of the cause of the failure according to the target equipment and situation.

[0008] Other issues, configurations, and effects not mentioned above will be clarified by the following description of the embodiments.

[0009] This is an overall configuration diagram including the fault cause analysis device and its peripheral devices. This is a configuration diagram of the fault cause analysis device. This is a schematic diagram of the functional configuration of the fault cause analysis device. This is a diagram detailing the work record input / output unit. This is a diagram detailing the fault analysis input / output unit. This is a diagram detailing the work record unit. This is a diagram detailing the cause / countermeasure estimation unit. This is a diagram detailing the user interface of the fault analysis input / output unit. This is a diagram detailing the work record information. This is a diagram detailing the proposal record information. This is a diagram detailing the learning data selection information. This is a diagram flowchart of the work record selection unit. This is a diagram flowchart of the proposal history selection unit. This is a diagram showing the output of the work record selection unit. This is a diagram flowchart of the estimated cause / countermeasure proposal generation unit. This is a diagram detailing the instruction to generate an estimated cause / countermeasure proposal report. This is a diagram detailing the estimated cause / countermeasure proposal report generation result. This is a diagram detailing the user interface of the fault analysis input / output unit. This is a diagram detailing the learning data selection information. This is a diagram showing the output of the work record selection unit. This is a diagram showing the output of the proposal history selection unit. This is a diagram detailing the instruction to generate an estimated cause / countermeasure proposal report.

[0010] The following examples will be described with reference to the drawings. The examples are illustrative for explaining the present invention, and have been omitted and simplified as appropriate for clarity of explanation. The present invention can also be implemented in various other forms. Unless otherwise specified, each component may be singular or plural.

[0011] The positions, sizes, shapes, and ranges of the components shown in the drawings may not represent their actual positions, sizes, shapes, and ranges in order to facilitate understanding of the invention. Therefore, the present invention is not necessarily limited to the positions, sizes, shapes, and ranges disclosed in the drawings.

[0012] Examples of various types of information may be described using terms such as "table" and "list," but these types of information may also be represented by other data structures. For example, various types of information such as "XX table" and "XX list" may also be referred to as "XX information." When describing identification information, terms such as "identification information," "identifier," "name," "ID," and "number" are used, but these terms are interchangeable.

[0013] When there are multiple components with the same or similar function, they may be described using the same symbol but with different subscripts. Furthermore, when it is not necessary to distinguish between these multiple components, the subscripts may be omitted in the description.

[0014] In the examples, the processes performed by executing a program may be described. Here, the computer executes the program using a processor (e.g., CPU, GPU) and performs the processing defined in the program using memory resources (e.g., memory) and interface devices (e.g., communication ports). Therefore, the main entity performing the processing by executing the program may be the processor. Similarly, the main entity performing the processing by executing the program may be a controller, device, system, computer, or node having a processor. The main entity performing the processing by executing the program may be an arithmetic unit, and may include dedicated circuits that perform specific processing. Here, dedicated circuits include, for example, FPGAs (Field Programmable Gate Arrays), ASICs (Application Specific Integrated Circuits), CPLDs (Complex Programmable Logic Devices), etc.

[0015] The program may be installed on the computer from the program source. The program source may be, for example, a program distribution server or a storage medium readable by the computer. If the program source is a program distribution server, the program distribution server includes a processor and storage resources for storing the program to be distributed, and the processor of the program distribution server may distribute the program to other computers. In addition, in the embodiment, two or more programs may be implemented as one program, or one program may be implemented as two or more programs.

[0016] Figure 1A is an overall configuration diagram of the fault cause analysis device 100 and its peripheral devices in this embodiment. The fault cause analysis device 100 is a computer device having a processing unit and a memory device, and is connected by wired or wireless means to input / output devices 200 such as a display device, keyboard, mouse, and camera for users (staff, maintenance personnel, etc.) to operate and input necessary data and to check input / output data. The input / output devices 200 may be tablet terminals or the like held by the user. Furthermore, the fault cause analysis device 100 is connected by wired or wireless means to a large-scale language model (LLM) 300 provided on the cloud via a communication network 400, making the LLM 300 available for use.

[0017] Figure 1B is a diagram showing the configuration of the fault cause analysis device 100, which includes a main memory 110, an auxiliary memory 130, an arithmetic processing unit 140, an input / output interface 150, a communication interface 160, and bus signal lines 170 connecting these.

[0018] The main memory 110 includes an input / output unit 111 which includes a work record input / output unit 112 and a fault analysis input / output unit 113, and a fault analysis unit 120 which has a work record unit 121 and a cause / countermeasure estimation unit 122. The cause / countermeasure estimation unit 122 further includes a learning data selection unit 123 which includes a work record selection unit 125 and a proposal history selection unit 126, and an estimated cause / countermeasure proposal generation unit 124. Each of these units in the main memory 110 is an application program (software), and these units function when the arithmetic processing unit 140, which will be described later, reads and executes these programs.

[0019] The auxiliary storage device 130 has a work record information database (DB) 131 and a proposal record DB 132. The arithmetic processing unit 140 is a processor device of a computer device, and in addition to reading and executing each program stored in the main memory device 110 described above, it also performs other controls on the fault cause analysis device 100 itself based on control programs (not shown) and performs predetermined processing based on predetermined programs. In the processing of the fault cause analysis device 100 in this embodiment, in cases where a specific processing entity is not specified in the description, the processing is performed by the arithmetic processing unit 140 executing a predetermined program.

[0020] The input / output interface 150 is an interface with the input / output device 200 and provides a user interface function for inputting predetermined information to the fault cause analysis device 100 and displaying processing results from the fault cause analysis device 100.

[0021] The communication interface 160 provides an interface function for connecting to other devices, such as a large-scale language model 300 provided in the cloud, via a communication network 400, either by wire or wirelessly.

[0022] Next, we will describe in more detail each part of the fault cause analysis device 100. Figure 1C is a schematic diagram of the functional configuration of the input / output unit 111 and the fault analysis unit 120, which are the main components of the fault cause analysis device 100 in this embodiment, and the arrows in the figure indicate the flow of processing and data.

[0023] The fault analysis unit 120 analyzes the cause of plant or equipment failures and proposes countermeasures. The input / output unit 111 performs input and output to and from the user. The large-scale language model 300 performs fault cause estimation and countermeasure proposal formulation as instructed by the fault analysis unit 120.

[0024] The work record input / output unit 112 receives work record input from the user, AI suggested results (estimated cause and proposed countermeasures), and the user's evaluation thereof, and transmits them to the work record unit 121.

[0025] The work record unit 121 stores the work record information and AI suggestion results received from the work record input / output unit 112 in a predetermined storage area, and also reads the work record information and AI suggestion results from the predetermined storage area and transmits them to the cause / countermeasure estimation unit 122.

[0026] The fault analysis input / output unit 113 receives fault status information and learning data selection information from the user and transmits it to the cause / countermeasure estimation unit 122. It also receives estimated cause and countermeasure proposals (AI proposed results) from the cause / countermeasure estimation unit 122 and presents them to the user. Furthermore, it transmits the estimated cause and countermeasure proposals to the work record input / output unit 5 for the user to evaluate them.

[0027] The cause and countermeasure estimation unit 122 uses a large-scale language model 300 to estimate the cause and formulate countermeasures based on the failure status information and learning data selection information received from the failure analysis input / output unit 113, and transmits them to the failure analysis input / output unit 113.

[0028] Figure 2 shows details of the work record input / output unit 112, which has a work record input unit 112A and a proposed record evaluation input unit 112B. The work record input unit 112A receives work record input information 112Ai from the user and transmits it to the work record unit 121 as work record input information 112Ao. The proposed record evaluation input unit 112B receives estimated cause / countermeasure information 113Co from the fault analysis input / output unit 113 and its evaluation input 112Bi from the user, and based on this, transmits it to the work record unit 121 via the work record input unit 112A as work record information, or transmits it to the work record unit 121 as proposed record information 112Bo. Here, work record input information 112Ai is information entered by a user (worker) after responding to a fault in a plant or equipment, including the fault situation, cause, and implemented countermeasures.

[0029] The Proposal Record Evaluation Input Unit 112B receives user evaluations (evaluation input 112Bi) regarding whether the estimated cause and countermeasure proposal (estimated cause and countermeasure information 113Co) output by the cause and countermeasure proposal generation unit 122 is useful as an estimation result, and outputs the estimated cause and countermeasure proposal as work record information or proposal record information according to the evaluation content. That is, if the estimated cause and countermeasure proposal (estimated cause and countermeasure information 113Co) output by the system is adopted by the worker (when the user's evaluation input (evaluation input 112Bi) is "ACCEPT"), it is sent to the work record input unit 112A as work record input information. Also, if the cause estimation and countermeasure proposal output by the system is evaluated as useful information but is not adopted (when the user's evaluation input is "GOOD"), it is sent to the work record unit 121 as proposal record information. If the cause estimation and proposed countermeasures output by the system are evaluated as not being useful information, for example, being unrelated to the current incident (i.e., if the user's evaluation input 112Bi is "BAD"), then those cause estimations and countermeasures will be discarded.

[0030] Figure 3 shows details of the failure analysis input / output unit 113, which includes a learning data selection information input unit 113A, a failure status information input unit 113B, and an estimated cause / countermeasure information output unit 113C. The learning data selection information input unit 113A and the failure status information input unit 113B each receive learning data selection information 113Ai and failure status information 113Bi from the user, and transmit them to the cause / countermeasure estimation unit 122 as learning data selection information 113Ao and failure status information 113Bo, respectively. The estimated cause / countermeasure information output unit 113C receives estimated cause / countermeasure information 113Ci from the cause / countermeasure estimation unit 122 and outputs it to the user and the work record input / output unit 112 as estimated cause / countermeasure information 113Co.

[0031] Here, the learning data selection information 113Ai is used by the user to input selection criteria for learning data to be used by the generating AI (large-scale language model 300) when the cause / countermeasure estimation unit 122 plans the cause estimation and countermeasures.

[0032] Furthermore, the fault status information 113Bi is used by the user, a worker, to input the details of a fault when responding to a new fault occurring in the plant or equipment.

[0033] Furthermore, the estimated cause and proposed countermeasure information 113Ci is information resulting from the cause and countermeasure estimation unit 122 generating the cause and proposed countermeasure using a generated AI (large-scale language model 300) based on selected data, i.e., past work records and proposed records, in response to the failure status information (113Bi, 113Bo) entered by the user.

[0034] Figure 4 shows details of the work record unit 121, which has a work record R / W unit 121A and a suggestion record R / W unit 121B. The work record unit 121 has the function of writing information to the work record database 131 and the suggestion record database 132 in the auxiliary storage device, and reading information from them and outputting it. That is, the work record R / W unit 121A writes work record input information 112Ao from the work record input / output unit 112 to the work record DB 131, and also reads work record information 131o stored in the work record DB 131 and transmits it to the cause / countermeasure estimation unit 122. Similarly, the suggestion record R / W unit 121B writes suggestion record information 112Bo from the work record input / output unit 112 to the suggestion record DB 132, and also reads suggestion record information 132o stored in the suggestion record DB 132 and transmits it to the cause / countermeasure estimation unit 122.

[0035] Figure 5 shows the details of the cause and countermeasure estimation unit 122, which consists of a learning data selection unit 123 and an estimated cause and countermeasure proposal generation unit 124, and the learning data selection unit 123 is further composed of a work record selection unit 125 and a proposal history selection unit 126.

[0036] The work record selection unit 125 selects work record information 131o read from the work record database 131 based on the learning data selection information 113Ao from the fault analysis input / output unit 113. The selected work record information 131o' is used as learning data when the estimated cause / countermeasure proposal generation unit 124 requests the large-scale language model unit 300 to estimate the cause and formulate a countermeasure proposal. In this embodiment, considering the possibility that only a small amount of work data, such as a few entries, may be stored in the work record database 131, the learning method is assumed to be Few-Shot learning, but the same applies even in situations where a large amount of data is stored in the work record database 131.

[0037] The proposal history selection unit 126 selects proposal record information 132o read from the proposal record database 132 based on the learning data selection information 113Ao from the failure analysis input / output unit 113. The selected proposal record information 132o' is used as data for few-shot learning when the estimated cause / countermeasure proposal generation unit 124 requests the large-scale language model unit 300 to estimate the cause and formulate a countermeasure proposal.

[0038] The estimated cause and proposed countermeasure generation unit 124, based on the work record information 131o' from the work record selection unit 125 and the proposed record information 132o' from the proposed history selection unit 126, instructs the large-scale language model 300 to perform few-shot learning. Based on this few-shot learning, it instructs the fault analysis input / output unit 113 to estimate the cause and formulate a proposed countermeasure for the fault status information 113Bo from the fault analysis input / output unit 113, and transmits the result to the fault analysis input / output unit 113 as estimated cause and proposed countermeasure information 113Ci.

[0039] Figure 6 shows details of the user interface 600 generated by the fault analysis input / output unit 113, and is an example of a screen displayed on the display device of the input / output device 200 via the input / output interface 150.

[0040] The status display unit 610 indicates the status of the current input screen. The "fault analysis mode" in FIG. 6 indicates that the current input screen is an input screen for the fault situation and an output screen for cause estimation. The user selects each input item and performs input using an input device such as the keyboard / mouse of the input / output device 200.

[0041] In the date / time input field 611, the current date and time are input. The date / time information can be automatically input as the current time in addition to manual input by the user.

[0042] In the device ID input field 612, the device ID of the work target is input. By pressing the barcode reading button 613, the camera of the input / output device 200 can be activated to read the barcode attached to the device.

[0043] In the operator input field 614, the ID of the operator is input. By pressing the barcode reading button 615, the camera of the input / output device 200 can be activated to read the barcode attached to the operator's name tag, work clothes, helmet, etc.

[0044] In the learning data selection field 616, as the learning data selection information 113Ai, which is the input information to the learning data selection information input unit 113A in FIG. 3, when inferring by the generation AI (large language model 300), the conditions of the work record used for few-shot learning and the proposal record by the generation AI are input. That is, as the date / time 616A, whether to specify the entire range or a specific date / time is specified. As the device ID 616B, whether to target all devices of the same type as the target device (ALL) or only the target device is specified. As the operator level 616C, whether to target all operators (ALL) or only senior operators is specified. As the past AI proposal 616D, whether to include the countermeasure proposals proposed by the past AI in the few-shot learning data or not is specified (in this example, it shows the state set not to include).

[0045] The user input field prompt section 620 and the support AI output field prompt section 630 indicate that the balloon on the left side of the screen represents user input and the balloon on the right side of the screen represents AI output. User input 621 indicates free-form input of the failure situation by the user, showing a state where the content "the safety valve of the pressure switch PS01 has activated" has been input as failure situation information 113Bi, which is the input information to the failure situation information input section 113B in FIG. 3.

[0046] AI output 631 indicates the results of the AI's analysis of the cause of the failure and proposed countermeasures, showing an example of the estimated cause and countermeasure case information 113Co, which is the output information from the estimated cause and countermeasure case information output section 113C in FIG. 3. That is, as failure situation information 631A, it indicates the failure situation information input by the user in user input 621. "(1) Clogging of the filter FL02" 631B and "(2) Pressure increase due to rising outside air temperature" 631C respectively indicate the estimated cause, countermeasure, and confirmation location generated by the generation AI.

[0047] The user evaluation input fields 631Ba and 631Ca are used to input the user evaluation (112Bi) of these cause analyses and proposals output by the generation AI. "ACCEPT" is selected when adopting the AI output as the work content. "GOOD" is selected when the AI output does not match (or matches but is not adopted) as the cause and countermeasure of this failure, but is correct as the inference result. "BAD" is selected when the AI output is incorrect as a candidate for the cause of the failure of the target device.

[0048] FIG. 7 shows an example of the work record information 700 stored in the work record database 131. The work record information 700 records the data ID, work date and time, device model number of the work target, device ID which is the unique ID of the work target device, failure situation, failure location, cause of failure, location of the cause of failure, countermeasures implemented as work, countermeasure location where the countermeasures were implemented, person in charge of the work, and the level of the person in charge of the work.

[0049] Figure 8 shows an example of proposal record information 800 recorded in the proposal record database 132. The proposal record information 800 shown in Figure 8 is obtained by adding the results of evaluation by an operator via the user interface (Figure 6) to the output result (estimated cause and countermeasure information 113Ci) of the estimated cause and countermeasure generation unit 124. The proposal record 221 consists of a record ID (#), date and time, model number of the equipment to be proposed, individual equipment ID to be proposed, failure situation in the proposal, failure location, estimated cause of failure, estimated cause location, estimated countermeasure, estimated countermeasure location, name of the operator who evaluated the proposal result, operator level, and evaluation result.

[0050] Figure 9 shows a specific example 900 of the learning data selection information 113Ao input to the learning data selection unit 123. In this example, the learning data selection information 900 records the selection range of the date and time specified in the learning data selection field 616, the selection range of the equipment ID, the selection range of the operator level, and whether or not past AI suggestions were selected. In this embodiment, it is shown that past AI suggestions are not adopted.

[0051] Figure 10 shows a flowchart of an example of the processing of the work record selection unit 125. The work record selection unit 125 selects work records to be used for Few-Shot learning of the generated AI based on the work record database 131 and the learning data selection information.

[0052] The process starts from step S1001. In step S1002, the record item ID to be read from the work record is set to 1. In step S1003, the setting date and time recorded in the learning data selection information 113Ao(900) is checked to see if the work record is within that range. If it is within the range of the setting date and time, the process proceeds to step S1004; otherwise, the process proceeds to step S1007. In step S1004, the setting device ID recorded in the learning data selection information 113Ao(900) is checked to see if the work record matches that setting device ID. If it matches the setting device ID, the process proceeds to step S1005; otherwise, the process proceeds to step S1007.

[0053] In step S31005, the worker level recorded in the learning data selection information 113Ao(900) is checked, and it is confirmed whether the work record is within that range. If it is within the set worker level range, the process proceeds to step S1006; otherwise, it proceeds to step S1007. In step S1006, the work record is added to the learning target work record. In step S1007, it is determined whether the current record item ID is the last item. If it is the last item, the process proceeds to step S1009; otherwise, the process proceeds to step S1008. In step S31008, 1 is added to the record item ID (+1), and the process returns to step S1003, and the subsequent processing is repeated. In step S1009, the process ends.

[0054] Figure 11 shows a flowchart of the selection process for proposal record information (132o(800)) in the proposal history selection unit 126. Based on the proposal record database 132 and the training data selection information, the proposal history selection unit 126 selects proposal record information to be used for few-shot training in the generating AI (large-scale language model 300).

[0055] The process starts from step S1101. In step S1102, it is checked whether the AI ​​proposal recorded in the learning data selection information 113Ao(900) is used. If the AI ​​proposal record is a target for learning, the process proceeds to step S1103; otherwise, the process proceeds to step S1110. As shown in Figure 9, this embodiment assumes that past AI proposals are not used as learning targets. However, as an explanation of the overall processing flow of the proposal history selection unit 126, the case where "YES" is used in step S1102 (when past AI proposals are used as learning targets) is also explained here. In step S1104, the set date and time recorded in the learning data selection information 113Ao(900) is checked to see if the proposal record is within that range. If it is within the set date and time range, the process proceeds to step S1105; otherwise, the process proceeds to step S1108.

[0056] In step S1105, the configured device ID recorded in the learning data selection information 113Ao(900) is checked, and it is confirmed whether the proposed record matches that configured device ID. If it matches the configured device ID, the process proceeds to step S1106; otherwise, the process proceeds to step S1108. In step S1106, the worker level recorded in the learning data selection information 113Ao(900) is checked, and it is confirmed whether the proposed record falls within that range. If it falls within the set worker level range, the process proceeds to step S1107; otherwise, the process proceeds to step S1108. In step S1107, the proposed record is added to the learning target proposed record. In step S1108, it is determined whether the current record item ID is the last item. If it is the last item, the process proceeds to step S1110; otherwise, the process proceeds to step S1109. In step S1109, 1 is added to the record item ID (+1), and the process returns to step S1104, and the subsequent processing is repeated. In step S1110, the process ends.

[0057] Figure 12 shows an example 1200 of the output result 131o' of the work record selection unit 125 in this embodiment. In this example, for the work record 700 shown in Figure 7, the data with "Data ID = 2" was not selected as a result of the selection based on the learning data selection information 113Ao(900), and the data with "Data ID = 1" and "Data ID = 3" were selected.

[0058] Figure 13 shows the processing flowchart in the estimated cause and proposed countermeasures generation unit 124.

[0059] The process starts from step S1301. In step S1302, the failure status information 113Bo is read. In step S1303, the work record 131o' selected by the work record selection unit 125 is read. In step S1304, the output of the proposal record 132o' selected by the proposal record selection unit 126 is read.

[0060] In step S1305, an instruction to generate an estimated cause / countermeasure report is generated based on the failure status information 113Bo, the work record selection result 131o', and the proposed record selection result 132o', and sent to the large-scale language model 300. In step S1306, the generation result of the estimated cause / countermeasure report is received from the large-scale language model 300. In step S1307, the process ends.

[0061] Figure 14 shows an example of the details of the estimated cause and proposed countermeasure report generation instruction 1400 that is sent to the large-scale language model 300 in step S1305. The estimated cause and proposed countermeasure report generation instruction 1400 consists of an instruction prompt 1400A for the large-scale language model unit 300, a newly occurring failure situation 1400B, past work records 1400C learned in Few-Shot learning, and an output format 1400D.

[0062] Figure 15 shows an example of the details of the estimated cause and proposed countermeasures report generated 1500, which is received from the large-scale language model 300 in step S1206.

[0063] As explained above, this embodiment makes it possible to narrow down the work records stored in the work record database according to the date and time the work was performed and use them for AI learning, thereby improving the accuracy of the failure cause and countermeasures estimated by the AI. In other words, in facilities such as factories, when a failure occurs, measures to prevent recurrence are sometimes taken. If reasoning is based on work records before the recurrence prevention measures were taken, there is a possibility of generating incorrect cause estimations and countermeasure proposals. However, by narrowing down the date and time of the work records used for AI learning, appropriate work records can be used for AI learning.

[0064] Furthermore, according to this embodiment, it becomes possible to use work records limited to the target equipment from the work record database for AI learning, thereby improving the accuracy of the failure causes and countermeasures estimated by the AI. In other words, in facilities such as factories, there are failures that are likely to occur in individual equipment depending on the country in which the equipment is installed, the installation and operating conditions, and individual differences. By narrowing down the equipment IDs used for AI learning, appropriate work records can be used for AI learning.

[0065] Furthermore, according to this embodiment, it becomes possible to use work records selected from the work record database according to the skill level of the worker for AI learning, thereby improving the accuracy of the failure causes and countermeasures estimated by the AI. In other words, in workplaces such as factories, there are differences in the skill levels of workers, and there is a possibility that incorrect work content or judgments may be recorded in the work records of less skilled workers. By narrowing down the skill levels of workers used for AI learning to those with a certain level of skill or higher, appropriate work records can be used for AI learning.

[0066] In the above-described Example 1, we explained a case where the training data used for the few-shot training of the large-scale language model 300 did not include past AI suggestions. In this example, however, we will explain a case where past AI suggestions are included. The configuration of each part and the overall processing flow are the same as in Example 1, so in the following explanation, we will assume the configuration of each part and the processing flow explained in Example 1, and will focus on the parts that differ from Example 1. Also, explanations of parts that overlap with Example 1 will be omitted.

[0067] Figure 16 shows the details of the user interface 1600 of the fault analysis input / output unit 113 in this embodiment. Since each item in this figure is the same as the user interface 600 shown in Figure 6 of Embodiment 1, no explanation is provided. In the user interface 1600 in Figure 16, the learning data selection unit 1616 shows the state in which past AI suggestions have been selected as the learning target.

[0068] Figure 17 shows a specific example 1700 of the learning data selection information 113Ao input to the learning data selection unit 123. The specific example 1700 of the learning data selection information 113Ao records the selection range of the date and time specified in the learning data selection field 1616, the selection range of the equipment ID, the selection range of the worker level, and whether or not past AI suggestions have been selected. In this embodiment, the case in which past AI suggestions are adopted is shown.

[0069] Figure 18 shows an example 1800 of the output result 131o' of the work record selection unit 125 in this embodiment. In this example, for the work record 700 shown in Figure 7, the data with "Data ID = 2" and "Data ID = 3" were not selected as a result of the selection based on the learning data selection information 113Ao (1700), and only the data with "Data ID = 1" was selected.

[0070] Figure 19 shows an example 1900 of the output result 132o' of the proposal record selection unit 126 in this embodiment. In this embodiment, past AI proposals are used as learning targets, so "YES" is determined in step S1102 of the processing flowchart in Figure 11, and the processing from step S1103 onwards is executed, resulting in the proposal record information 800 shown in Figure 8 being selected based on the learning data selection information 113Ao (1700).

[0071] Figure 20 shows the details of the estimated cause / countermeasure report generation instruction 2000 that is sent to the large-scale language model 300 in step S1305 of Figure 13 in this embodiment. The estimated cause / countermeasure report generation instruction 2000 consists of an instruction prompt 2000A for the large-scale language model 300, a newly occurring failure situation 2000B, past work records 2000C learned in Few-Shot learning, past proposal records 2000D, and an output format 200E. It differs from the estimated cause / countermeasure report generation instruction 1400 of Embodiment 1 shown in Figure 14 in that past proposal records 2000D are added.

[0072] As explained above, this embodiment makes it possible to narrow down the proposal records stored in the proposal record database according to the date and time the proposal was made, the ID of the equipment to which the proposal was made, and the level of the worker who performed the evaluation, and use them for AI learning, thereby improving the accuracy of the failure cause and countermeasures estimated by the AI. In other words, in equipment such as factories, if the equipment has a low failure frequency, there is a possibility that there will be few work records that can be used for AI learning. On the other hand, if all past proposal records are used as cache information, there is a problem that incorrect proposal results will be repeated and the quality of the proposals will decline. In this embodiment, by using proposals that have been evaluated as excellent from past proposal records for AI learning, it is possible to use appropriate proposal records for AI learning as inference results.

[0073] Furthermore, according to this embodiment, it becomes possible to use work records and suggestion records, whose number has been reduced according to user-defined conditions, for AI learning from the work record database and suggestion record database, thereby reducing the cost of AI learning and inference, and shortening the response time. In other words, when using AI learning such as Few-Shot learning or Retrievable-Augmented Generation (RAG), computation costs are required according to the amount of information used for AI learning, resulting in high service usage fees. Also, as the amount of AI learning (number of examples in Few-Shot learning) increases, the time required for inference increases accordingly. In this embodiment, by reducing the work records and suggestion records used for AI learning according to user-defined conditions, it becomes possible to balance the quality of AI-based fault cause analysis with the inference cost (expense and time).

[0074] 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 described in detail to make the present invention easier to understand, and are not necessarily limited to those having all the configurations described. Furthermore, it is possible to replace parts of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add configurations from other embodiments to the configuration of one embodiment. In addition, it is possible to add, delete, or replace parts of the configuration of each embodiment with other configurations.

[0075] Furthermore, each of the above configurations, functions, processing units, and processing means may be implemented in hardware, either partially or entirely, for example, by designing them as integrated circuits. Information such as programs, tables, and files that implement each function can be stored in the processor's internal memory, external flash memory, or SSD (Solid State Drive) storage devices. Additionally, each of the above configurations and functions can be configured on a PC, an on-premise server, or a cloud server.

[0076] 100: Failure Cause Analysis Device 110: Main Memory 111: Input / Output Unit 112: Work Record Input / Output Unit 113: Failure Analysis Input / Output Unit 120: Failure Analysis Unit 121: Work Record Unit 122: Cause / Countermeasure Estimation Unit 123: Learning Data Selection Unit 124: Estimated Cause / Countermeasure Proposal Generation Unit 125: Work Record Selection Unit 126: Proposal History Selection Unit 130: Auxiliary Storage Device 131: Work Record Information DB 132: Proposal Record DB 140: Arithmetic Processing Unit 150: Input / Output Interface 160: Communication Interface 200: Input / Output Device 300: Large-Scale Language Model 400: Communication Network

Claims

1. A failure cause analysis device for analyzing the cause of failure and countermeasures for a target system, comprising: a work record storage unit that stores work records including the occurrence of a failure event, the cause of the failure, and information on countermeasures for the failure event; a work record selection unit that selects work records to be used for training a generating AI from the work records stored in the work record storage unit based on predetermined conditions; and an estimation unit that, in response to a newly occurring failure event, trains the generating AI based on the work records selected by the work record selection unit, and estimates the cause of the failure and countermeasures for the newly occurring failure event using the trained generating AI.

2. A fault cause analysis device according to claim 1, wherein the work record selection unit selects a work record to be used for training the generating AI based on at least one of the following items set by the user: work date and time, individual identification number of the equipment to be worked on, and worker's skill level.

3. A fault cause analysis apparatus according to claim 1, comprising: a proposal record evaluation input unit that accepts evaluations by a user for proposed records of fault causes and countermeasures estimated by the estimation unit; and a proposal record storage unit that stores the evaluated proposal records based on the evaluation results by the user.

4. A fault cause analysis apparatus according to claim 3, comprising a proposal record selection unit that selects proposal records to be used for learning the generating AI from proposal records stored in the proposal record storage unit based on predetermined conditions, and the estimation unit performs learning of the generating AI based on the proposal records selected by the proposal record selection unit.

5. A fault cause analysis device according to claim 4, wherein the proposal record selection unit selects a proposal record to be used for training the generated AI based on at least one of the following items set by the user: whether or not to use AI proposals, the date and time of the proposal, the individual identification number of the proposed equipment, and the skill level of the evaluator.

6. A fault cause analysis apparatus according to claim 3, comprising a work record input unit that receives work records from the user, and a proposed record evaluation input unit that transmits the proposed record adopted by the user as an actual work to the work record input unit as a work record.

7. A failure cause analysis method in a failure cause analysis device having a work record storage unit, a work record selection unit, and an estimation unit, for analyzing the cause of failure and countermeasures for a target system, characterized in that: the work record storage unit stores work records including the occurrence of a failure event, the cause of the failure, and information on countermeasures for the failure event; the work record selection unit selects work records to be used for training a generating AI from the work records stored in the work record storage unit based on predetermined conditions; and the estimation unit trains the generating AI based on the work records selected by the work record selection unit for newly occurring failure events, and estimates the cause of the failure and countermeasures for the newly occurring failure event using the trained generating AI.

Images