Intelligent predictive-diagnosis device and method based on ultra-large language model

Abstract

The present invention provides an intelligent predictive-diagnosis system based on an ultra-large language model. The intelligent predictive-diagnosis device based on an ultra-large language model, according to the present invention, comprises: a preprocessing unit for extracting, upon occurrence of an early warning corresponding to closed-type industrial equipment, tag information related to the early warning, then converting the extracted tag information into a predetermined standard signal name, converting, on the basis of signal data included in the tag information, a result of analyzing a data trend into text information, and generating predictive-diagnosis request information including the converted standard signal name and text information; a first processing unit for outputting warning cause analysis information as a result of inquiring and analyzing, by means of the predictive-diagnosis request information, a first predictive-diagnosis database pre-stored in correspondence to the closed-type industrial equipment; and a second processing unit for outputting a predictive diagnosis for the occurrence of the early warning and one or more pieces of equipment failure action information corresponding thereto by inputting the predictive-diagnosis request information and / or the warning cause analysis information into a diagnosis prediction module based on the ultra-large language model.

Description

Intelligent predictive diagnosis device and method based on a super-large language model

[0001] The present invention relates to an intelligent predictive diagnosis device and method based on a large-scale language model, and more particularly, to an intelligent predictive diagnosis device and method based on a large-scale language model applied to national and public facility industries.

[0002] In the national and public utility industries (e.g., nuclear power plants), the requirements of these industries are evolving in line with the advancement of AI technology, and the level of technology is advancing to reflect this.

[0003] At present, intelligent predictive diagnostic systems being applied or developed in the national and public facility industries (e.g., nuclear power plants) are still at the stage of ‘predicting and predicting facility failures.’

[0004] In addition, when an alarm is generated through an intelligent predictive diagnosis system, a specialized analyst with domain knowledge of the relevant facility must analyze the cause of the alarm by comparing it with data trends, related system signal data, or past data to analyze changes in data trends, which takes a lot of time to secure action time after the alarm is generated.

[0005] In addition to trend analysis using domain knowledge, the cause of the alarm occurrence is analyzed by reviewing the facility's system diagram, procedure documents, etc., and a detailed inspection is requested from the plant site to identify the actual cause and take action. Therefore, the process of analyzing the cause of the alarm occurrence requires analysis based on a lot of data and materials, which also causes the problem of time consumption.

[0006] And, there are various online-based LM models currently in use, such as GPT, Bard, LLaMa, and Vicuna, and in order to utilize the models, you can get answers by utilizing APIs in an environment where the Internet is available, but in the case of data that is not common, such as nuclear power plants, there is a problem that the model outputs incorrect answers, and although learning must be done based on the procedures, instructions, and documents of the actual facility, in the case of national infrastructure 'A' grade, there is a problem that it is difficult to export the relevant data, making it difficult to further advance the intelligent prediction diagnosis system.

[0007] This is a time when a solution that reflects this is needed.

[0008] Accordingly, the present invention was created to solve the above problems, and the purpose of the present invention is to provide an intelligent predictive diagnosis system based on a super-large language model that can be applied to national and public facility industries and even suggests countermeasures for facility failures.

[0009] The purpose of the invention is not limited to the purposes mentioned above, and other purposes not mentioned will be clearly understood by those skilled in the art from the description below.

[0010] In order to achieve the above object, an intelligent predictive diagnosis device based on a super-large language model according to a first aspect of the present invention includes a preprocessing unit which extracts tag information related to an early warning when an early warning corresponding to a closed industrial facility occurs, converts the extracted tag information into a predetermined standard signal name, analyzes data trends based on signal data included in the tag information, converts the results into text information, and generates predictive diagnosis request information including the converted standard signal name and the text information, a first processing unit which searches and analyzes a first predictive diagnosis database stored in advance corresponding to the closed industrial facility with the predictive diagnosis request information and outputs alarm cause analysis information as a result, and a second processing unit which inputs at least one of the predictive diagnosis request information and the alarm cause analysis information into a diagnosis prediction module based on a super-large language model, thereby outputting a predictive diagnosis for the occurrence of the early warning and at least one piece of equipment failure response information corresponding thereto.

[0011] The second processing unit can update and store the prediction diagnosis for the occurrence of the early warning and the one or more equipment failure response information in the second prediction diagnosis database included in the diagnosis prediction module.

[0012] The above second predictive diagnosis database is managed by the closed industrial facility or by industrial sector including the closed industrial facility, and when managed by industrial sector, it can be used for intelligent predictive diagnosis of other closed industrial facilities similar to the closed industrial facility.

[0013] The second processing unit may perform processing including encryption or a similar processing method to block information that is prohibited from being exposed to the public and to identify the closed industrial facility through one or more pieces of information from among the prediction diagnosis request information and the alarm cause analysis information input to the diagnosis prediction module.

[0014] The above diagnostic prediction module may be provided in an external system and linked to the second processing unit, or may be provided within the second processing unit.

[0015] And, according to a second aspect of the present invention for achieving the above object, an intelligent predictive diagnosis method based on a super-large language model may include a preprocessing step of extracting tag information related to an early warning when an early warning corresponding to a closed industrial facility occurs in an intelligent predictive diagnosis device, converting the extracted tag information into a predetermined standard signal name, analyzing a data trend based on signal data included in the tag information, converting the result into text information, and generating predictive diagnosis request information including the converted standard signal name and the text information, a first processing step of searching and analyzing a first predictive diagnosis database stored in advance corresponding to the closed industrial facility with the predictive diagnosis request information, and outputting cause analysis information as a result, and a second processing step of inputting at least one of the predictive diagnosis request information and the cause analysis information into a diagnosis prediction module based on a super-large language model to output a predictive diagnosis for the occurrence of the early warning and at least one piece of equipment failure response information corresponding thereto.

[0016] The second processing step may include a step of updating and storing the prediction diagnosis for the occurrence of the early warning and the one or more equipment failure response information in a second prediction diagnosis database included in the diagnosis prediction module.

[0017] The above second predictive diagnosis database is managed by the closed industrial facility or by industrial sector including the closed industrial facility, and when managed by industrial sector, it can be used for intelligent predictive diagnosis of other closed industrial facilities similar to the closed industrial facility.

[0018] The second processing step may include a processing step including encryption or a processing method equivalent thereto for blocking the identification of the closed industrial facility and the prohibited external exposure information related to the closed industrial facility through one or more pieces of information from the prediction diagnosis request information and the alarm cause analysis information input to the diagnosis prediction module.

[0019] The second processing step may include a step of linking with the diagnosis prediction module provided in an external system or calling the diagnosis prediction module provided in the intelligent prediction diagnosis device.

[0020] Accordingly, the present invention not only provides information on prediction and forecasting of facility failure when an alarm is generated due to facility failure for closed industrial facilities such as national and public facility industries (e.g., nuclear power plants), but also provides one or more pieces of facility failure response information that can be used to immediately take action against the facility failure, thereby significantly reducing the time taken to take actual action after the occurrence of an early warning, and in cases where there is no learning data to provide as result data for an early warning, there is an advantage in that one or more pieces of facility failure response information can be provided by automatically learning documents and data with high similarity.

[0021] The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description of the claims.

[0022] FIG. 1 is a schematic diagram showing an intelligent predictive diagnosis device based on a super-large language model according to one embodiment of the present invention.

[0023] Fig. 2 is a configuration diagram showing an example of a database structure provided in the intelligent predictive diagnosis device of Fig. 1.

[0024] Figure 3 is a diagram showing an example of learning and application expansion based on the database structure of Figure 2.

[0025] Fig. 4 is an exemplary diagram showing an example of the process of generating equipment failure response information through the intelligent predictive diagnosis device of Fig. 1.

[0026] Figure 5 is a block diagram showing the process of generating analysis information as an alarm source through the intelligent predictive diagnosis device of Figure 1.

[0027] Figure 6 is a block diagram showing the process of generating equipment failure response information through the intelligent predictive diagnosis device of Figure 1.

[0028] Figure 7 is an example diagram showing the results of the time series data analysis of Figure 6.

[0029] Figure 8 is an example diagram showing an example of a target for text data analysis of Figure 6.

[0030] Figure 9 is an example diagram including a process in which expert feedback related to a closed industrial facility is reflected in the process of learning diagnostic prediction data based on the ultra-large language model of the present invention.

[0031] Figure 10 is a block diagram showing an example of a process for obtaining diagnostic prediction data based on the ultra-large language model of the present invention.

[0032] Figure 11 is a block diagram showing an example of a process for generating result data for an unlearned request of the present invention.

[0033] Figure 12 is an example diagram showing a method for learning technical documents related to the closed industrial facility field.

[0034] And, FIG. 13 is a flowchart showing an intelligent prediction diagnosis method based on a super-large language model according to one embodiment of the present invention.

[0035] The advantages and features of the present invention, and the methods for achieving them, will become clearer with reference to the embodiments described in detail below together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various different forms. These embodiments are provided only to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Like reference numerals designate like elements throughout the specification.

[0036] Furthermore, the embodiments described herein will be described with reference to cross-sectional and / or schematic drawings, which are ideal illustrations of the present invention. Therefore, the form of the illustrations may be modified due to manufacturing techniques and / or tolerances. Furthermore, in each drawing illustrated in the present invention, each component may be depicted somewhat enlarged or reduced for convenience of explanation.

[0037] The intelligent predictive diagnosis device based on the ultra-large language model of the present invention is configured to provide information on equipment failure measures so that rapid action can be taken after an early warning is generated.

[0038] FIG. 1 is a schematic diagram showing an intelligent predictive diagnosis device based on a super-large language model according to one embodiment of the present invention.

[0039] As illustrated in FIG. 1, an intelligent predictive diagnosis device (100) based on a large-scale language model includes a preprocessing unit (110) which extracts tag information related to an early warning when an early warning corresponding to a closed industrial facility (300) occurs, converts the extracted tag information into a predetermined standard signal name, analyzes data trends based on signal data included in the tag information, converts the results into text information, and generates predictive diagnosis request information including the converted standard signal name and text information, a first processing unit (120) which searches and analyzes a first predictive diagnosis database stored in advance corresponding to the closed industrial facility (300) for the predictive diagnosis request information and outputs alarm cause analysis information as a result, and a second processing unit (130) which inputs at least one of the predictive diagnosis request information and the alarm cause analysis information into a diagnosis prediction module (200) based on a large-scale language model, thereby outputting a predictive diagnosis for the occurrence of an early warning and at least one piece of equipment failure response information corresponding thereto.

[0040] Here, the closed industrial facility (300) refers to an industrial facility in the national and public facility industry (e.g., nuclear power plant), and may also include facilities in other industrial fields that require such closed infrastructure.

[0041] Conventional intelligent predictive diagnostic devices are devices that notify of abnormal signs in the measurement signals of industrial facilities. They can provide information on the measurement signals of industrial facilities (e.g., temperature rise, pressure rise, etc.), but the cause of the abnormal signs required a considerable amount of time to be analyzed by experts in the relevant field, which resulted in a considerable amount of time required before actual action could be taken.

[0042] However, the intelligent predictive diagnosis device (100) based on the ultra-large language model of the present invention described above informs experts in the relevant field of the cause of abnormal signs in industrial facilities, thereby enabling significantly faster action since the expert's analysis process can be omitted based on the data.

[0043] Here, a "super-large language model" refers to a model with tens to hundreds of billions of parameters, distinguishing it from a "huge language model" (a model with hundreds of millions to billions of parameters). For example, GPT-3, with 175 billion parameters, and GPT-4, with even more parameters, are examples of super-large language models, while GPT-2, with 1.5 billion parameters, is considered a large language model.

[0044] The present invention relates to industrial facilities mainly applied to public areas such as nuclear power plants, and in order to generate precise facility failure response information, it is preferable to use an ultra-large language model that is capable of understanding the context of technical documents, complex queries, and responses thereto, and is capable of zero-shot learning, rather than a large language model.

[0045] The equipment failure response information generated by the intelligent predictive diagnosis device (100) based on the ultra-large language model of the present invention may be provided as a single piece of information, but preferably, it may be provided as multiple pieces of equipment failure response information in order of priority. This provides an efficient system based on AI analysis for the response process for equipment showing abnormal signs. However, it is not easy to provide 100% error-free and perfect equipment failure response information with current technology. Therefore, by providing multiple pieces of equipment failure response information in order of priority as described above, it is possible to process early warnings in a safe and efficient manner.

[0046] The second processing unit (130) can update and store the prediction diagnosis for the occurrence of an early warning and one or more pieces of equipment failure response information in the second prediction diagnosis database included in the diagnosis prediction module, thereby additionally collecting AI learning data to continuously update the accuracy of the equipment failure response information.

[0047] Fig. 2 is a configuration diagram showing an example of a database structure provided in the intelligent predictive diagnosis device of Fig. 1.

[0048] As illustrated in FIG. 2, a first predictive diagnosis database (121) corresponding to the closed industrial facility (300) and a second predictive diagnosis database (131) separately provided for the purpose of accumulating and storing the results of predictive diagnosis and facility failure response information and utilizing them as AI learning data can be distinguished.

[0049] In addition, the second predictive diagnosis database (131) can be managed by closed industrial facility (300) or by industrial field including closed industrial facility (300), and when managed by industrial field, it can also be applied to predictive diagnosis of other closed industrial facilities similar to the closed industrial facility (300).

[0050] In other words, rather than collecting AI learning data for each closed industrial facility and learning from it in a limited manner, it may be more beneficial to improve the performance of intelligent predictive diagnostic devices by learning from various cases created in multiple closed industrial facilities by industrial sector, the corresponding action information, and the accuracy of the provided action information.

[0051] To this end, the second processing unit (130) can execute processing that includes encryption or a processing method equivalent thereto to block information that cannot be specified for the closed industrial facility (300) and is prohibited from being exposed to the public related to the closed industrial facility (300) through one or more pieces of information from among the prediction diagnosis request information and the alarm cause analysis information input to the diagnosis prediction module, and it is preferable to execute the processing above in consideration of the industrial structure in which the closed industrial facility (300) is operated.

[0052] Figure 3 is a diagram showing an example of learning and application expansion based on the database structure of Figure 2.

[0053] As illustrated in Figure 3, a first predictive diagnosis database (121) corresponding to each closed industrial facility can be provided, and as described above, a second predictive diagnosis database (131) that collects AI learning data by integrating these can be provided in an integrated manner.

[0054] Among the data updated in the second predictive diagnosis database (131), data matching the analysis information that is the cause of the alarm can be provided to each first predictive diagnosis database, thereby verifying and supplementing the data in the first predictive diagnosis database (121).

[0055] In addition, when constructing a new 1-N predictive diagnosis database (121-4) corresponding to the Nth closed industrial facility (340), it can also be utilized to construct the 1-N predictive diagnosis database (121-4) with efficient and verified data based on the 2nd predictive diagnosis database (131).

[0056] Fig. 4 is an exemplary diagram showing an example of the process of generating equipment failure response information through the intelligent predictive diagnosis device of Fig. 1.

[0057] As illustrated in FIG. 4, when an early warning corresponding to a closed industrial facility (300) occurs, the preprocessing unit (110) extracts tag information related to the early warning, such as 'TagName: FWT0011', and then converts the extracted tag information into a predetermined standard signal name, and analyzes data trends based on signal data, which is time series data included in the tag information, and converts the analyzed results into text.

[0058] The first processing unit (120) receives the standard signal name converted from the preprocessing unit (110) as 'inlet temperature of main water pump C' and the text converted as '60 degrees lower than inlet temperature of main water pump B' as predictive diagnosis request information, and can combine and reconstruct this information to output it as alarm cause analysis information.

[0059] Thereafter, the second processing unit (130) inputs the information provided from the first processing unit (120) into a diagnosis prediction module based on a super-large language model, thereby generating a prediction diagnosis corresponding to the occurrence of an early warning and one or more pieces of equipment failure response information corresponding thereto.

[0060] In addition, when an early warning occurs and the facility operator (user) inquires about the cause of the early warning to the intelligent predictive diagnosis device (100), the facility failure response information is output directly on the screen of the facility operator (user)'s user device without requiring an analysis process by the facility operator (user) or an expert related to the facility, thereby enabling the facility operator (user) to immediately take action on a specific facility configuration.

[0061] Figure 5 is a block diagram showing the process of generating analysis information as an alarm source through the intelligent predictive diagnosis device of Figure 1.

[0062] As illustrated in FIG. 5, when a request for identifying the cause of an early warning is received, the intelligent predictive diagnosis device (100) can generate alarm cause analysis information by analyzing signal data and system-related information corresponding to the closed industrial facility (300) by querying the first predictive diagnosis database (121) prepared in advance and outputting the same.

[0063] At this time, the alarm cause analysis information is information that has not passed through the diagnosis prediction module (200) based on the ultra-large language model, and may be notified to the facility operator (user) of the closed industrial facility (300) through an expert judgment process on the part of the service provider operating the intelligent prediction diagnosis device (100).

[0064] Figure 6 is a block diagram showing the process of generating equipment failure response information through the intelligent predictive diagnosis device of Figure 1.

[0065] As illustrated in FIG. 6, the intelligent predictive diagnosis device (100) can operate a diagnosis prediction module (200) based on a super-large language model based on predictive diagnosis request information to provide one or more pieces of equipment failure response information including the cause of an abnormality, ripple effects due to an abnormality or failure, and response measures.

[0066] To this end, the second processing unit (120) analyzes the time series data trend of the signal data analysis results of the closed industrial facility (300), performs qualitative analysis to generate results for predictive diagnosis request information through text data analysis of the system-related information (e.g., drawings, procedure documents, etc.) of the closed industrial facility (300), and can generate one or more pieces of facility failure response information as a result.

[0067] Fig. 7 is an example diagram showing an example of the results for the time series data analysis of Fig. 6, and Fig. 8 is an example diagram showing an example of the target for the text data analysis of Fig. 6.

[0068] As illustrated in FIG. 7, the second processing unit (120) can convert data on time-series trends into text data through time-series variation analysis of each signal data, and can additionally provide the converted text data in addition to the text data on system-related information for the qualitative analysis described above.

[0069] Figure 9 is an example diagram including a process in which expert feedback related to a closed industrial facility is reflected in the process of learning diagnostic prediction data based on the ultra-large language model of the present invention.

[0070] As illustrated in FIG. 9, in order to obtain predictive diagnosis data, technical documents including procedures, drawings, and instructions corresponding to the industrial facility (300) are secured in a closed structure within the system, data preprocessing and labeling are performed, and then the data is converted into data through a data learning process, and a question / answer process is repeatedly performed to reflect expert feedback related to the industrial facility (300), thereby supplementing the data learning environment based on local resources without being connected to the limited technical documents and online, thereby enabling accurate and specific data collection and learning.

[0071] FIG. 10 is a block diagram showing an example of a process for preparing diagnostic prediction data based on the ultra-large language model of the present invention, and FIG. 11 is a block diagram showing an example of a process for generating result data for an unlearned request of the present invention.

[0072] As shown in FIGS. 10 and 11, the technical documents including instructions, procedures, logic, and drawings related to the industrial facility (300) are digitized through data preprocessing and labeling, and the diagnostic prediction data is constructed through an embedding model that goes through the APR-1400 LDM document collation process through text chunking, and then the diagnostic prediction data is digitized and stored in a vector store.

[0073] In addition, the intelligent prediction diagnosis device (100) of the present invention searches for highly relevant technical documents stored by chunking with the E5-Large model in a situation where it is necessary to derive results that have not been learned in advance, and can also process responses to diagnosis prediction and result data that are not inherent in the E5-Large model.

[0074] Figure 12 is an example diagram showing a method for learning technical documents related to the closed industrial facility field.

[0075] As illustrated in FIG. 12, the intelligent predictive diagnosis device (100) can learn the corresponding information in the RAG (retrieval-augmented generation) database to perform learning within the LLM model based on the HTML Parsing program based on the classified unstructured data.

[0076] And, FIG. 13 is a flowchart showing an intelligent prediction diagnosis method based on a super-large language model according to one embodiment of the present invention.

[0077] As illustrated in FIG. 13, the intelligent predictive diagnosis method based on the ultra-large language model of the present invention extracts tag information related to the early warning when an early warning corresponding to a closed industrial facility (300) occurs in an intelligent predictive diagnosis device (100), converts the extracted tag information into a predetermined standard signal name, analyzes the data trend based on the signal data included in the tag information, converts the result into text information, and generates predictive diagnosis request information including the converted standard signal name and text information (S100).

[0078] At step S100, a first processing step is performed to output alarm cause analysis information as a result of searching and analyzing the first predictive diagnosis database (121) stored in advance in correspondence with the closed industrial facility (300) using the generated predictive diagnosis request information (S102).

[0079] Thereafter, by inputting at least one of the predictive diagnosis request information and the alarm cause analysis information into the diagnosis prediction module (200) based on the ultra-large language model, a second processing step is performed to output a predictive diagnosis for the occurrence of an early warning and at least one piece of equipment failure response information corresponding thereto (S104).

[0080] And, if the process up to step S104 is completed to take measures for early warning, the process can be terminated thereafter, and if the early warning cannot be resolved even after taking measures through one or more pieces of equipment failure action information, the process of generating and providing one or more pieces of equipment failure action information that have been supplemented through steps S100 to S104 is performed again (S106).

[0081] The detailed step description for each step described above is based on the previously mentioned Figures 1 to 12 and the detailed descriptions therefor.

[0082] Although the embodiments of the present invention have been described with reference to the above and the attached drawings, those skilled in the art will understand that the present invention can be implemented in other specific forms without altering the technical concept or essential characteristics thereof. Therefore, the embodiments described above should be understood to be illustrative in all respects and not restrictive.

[0083] In addition, the present invention is applicable to the national and public facility industries, and is intended to provide an intelligent predictive diagnosis system based on a super-large language model that even suggests countermeasures for facility failures. Therefore, the invention has sufficient potential for commercialization or sales, and is thus an invention with industrial applicability as it can be clearly implemented in reality.

Claims

1. A preprocessing unit that extracts tag information related to the early warning when an early warning corresponding to a closed industrial facility occurs, converts the extracted tag information into a predetermined standard signal name, analyzes data trends based on signal data included in the tag information, converts the results into text information, and generates predictive diagnosis request information including the converted standard signal name and the text information; A first processing unit that outputs analysis information, which is the cause of the alarm, as a result of searching and analyzing the first predictive diagnosis database stored in advance in correspondence with the closed industrial facility based on the predictive diagnosis request information; and An intelligent predictive diagnosis device based on a large-scale language model, comprising a second processing unit that outputs a predictive diagnosis for the occurrence of the early warning and one or more pieces of equipment failure response information corresponding thereto by inputting at least one of the above predictive diagnosis request information and the above alarm cause analysis information into a diagnosis prediction module based on a large-scale language model.

2. In paragraph 1, The second processing unit is an intelligent predictive diagnosis device based on a large-scale language model that updates and stores predictive diagnosis for the occurrence of the early warning and the information on the one or more equipment failure measures in a second predictive diagnosis database included in the diagnosis prediction module.

3. In paragraph 2, The above second predictive diagnosis database is managed by the closed industrial facility or by industrial sector including the closed industrial facility, and when managed by industrial sector, is an intelligent predictive diagnosis device based on a large-scale language model that can be used for intelligent predictive diagnosis of other closed industrial facilities similar to the closed industrial facility.

4. In paragraph 1 or paragraph 3, The second processing unit is an intelligent predictive diagnosis device based on a large-scale language model that processes information including encryption or a processing method equivalent thereto to prevent the identification of the closed industrial facility and block external exposure prohibited information related to the closed industrial facility through at least one of the predictive diagnosis request information and the alarm cause analysis information input to the diagnosis prediction module.

5. In paragraph 1 or 2, The above diagnosis prediction module is an intelligent prediction diagnosis device based on a large-scale language model that is installed in an external system and linked to the second processing unit, or installed in the second processing unit.

6. In an intelligent predictive diagnosis device, when an early warning corresponding to a closed industrial facility occurs, a preprocessing step of extracting tag information related to the early warning, converting the extracted tag information into a predetermined standard signal name, analyzing the data trend based on the signal data included in the tag information, and converting the result into text information, and generating predictive diagnosis request information including the converted standard signal name and the text information; A first processing step of outputting analysis information, which is the cause of the alarm, as a result of searching and analyzing the first predictive diagnosis database stored in advance in correspondence with the closed industrial facility based on the predictive diagnosis request information; and An intelligent predictive diagnosis method based on a large-scale language model, comprising a second processing step of inputting at least one of the above predictive diagnosis request information and the above alarm cause analysis information into a diagnosis prediction module based on a large-scale language model, thereby outputting a predictive diagnosis for the occurrence of the above early warning and at least one piece of equipment failure response information corresponding thereto.

7. In paragraph 6, An intelligent predictive diagnosis method based on a super-large language model, wherein the second processing step includes a step of updating and storing predictive diagnosis for the occurrence of the early warning and the one or more equipment failure response information in a second predictive diagnosis database included in the diagnosis prediction module.

8. In paragraph 7, The above second predictive diagnosis database is managed by the closed industrial facility or by industrial sector including the closed industrial facility, and when managed by industrial sector, is an intelligent predictive diagnosis method based on a super-large language model that can be used for intelligent predictive diagnosis of other closed industrial facilities similar to the closed industrial facility.

9. In paragraph 6 or paragraph 8, An intelligent prediction diagnosis method based on a large-scale language model, wherein the second processing step includes a processing step including encryption or a processing method equivalent thereto for preventing the identification of the closed industrial facility and blocking external exposure prohibited information related to the closed industrial facility through at least one of the prediction diagnosis request information and the alarm cause analysis information input to the diagnosis prediction module.

10. In paragraph 6 or 7, An intelligent prediction diagnosis method based on a large-scale language model, wherein the second processing step includes a step of linking with the diagnosis prediction module provided in an external system or calling the diagnosis prediction module provided in the intelligent prediction diagnosis device.

Images