Intelligent FMEA analysis content generation system and method

Abstract

The invention discloses an intelligent FMEA analysis content generation system and method, and relates to the technical field of FMEA analysis. Comprises: a failure data capturing module, which captures product failure information from multiple sources by using a semantic understanding-based failure knowledge large model; the expansion entity triad generation module is used for converting the failure information into a dynamic FMEA expansion entity triad containing weights based on FMEA analysis specification requirements; the data updating module is used for checking the expansion entity triad and updating the database; the data export module retrieves information from a database, complements risk coefficient scores in combination with an AI technology, and finds weak links; and the information interaction module is used for completing analysis information and standard input and displaying an analysis result. According to the method, the semantic large model and the AI technology are combined, fault information automatic grabbing and weak link automatic searching are achieved, the method is suitable for FMEA analysis of complex equipment, the FMEA analysis threshold is lowered, and the analysis efficiency and accuracy are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of FMEA analysis, and particularly relates to an intelligent FMEA analysis content generation system and method. BACKGROUND

[0002] With the continuous increase of equipment product functions, the number of subsystems is increasing, the failure modes of constituent units are various, and the determination difficulty of key failure modes is greatly improved. And the application proportion of new technology, new design and new method is increasing, and the experience data available for reference of products is gradually reduced. How to quickly and accurately find possible design weak links through qualitative analysis method under complex design requirements without omission, and propose corresponding design improvement and control measures, is a problem that needs to be solved in the field of equipment design and manufacturing.

[0003] As one of the analysis methods that must be carried out in the field of equipment design and manufacturing, the traditional reliability analysis method often uses failure mode and effects analysis (FMEA) to carry out risk identification analysis. Design personnel can quickly determine the design weak link through the analysis method of FMEA, and propose corresponding improvement measures. However, in the process of a large number of engineering practices, when facing complex systems, the traditional FMEA method has the following shortcomings: (1) The traditional FMEA analysis method finds possible failure modes and effects through artificial deductive reasoning, and the analysis process is carried out by the design experience of the analyst, and the accuracy of the analysis result has a high dependence on the analyst. For the analyst who lacks design experience, the failure modes and influence factors are easy to be omitted in the analysis process, which leads to inaccurate FMEA analysis results; (2) The FMEA analysis process requires a comprehensive analysis of possible failure modes and influence factors at each level from the product as a whole to the smallest constituent parts, and the analysis steps are tedious and the analysis time cost is high; (3) Design personnel carry out FMEA analysis around possible failure modes of products, and store FMEA analysis results in the form of paper reports, which limits the sharing and reuse of results in form. At the same time, in the content such as field failure data, similar product reliability test data, product design report and the like, there may also be useful product failure information. These information is difficult to reuse in the process of FMEA analysis, which may cause the product failure analysis to be incomplete.

[0004] Therefore, it is necessary to improve the existing FMEA analysis method, so that the designer can quickly and accurately locate the key failure mode during risk identification analysis, and propose the corresponding design improvement or use compensation measures. In addition, considering the digitalization and intelligentization demand in the equipment design process, if the FMEA design method can be intelligently improved, the FMEA analysis information extraction and automatic generation system is designed, and the intelligentization of the whole process of FMEA analysis is realized, the FMEA analysis process will be greatly simplified, and the development of FMEA analysis automation will be promoted. In some researches, an FMEA analysis database based on a knowledge graph has been constructed, and part of the intelligence of FMEA analysis is realized by constructing a fixed triple form. However, the entity fusion technology under the fixed alignment rule cannot realize the dynamic conversion of failure information to FMEA special extended triples. When facing FMEA analysis specifications in different industries, it is difficult to represent the industry-specific triple relationship, and the applicability to different equipment is low, and it is difficult to identify key failures.

[0005] Based on this, the present application provides an intelligent FMEA analysis content generation system and method to solve the problems existing in the prior art. SUMMARY

[0006] To solve the above technical problems, the present application provides an intelligent FMEA analysis content generation system and method to solve the problems of the entity fusion technology under the fixed alignment rule, which cannot realize the dynamic conversion of failure information to FMEA special extended triples; when facing FMEA analysis specifications in different industries, it is difficult to represent the industry-specific triple relationship, and the applicability to different equipment is low, and it is difficult to identify key failures.

[0007] To achieve the above purpose, the present application provides the following technical solutions:

[0008] The present application provides a first scheme: an intelligent FMEA analysis content generation system, comprising:

[0009] The failure data extraction module is used to train the failure knowledge big model based on semantic understanding in advance with existing data as the training set, and extract product failure information from multiple sources;

[0010] The extended entity triple generation module is used to structure and normalize the extracted product failure information, convert it into structured data, and calculate the dynamic weight of the failure information; according to the requirements of the selected FMEA analysis specification, generate self-defined RDF triple mapping rules including special requirements in the analysis specification, and convert the processed structured data into FMEA analysis information extended entity triples including weight information according to the processing rules;

[0011] a data updating module configured to determine whether the candidate triples meet requirements through multi-dimensional auditing, and store data meeting the requirements in the FMEA analysis content knowledge base;

[0012] a data exporting module configured to search for required analysis content from the FMEA analysis content knowledge base, read system composition information and execution function information, combine the searched failure information to give a hazard degree score of the failure mode, search for a design weak link according to the score, and finally generate an FMEA analysis report according to the FMEA analysis specification requirements;

[0013] an information interaction module configured to input, select and display data extraction content, input and select FMEA analysis specification content and requirements, display final output report content, and complete specified user interaction.

[0014] In a preferred embodiment of the present application, the failure data extraction module comprises:

[0015] a failure knowledge large model based on semantic understanding, configured to receive a specified extraction object from a data input interface of the information input interaction module, and find, identify and extract relevant failure information from multiple sources by using the knowledge large model;

[0016] an entity recognition and relationship extraction module, configured to label FMEA entity tag types in the extracted failure information by using a domain ontology model constructed in the FMEA analysis content knowledge base and combining a pre-trained language model, and identify possible relationships between entities by using a deep learning model.

[0017] In a preferred embodiment of the present application, the failure knowledge large model based on semantic understanding comprises:

[0018] an active learning training module, configured to receive pre-training data from the FMEA analysis content knowledge base and input the pre-training data into the training module, judge data confidence, process and train the failure knowledge large model according to the confidence level, and optimize the failure knowledge large model through an adversarial training mechanism and update the training data;

[0019] a multi-source heterogeneous data analysis module, comprising an OCR and PDF analysis module and a table and chart recognition module; the OCR and PDF analysis module is configured to analyze scanned documents by using an AI optical character recognition technology, and extract text content from different data sources by combining a PDF text extraction technology; the table and chart recognition module is configured to recognize and extract key information in chart content by using a computer vision model, and output information content to a subsequent module;

[0020] The semantic understanding and information filtering module is configured to parse unstructured text by using a pre-trained model, extract context semantic features to form relevant fault parameters, and determine whether the parsed fault information is invalid information.

[0021] In a preferred embodiment of the present application, the extended entity triple generation module comprises:

[0022] The data structuring and dynamic weight updating module is configured to receive the preliminary processed failure information from the entity recognition and relationship extraction module, convert the failure information into structured JSON format data by using a Python program, and perform risk scoring; or generate entity correlation weights by using a neural network algorithm according to the context semantic feature information, add the entity correlation weights to the structured data, and transmit the structured data and the entity correlation weights to the FMEA analysis information extended triple generation module.

[0023] The FMEA analysis information extended triple generation module is configured to receive the user-defined triple relationship and mapping rule from the self-defined triple relationship module, combine the entity correlation weights generated by the data structuring and dynamic weight updating module, automatically generate extended triples including a head entity, a relationship, a tail entity and a weight, replace the traditional fixed format triple structure, and transmit the candidate triple information generated by combining the language model to the FMEA information multi-dimensional verification module.

[0024] In a preferred embodiment of the present application, the data updating module comprises:

[0025] The FMEA information multi-dimensional verification module is configured to automatically and / or manually audit and judge the accuracy and expandability of the candidate triples from multiple dimensions, output a judgment result for each candidate triple after the judgment, output a further instruction if the triple meets the requirements, and output the triple information to the FMEA analysis content updating module; if the analysis content does not meet the requirements, terminate and proceed to the next information judgment.

[0026] The FMEA analysis content updating module is configured to receive the FMEA analysis information extended entity triple that meets the requirements after the judgment, and update and add the FMEA analysis information extended entity triple to the FMEA analysis knowledge base based on the ontology model and the knowledge graph according to the pre-filled module information.

[0027] In a preferred embodiment of the present application, the data exporting module comprises:

[0028] The data retrieval module is configured to read the product information to be searched by using the semantic understanding large model, search the extended triple information of the required component from the FMEA analysis content knowledge base, and transmit the found information to the weak link identification module based on the large model.

[0029] The big model-based weak link identification module is used for identifying system components and functions to be performed through a semantic understanding-based failure knowledge big model; weight information in the retrieved component unit is expanded, a neural network algorithm is used to generate a hazard degree score of each piece of failure information, and the hazard degree scores are sorted to find main weak links of the system, and finally the analysis result is transmitted to the FMEA analysis report filling module;

[0030] The FMEA analysis report filling module is used for filling the searched and analyzed FMEA analysis information into a corresponding report template under the FMEA analysis specification, and transmitting the final result to the result export interaction module.

[0031] In a preferred embodiment of the present application, the information interaction module comprises

[0032] The information input interaction module is used for providing an interactive interface, displaying a range of information to be captured, selecting relevant failure information content to be captured, inputting and selecting FMEA analysis specification content to be used, providing a window for inputting failure information, and displaying system errors in the generation process;

[0033] The result export interaction module is used for providing an interactive interface and displaying FMEA analysis report content finally output.

[0034] In a preferred embodiment of the present application, the FMEA analysis content generation system further comprises:

[0035] The FMEA analysis specification module is connected with the information input interaction module, the custom three tuple relationship module and the FMEA analysis report filling module, and is used for manually selecting an FMEA analysis specification standard to be used and outputting specification requirements to the custom three tuple relationship module and the FMEA analysis report filling module;

[0036] The custom three tuple relationship module is connected with the FMEA analysis specification module, the FMEA analysis content knowledge base and the expanded entity three tuple generation module, and is used for receiving possible special entities and entity relationships from the FMEA analysis specification requirements, automatically generating and / or manually inputting RDF three tuple mapping rules suitable for special relationships of a certain FMEA analysis specification, constructing a directed edge in the FMEA analysis content knowledge base for storing related content, and finally outputting the custom RDF three tuple mapping rules to the expanded entity three tuple generation module.

[0037] The present application provides a second scheme: an intelligent FMEA analysis content generation method, which is realized based on an intelligent FMEA analysis content generation system.

[0038] In a preferred embodiment of the present application, the method comprises the following steps:

[0039] Step S1. Establish the FMEA analysis content knowledge base;

[0040] Determine the reference FMEA analysis specification requirements to be selected, extract the relevant rules and relationships in the analysis specification, and combine the collected existing FMEA analysis data to construct an initial version of the FMEA analysis content knowledge base based on the ontology model and knowledge graph, and store the FMEA analysis information in modules;

[0041] Step S2. Identify and extract failure information through the failure knowledge large model based on semantic understanding trained in advance in the failure data grabbing module;

[0042] Through the information input interaction module, the user inputs the source file of the failure data information to be grabbed, and delimits the search range; then the failure knowledge large model based on semantic understanding trained in advance is used to find the required failure information within the delimited search range, and the failure data information meeting the condition requirements is preliminarily screened out; and the entity recognition and relationship extraction model is used for failure data information extraction;

[0043] Step S3. Use the extended entity triple generation module to convert the product failure information grabbed into FMEA analysis information extended entity triples including weight information;

[0044] Through the extended entity triple generation module, the product failure information grabbed is structured and normalized, converted into structured data, and the correlation weight of the failure information is calculated; then according to the requirements of the selected FMEA analysis specification, the self-defined RDF triple mapping rules including special requirements in the analysis specification are generated, and the processed structured data is converted into FMEA analysis information extended entity triples in the form of "head entity-relation-tail entity-weight" according to the processing rules;

[0045] Step S4. Use the data update module to review the candidate triples, and store the data meeting the requirements into the FMEA analysis content knowledge base;

[0046] Step S5. Use the data export module to give the hazard rating of the failure mode, automatically find the weak design link, and generate the FMEA analysis report;

[0047] Use the large model technology to read the system composition information and execution function information, and search for the failure analysis content of the related composition units and functions from the FMEA analysis content knowledge base through the data export module; combine the failure information and system composition, use the neural network algorithm to give the hazard rating of the failure mode, search for the weak design link according to the rating, and finally automatically generate the FMEA analysis report according to the FMEA analysis specification requirements;

[0048] Step S6. Input and select FMEA analysis specification content and requirements using the information interaction model, display the final output report content, and complete the specified user interaction.

[0049] Compared with the prior art, the intelligent FMEA analysis content generation system and method provided by the application has the following beneficial effects:

[0050] 1. The intelligent failure information recognition system is constructed by introducing a multi-source heterogeneous data analysis method and a semantic large model, the function of filtering and extracting fault data information from various data sources is realized, a stable information source is provided for the automatic filling and generation of FMEA analysis information, and an information framework foundation is established for the realization of the intelligent FMEA analysis method.

[0051] 2. The application proposes an extended entity triple model including weight information, models the relationship between failure modes and importance, gives the failure mode weight through a neural network algorithm, facilitates rapid positioning of key faults, and improves reliability design efficiency and FMEA analysis accuracy.

[0052] 3. The application establishes a method that can define triple mapping rules according to FMEA analysis specification requirements, compared with the triple information formed by the fixed mapping rules in the traditional method, it can better reflect the industry characteristics, and improve the adaptability of the method to FMEA analysis results of different equipment products.

[0053] 4. The application realizes machine understanding of system composition, functional principle and failure mode by establishing a failure knowledge large model based on semantic understanding. Further, by introducing neural network technology, a component unit failure mode risk score algorithm is formed, realizing the automatic search function of system design weak links, enabling designers to quickly locate key failure modes, greatly improving the accuracy and efficiency of FMEA analysis results. The application solves the problems of the prior art, such as the inability of entity fusion technology under fixed alignment rules to realize dynamic conversion of failure information to FMEA special extended triple, the inability to represent industry-specific triple relationships when facing FMEA analysis specifications of different industries, and the low applicability to different equipment, which cannot identify key faults. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 The intelligent FMEA analysis content generation system structure diagram provided by the application;

[0055] Figure 2 The intelligent FMEA analysis content generation method flowchart provided by the application;

[0056] Figure 3The workflow diagram of the failure knowledge large model based on semantic understanding established by the present application is shown in the figure;

[0057] Figure 4 The internal composition and working principle diagram of the failure knowledge large model based on semantic understanding established by the present application are shown in the figure;

[0058] Figure 5 The workflow diagram of the active learning training module established by the present application is shown in the figure;

[0059] Figure 6 The information interaction interface provided by the information display module established by the present application is shown in the figure.

[0060] In the drawings: 1, failure knowledge large model based on semantic understanding; 2, entity recognition and relationship extraction module; 3, FMEA analysis specification module; 4, self-defined triple relationship module; 5, data structuring and dynamic weight updating module; 6, FMEA analysis information expansion triple generation module; 7, FMEA information multi-dimensional verification module; 8, FMEA analysis content updating module; 9, data retrieval module; 10, weak link identification module based on large model; 10, FMEA analysis report filling module; 12, information input interaction module; 13, result export interaction module. DETAILED DESCRIPTION

[0061] The present application will be further described below in combination with the drawings and specific embodiments.

[0062] Please refer to Figures 1-6 The embodiment provides an intelligent FMEA analysis content generation system, which comprises a failure data grabbing module, an expanded entity triple generation module, a data updating module, a data exporting module and an information interaction module. Specifically,

[0063] The failure data grabbing module is connected with the expanded entity triple generation module, the information interaction module and the FMEA analysis content knowledge base, and is used for training the failure knowledge large model based on semantic understanding in advance by taking existing data as a training set, and extracting product failure information from similar product FMEA analysis reports, field failure data, reliability test data and other sources, and delivering the grabbed content to the expanded entity triple generation module.

[0064] The expansion entity triple generation module is connected with the failure data crawling module, the self-defined triple relationship module 4 and the data updating module, is used for structuring and normalizing the crawled product failure information, converting into structured data, and calculating the dynamic weight of the failure information; then, according to the requirements of the selected FMEA analysis specification, a self-defined RDF triple mapping rule including special requirements in the analysis specification is generated, the processed structured data is converted into FMEA analysis information expansion entity triples containing weight information according to the processing rule, and the candidate triple content is passed back;

[0065] The data updating module is connected with the expansion entity triple generation module and the FMEA analysis content knowledge base, is used for auditing whether the candidate triple meets the requirements from multiple dimensions such as semantic conflict detection, cross-data source consistency verification and industry standard compliance checking, and storing the data meeting the requirements into the FMEA analysis content knowledge base;

[0066] The data export module is connected with the FMEA analysis content knowledge base, the FMEA analysis specification module 3 and the information export interaction module 13, is used for searching the required analysis content from the FMEA analysis content knowledge base, reading the system composition information and execution function information by using the large model technology, combining the retrieved fault information to give a hazard degree score of the fault mode, searching the design weak link by itself, and finally generating an FMEA analysis report according to the requirements of the FMEA analysis specification;

[0067] The information interaction module is connected with the failure data crawling module and the data export module, is used for inputting, selecting and displaying the data crawling content, inputting and selecting the FMEA analysis specification content and requirements, displaying the final output report content, and can complete the specified user interaction.

[0068] In a preferred embodiment, as shown in Figure 3 and Figure 4 The failure data crawling module includes a semantic understanding-based failure knowledge large model 1 and an entity recognition and relationship extraction module 2. Specifically:

[0069] As shown in Figure 3 The semantic understanding-based failure knowledge large model 1 is connected with the entity recognition and relationship extraction module 2, the information input interaction module 12 and the FMEA analysis content knowledge base, is used for receiving the specified crawling object from the data input interface in the information input interaction module 12, and finding, recognizing and crawling the related fault information from various sources such as field fault data, similar product fault data, reliability test data and similar product FMEA analysis report by using the pre-trained knowledge large model.

[0070] Specifically in use, first of all, the user inputs the source file of the failure data information that needs to be captured, and the search range is delimited; then the pre-trained failure knowledge large model based on semantic understanding 1 is used to find the required failure information in the delimited search range, and the failure data information meeting the condition requirements is preliminarily screened out and transmitted to the entity recognition and relationship extraction module 2 for further processing of the information.

[0071] The entity recognition and relationship extraction module 2 is connected with the failure knowledge large model based on semantic understanding 1, the FMEA analysis content knowledge base and the data structuring and dynamic weight updating module 5, and is used to use the domain ontology model already constructed in the FMEA analysis content knowledge base, in combination with the pre-training language model, to label the FMEA entity label types of no less than 15 in the failure information captured, such as failure mode, failure cause and failure impact. And through the deep learning model, the possible relationship between entities is identified, such as the causal relationship between "bearing wear" and "abnormal noise", etc., which is transmitted to the data structuring processing module.

[0072] Specifically, the failure knowledge large model based on semantic understanding 1 includes an active learning training module, a multi-source heterogeneous data analysis module and a semantic understanding and information filtering module. More specifically:

[0073] As shown in Figure 4 The active learning training module is connected with the FMEA analysis content knowledge base, and is used to receive the pre-set training data from the FMEA analysis content knowledge base and bring it into the training module, judge the data confidence, process and train the failure knowledge large model according to the confidence level, and optimize the failure knowledge large model through the adversarial training mechanism, and update the training data in a timely manner for possible emerging failure modes;

[0074] As shown in Figure 4 The multi-source heterogeneous data analysis module is connected with the semantic understanding and relationship extraction module and the active learning training module, and includes an OCR and PDF analysis module and a table and chart recognition module.

[0075] The semantic understanding and relationship extraction module is connected with the multi-source heterogeneous data analysis module and the entity recognition and relationship extraction module 2, and is used to judge whether the parsed failure information is invalid information.

[0076] Further, as shown in Figure 5As shown, in use, the active learning training module is configured to receive training data from the FMEA analysis content knowledge base, input the data into a model prediction function of the module, and determine the confidence of the data. For high-confidence data, the data is directly used as training data for the model, and the failure knowledge large model based on semantic understanding is iterated. For low-confidence data, the data is returned to a display area for manual modification or annotation, and the annotated model is re-added to the training set for model training, achieving the purpose of iterating the failure knowledge large model. After obtaining the training data, the failure knowledge large model is continuously optimized through an adversarial training mechanism, so that the model can adapt to new objects and failure modes.

[0077] Exemplarily, in use, the active learning training module is configured to perform the following operation steps S2111-S2115:

[0078] Step S2111. First, a basic failure knowledge large model is established using semantic large model technology.

[0079] Step S2112. Training data information is extracted from the initial version of the FMEA analysis content knowledge base and input into the basic failure knowledge large model, and the confidence of the training data is determined.

[0080] Step S2113. For failure data with high confidence, the data is directly added to the training set to train and iterate the basic model, forming a new failure knowledge large model based on semantic understanding.

[0081] Step S2114. For failure data with low confidence, the data is returned to the FMEA analysis content knowledge base for manual modification. For the modified data returned, the confidence is again determined. If the data meets the requirements, the basic model is trained and iterated to form a new failure knowledge large model. If the data still does not meet the requirements, the data is discarded and the next set of data is processed.

[0082] Step S2115. The failure knowledge large model is continuously optimized using an adversarial training mechanism, while reserving the model update capability when new failure modes or new objects appear.

[0083] Further, the OCR and PDF analysis module is configured to analyze scanned documents using AI optical character recognition (OCR) technology, and extract text content from similar product FMEA reports, field failure data reports, reliability test reports, and other different data sources using PDF text extraction technology.

[0084] Exemplarily, in use, the OCR and PDF analysis module is configured to perform the following operation steps S2121-S2122:

[0085] Step S2121. Input the failure data analysis file, specify the capture range of the failure data, and use the trained failure knowledge large model based on semantic understanding to capture the failure data.

[0086] Step S2122: Use the optical character recognition (OCR) technology in the failure knowledge large model to analyze the scanned document, and combine the PDF text extraction technology to extract the text content from the specified capture range.

[0087] Further, the table and chart recognition module is used to use a computer vision (CV) model to identify and extract key information from FMEA analysis tables, fault trees, product structure composition diagrams, task reliability models, and other chart contents, and output the information content to the subsequent module.

[0088] Illustratively, the table and chart recognition module, when in use, is used to perform the following operation steps S2131:

[0089] Step S2131. Use a computer vision (CV) model to identify and extract information content in FMEA analysis tables, fault trees, product structure composition diagrams, task reliability models, and other charts in the specified capture range.

[0090] Further, the semantic understanding and relationship extraction module uses a FMEA field-specific pre-trained model to analyze unstructured text, identify core entities such as "failure mode", "fault cause", "impact consequence", and extract context semantic features (such as "high temperature" in "high temperature environment bearing wear" as a failure condition), form more fault-related parameters such as fault occurrence condition, frequency, severity, etc., and finally pass the final information to the entity recognition and relationship extraction module.

[0091] Illustratively, the semantic understanding and relationship extraction module, when in use, is used to perform the following operation steps S2132 and S2133:

[0092] Step S2132. Determine whether the captured text and chart information is invalid information. If it is determined to be invalid information, it is rejected and re-identified. If it is valid information, the next step is performed;

[0093] Step S2133. Use a sequence labeling model to extract more fault-related parameters such as fault occurrence condition, frequency, severity, etc.

[0094] Further, the entity recognition and relationship extraction module, when in use, is used to perform the following operation steps S221-S222:

[0095] Step S221. Utilize a pre-built FMEA-specific pre-trained model to parse the captured unstructured failure data information, identify and annotate the core entities such as failure mode, failure cause, and impact in the captured failure information;

[0096] Step S222: Utilize a deep learning model, combined with a neural network algorithm, to construct possible relationship paths, such as causal relationship paths, hierarchical relationship paths, etc. The annotated failure information is placed in the paths for comparison, and possible relationships between entities are identified.

[0097] In a preferred embodiment, Figure 1 As shown, the expanded entity triple generation module includes a data structuring and dynamic weight updating module 5 and a FMEA analysis information expanded triple generation module 6. Specifically:

[0098] The data structuring and dynamic weight updating module 5 is connected to the entity recognition and relationship extraction module 2 and the FMEA analysis information expansion triple generation module 6, and is used to receive the failure information after preliminary processing from the entity recognition and relationship extraction module 2, use the Python program to convert the failure information into structured JSON format data, and score the failure information based on the severity, frequency of occurrence, detection difficulty, etc., or generate entity association weights through a neural network algorithm based on contextual semantic feature information, add them to the structured data, and pass them together to the FMEA analysis information expansion triple generation module 6.

[0099] The FMEA analysis information extended triple generation module 6 is connected to the data structuring and dynamic weight updating module 5, the custom triple relationship module 4 and the FMEA information multi-dimensional verification module 7, and is used to receive user-defined triple relationships and mapping rules from the custom triple relationship module 4, and automatically generate an extended triple including "head entity-relationship-tail entity-weight" in combination with the entity association weight generated by the data structuring and dynamic weight updating module 5, such as "bearing→wear→abnormal noise, weight 0.8", to replace the traditional fixed format triple structure, and transmit the candidate triple information generated in combination with the language model to the FMEA information multi-dimensional verification module 7.

[0100] Furthermore, the data structuring and dynamic weight updating module 5 utilizes Python programming during use, and the entire process employs streaming file parsing and exception handling mechanisms to ensure data consistency and robustness across multiple rounds of model calls. When calculating the weight of failure information, a neural network algorithm generates associated weights based on the severity (S), frequency (O), and detection difficulty (D) of the failure data. For example, the weight of "engine fracture" is set to 0.9 (high severity), while "bolt loosening" is set to 0.6 (medium risk), enabling the system to locate critical fault information more quickly.

[0101] Exemplarily, the data structuring and dynamic weight updating module 5, when in use, is configured to perform the following operation steps S311-S312:

[0102] Step S311. Convert the FMEA analysis information after entity recognition and relationship extraction into structured data in JSON format.

[0103] Step S312. According to the frequency, severity, and detectability information in the failure information, generate entity correlation weights using a neural network algorithm, and add them to the structured data.

[0104] Exemplarily, the FMEA analysis information expansion triple generation module 6, when in use, is configured to perform the following operation step S321:

[0105] Step S321. According to the selected FMEA analysis specification requirements, generate a filled report template and custom relationship mapping rules. Bring the extracted structured failure data into the custom mapping rules, combine the calculated correlation weights, and map them into expansion triples in the form of "head entity-relation-tail entity-weight", and generate candidate triple information combined with the language model.

[0106] In a preferred embodiment, as shown in Figure 1 The data updating module includes an FMEA information multi-dimensional verification module 7 and an FMEA analysis content updating module 8. Specifically:

[0107] The FMEA information multi-dimensional verification module 7 is connected to the FMEA analysis information expansion triple generation module 6 and the FMEA analysis content updating module 8, and is configured to automatically and / or manually audit and judge the accuracy and scalability of the candidate triples from multi-dimensional aspects such as semantic conflict detection, cross-data source consistency verification, and industry standard compliance checking. After judgment, output the judgment result for each candidate triple. If the triple meets the requirements, output further instructions and output the triple information to the FMEA analysis content updating module 8. If the analysis content does not meet the requirements, terminate and proceed to the next information judgment.

[0108] The FMEA analysis content updating module 8 is connected to the FMEA information multi-dimensional verification module 7 and the FMEA analysis content knowledge base, and is configured to receive the FMEA analysis information expansion entity triples that meet the requirements after judgment, and update and add them to the already established FMEA analysis knowledge base based on the ontology model and knowledge graph according to the pre-filled module information.

[0109] Further, the FMEA information multi-dimensional verification module 7 is mainly used to design a four-layer verification mechanism for the candidate extended triple information. The first layer performs uniqueness inspection. This step compares the candidate triple information with the existing information in the FMEA analysis content knowledge base, confirms whether it is repeated, and if there is repetition, returns a conflict: similar information alarm, and returns manual judgment. The second layer performs semantic conflict detection. Based on the domain ontology model constructed in the FMEA analysis content knowledge base, the logical rationality of the entity relationship is verified. For example, if the head entity is “bearing” and the tail entity is “software error”, a conflict alarm is triggered. Because the bearing is a mechanical component, it has no direct connection with the software error. The third layer performs consistency verification across data sources. This step needs to compare the related parameters of the same failure mode in multi-source data (such as field failure report, reliability test data, and historical FMEA report) to check whether the parameters are consistent. For example, if the occurrence frequency of “gear fracture” is 0.1% in the test data, and 5% in the field report, it is marked as a data conflict, and manual review is triggered. The fourth layer performs industry standard compliance check. According to the selected FMEA analysis specification requirements, it verifies whether the data meets the standard format and content requirements. For example, it checks whether the “detection measures” include “automatic diagnosis system” and other mandatory words to ensure that the report can be directly used for compliance certification.

[0110] For example, the FMEA information multi-dimensional verification module 7 is used to perform the following operation steps S41 when in use:

[0111] Step S41. Manually review the candidate triple information from independence inspection, semantic conflict detection, cross-data source consistency verification, and industry standard compliance check, and judge the accuracy and expandability.

[0112] Further, the FMEA analysis content update module 8 receives the required extended triple information from the FMEA information multi-dimensional verification module 7, and directly connects to the corresponding information classification node of the FMEA analysis knowledge base based on the ontology model and knowledge graph according to the content of the head entity (failure unit) and tail entity (failure phenomenon) in the extended triple, and stores it in the FMEA analysis content knowledge base in the form of an instance.

[0113] For example, the FMEA analysis content update module 8 is used to perform the following operation steps S42 when in use:

[0114] Step S42. Store the FMEA analysis triple information that meets the requirements after review in the FMEA analysis content knowledge base, and update the initial version of the FMEA analysis content knowledge base.

[0115] In a preferred embodiment, as Figure 1As shown, the data export module includes a data retrieval module 9, a large model-based weak link identification module 10, and an FMEA analysis report filling module 11. Specifically:

[0116] The data retrieval module 9 is connected to the FMEA analysis content knowledge base and the large model-based weak link identification module 10, and is used to read the composition information of the product to be searched through semantic understanding of the large model, and search the expansion triple information of the components to be analyzed from the FMEA analysis content knowledge base, and pass the found information to the next module;

[0117] The large model-based weak link identification module 10 is connected to the data retrieval module 9 and the FMEA analysis report filling module 11, and is used to identify the system composition and functions to be performed through the failure knowledge large model 1 based on semantic understanding. In combination with the weight information in the retrieved component unit expansion triple, the neural network algorithm is used to generate the hazard degree score of each failure information, and the hazard degree score is sorted to find the main weak link of the system, and finally the analysis result is transmitted to the FMEA analysis report filling module 11;

[0118] The FMEA analysis report filling module 11 is connected to the large model-based weak link identification module 10, the FMEA analysis specification module 3, and the result export interaction module 13, and is used to fill the searched and analyzed FMEA analysis information into the corresponding report template under the previously selected FMEA analysis specification, and transmit the final result to the result export interaction module.

[0119] Further, the data retrieval module 9 is used to use the knowledge large model based on semantic understanding to perform fuzzy processing on the input sentence, extract the system component unit name as the main keyword, such as "bearing", "spring", etc., transmit it back to the FMEA analysis content knowledge base, search for the matching keyword, extract the related expansion triple information, and output the query result to the large model-based weak link identification module.

[0120] Illustratively, the data retrieval module 9, when in use, is used to perform the following operation steps S511-S512:

[0121] Step S511. Determine the product composition information according to the input product composition and analysis needs, search for the FMEA analysis information and failure information of the related product in the FMEA analysis content knowledge base;

[0122] Step S512. Fill the searched failure data information into the corresponding report template under the previously selected FMEA analysis specification.

[0123] Further, the weak link identification module 10 based on a large model is used to identify the system components, structure, working principle and the function of each component unit by using the established failure knowledge model 1 based on semantic understanding, to build a system working path similar to an event tree, and each node on the path is composed of a component unit, and each path is a function performed by a node. In combination with the failure mode that may occur in the retrieved component unit, the corresponding adjustment is made on each path, and it is determined whether the system failure occurs. The neural network algorithm is used to evaluate the probability of each path, and the weight information carried in the extended triple is combined to score the hazard degree. The hazard degree scores of all failure modes are sorted to find the single or multiple failure modes with the highest hazard degree score as the main design weak link of the system, and the analysis result is passed to the next module.

[0124] For example, the weak link identification module 10 based on a large model is used to perform the following operation steps S521-S524:

[0125] Step S521. The OCR and PDF reading technology and chart recognition technology in the established failure knowledge model 1 based on semantics are used to identify the input system components, structure, working principle and the function of each component unit, and read the extended triple information of the retrieved failure mode;

[0126] Step S522. The neural network algorithm model is trained in advance by using the normalized FMEA analysis data to achieve the purpose of calculating the failure mode hazard degree score;

[0127] Step S523. The system working path diagram is established, each failure mode is imported into the system working path diagram as input, the trained neural network algorithm model is used to judge the influence of each failure mode on the system failure, and the hazard degree score of each failure mode is given;

[0128] Step S524. All hazard degree scores are sorted to find the single or multiple failure modes with the highest hazard degree score as the main design weak link of the system, and the analysis result is passed to the next module.

[0129] Further, the FMEA analysis report filling module 11 is used to perform the following operation step S53 when in use:

[0130] Step S53. According to the report template requirements provided by the FMEA analysis specification module 3, the FMEA analysis report is output, and the automatic generation of FMEA analysis is completed.

[0131] In a preferred embodiment, as Figure 1As shown, the information interaction module includes an information input interaction module 12 and a result derivation interaction module 13. Specifically,

[0132] The information input interaction module 12 is connected with the failure data grabbing module and the FMEA analysis specification module 3, and is used to provide an interactive interface, display the information range to be grabbed, select the relevant failure information content to be grabbed, input and select the FMEA analysis specification content to be used, provide a window for inputting the failure information, and display the system error in the generation process;

[0133] The result derivation interaction module 13 is connected with the FMEA analysis report filling module 11, and is used to provide an interactive interface and display the FMEA analysis report content finally output.

[0134] In a preferred embodiment, as Figure 1 As shown, the FMEA analysis content knowledge base is connected with the failure data grabbing module, the self-defined triple relationship module 4, the FMEA analysis content updating module 8, and the data retrieval module 9, and is constructed based on the ontology model technology and the knowledge graph technology, and is used to represent the entity model of the FMEA analysis, provide training data for the failure data grabbing module, store the updated FMEA analysis information expansion triple, and output the product failure information to the data derivation module.

[0135] Specifically, the FMEA analysis content knowledge base first constructs the FMEA analysis method ontology model based on the analysis specification requirements. The ontology model is used to represent each analysis content in the FMEA analysis. Further, the ontology model is connected with the knowledge graph to construct the FMEA analysis content knowledge base based on the knowledge graph. Finally, part of the standardized FMEA analysis information audited by experts is stored in different modules to form the initial FMEA analysis content knowledge base. The basic training data is provided to the failure knowledge large model based on semantic understanding 1.

[0136] Illustratively, the FMEA analysis content knowledge base is used to perform the following operation steps S11 when in use:

[0137] Step S11. Determine the reference FMEA analysis specification requirements to be selected, extract the relevant rules and relationships in the analysis specification, combine the collected existing FMEA analysis data, construct the initial version of the FMEA analysis content knowledge base based on the ontology model and the knowledge graph, and store the FMEA analysis information in modules.

[0138] In a preferred embodiment, as Figure 1As shown, the FMEA analysis specification module 3 is connected with the information input interaction module 12, the custom triple relationship module 4 and the FMEA analysis report filling module 11, for manually selecting the FMEA analysis specification standard to be adopted, and outputting the specification requirements to the custom triple relationship module and the FMEA analysis report filling module.

[0139] Specifically, in the working process of the FMEA analysis specification module 3, the FMEA analysis specification to be adopted is selected and / or imported through a drop-down menu and / or by importing a doc or PDF format file. Then the FMEA analysis specification text is automatically parsed by using natural language processing technology, and the content to be filled and the possible relationships are extracted. Finally, the corresponding result content is output to the FMEA analysis report filling module 11 and the custom triple relationship module 4 respectively.

[0140] In a preferred embodiment, as shown in Figure 1 As shown, the custom triple relationship module 4 is connected with the FMEA analysis specification module 3, the FMEA analysis content knowledge base and the extended entity triple generation module, for receiving the possible special entities and entity relationships from the FMEA analysis specification requirements, automatically generating and / or manually specifying the RDF triple mapping rules suitable for the special relationships in a certain FMEA analysis specification, and constructing directed edges in the FMEA analysis content knowledge base for storing the related content. Finally, the custom RDF triple mapping rules are output to the extended entity triple generation module.

[0141] Specifically, in the working process of the custom triple relationship module 4, the information input interaction interface in the information input interaction module 4 is connected, for providing a visual interface to support the creation of new relationship types such as "function failure-safety level" in the aerospace field by dragging and dropping, importing configuration files and the like. Then the module will automatically parse the input relationship, generate the corresponding RDF triple mapping rules, and construct directed edges in the knowledge graph. This module can support the import of mainstream FMEA analysis specifications such as AIAG-VDA, NASA, ISO26262, and generate adaptive relationship templates.

[0142] The information display interface provided by the information display module established by the present application is as shown in Figure 6 The user can input the failure information of the product to be queried in the information input box on the lower left. The system can display the query result in the interface box on the upper left.

[0143] As shown in Figures 1-6 The present embodiment provides an intelligent FMEA analysis content generation method, which is realized based on the intelligent FMEA analysis content generation system as described above. The method specifically includes the following steps:

[0144] Step S1. Establish the FMEA analysis content knowledge base;

[0145] Determine the reference FMEA analysis specification requirements to be selected, extract the relevant rules and relationships in the analysis specification, and combine the collected existing FMEA analysis data to construct an initial version of the FMEA analysis content knowledge base based on the ontology model and knowledge graph, and store the FMEA analysis information in modules;

[0146] Step S2. Identify and extract failure information through the failure knowledge large model based on semantic understanding trained in advance in the failure data grabbing module;

[0147] Through the information input interaction module 12, the user inputs the source file of the failure data information to be grabbed, and delimits the search range; then the failure knowledge large model based on semantic understanding trained in advance is used to find the required failure information within the delimited search range, and the failure data information meeting the condition requirements is preliminarily screened out; and the entity recognition and relationship extraction model is used for failure data information extraction;

[0148] Step S3. Use the extended entity triple generation module to convert the product failure information grabbed to FMEA analysis information extended entity triples including weight information;

[0149] Through the extended entity triple generation module, the product failure information grabbed is structured and normalized, converted into structured data, and the correlation weight of the failure information is calculated; then according to the requirements of the selected FMEA analysis specification, the self-defined RDF triple mapping rules including special requirements in the analysis specification are generated, and the processed structured data is converted into FMEA analysis information extended entity triples in the form of "head entity-relation-tail entity-weight" according to the processing rules;

[0150] Step S4. Use the data update module to review the candidate triples, and store the data meeting the requirements into the FMEA analysis content knowledge base;

[0151] Step S5. Use the data export module to give the hazard rating of the failure mode, automatically find the weak design link, and generate the FMEA analysis report;

[0152] Use the large model technology to read the system composition information and execution function information, and search for the failure analysis content of the related composition units and functions from the FMEA analysis content knowledge base through the data export module. Combine the fault information and system composition, use the neural network algorithm to give the hazard rating of the failure mode, search for the weak design link according to the rating, and finally automatically generate the FMEA analysis report according to the FMEA analysis specification requirements;

[0153] Step S6. Input and select FMEA analysis specification content and requirements using the information interaction model, display the final output report content, and complete the specified user interaction.

[0154] In a preferred embodiment, the process of identifying and extracting failure information by the failure data crawling module in step S2 includes:

[0155] Step S21. Use the semantic understanding-based failure knowledge large model 1 to search for the required failure information within the designated search range, and preliminarily screen out failure data information that meets the conditions and requirements;

[0156] Step S22. Use the entity recognition and relationship extraction model to extract specific failure data information.

[0157] Specifically, the process of step S21 using the semantic understanding-based failure knowledge large model 1 to search for the required failure information within the designated search range and preliminarily screen out failure data information that meets the conditions and requirements includes:

[0158] Step S211. Use the active learning model to receive basic training data from the FMEA analysis content knowledge base, input it into the model prediction and training set of the module, judge the confidence of the data, and continuously train the model using effective data;

[0159] Step S2111. First, use semantic large model technology to establish a basic failure knowledge large model;

[0160] Step S2112. Extract training data information from the initial version of the FMEA analysis content knowledge base and input it into the basic failure knowledge large model, and judge the confidence of the training data;

[0161] Step S2113. For failure data with high confidence, directly add it to the training set to train and iterate the basic model, forming a new semantic understanding-based failure knowledge large model;

[0162] Step S2114. For failure data with low confidence, return it to the FMEA analysis content knowledge base for manual modification. For the data returned after modification, judge the confidence again. If it meets the requirements, train and iterate the basic model to form a new failure knowledge large model. If it still does not meet the requirements, discard this data and proceed to the next set of data processing work;

[0163] Step S2115. Use the adversarial training mechanism to continuously optimize the failure knowledge large model, while reserving the model updating capability when new failure modes or new objects appear.

[0164] Step S212. Text content extraction is performed using the OCR and PDF parsing module.

[0165] Step S2121. The failure data analysis file is input, the capture range of the failure data is specified, and the trained failure knowledge large model based on semantic understanding is used to capture the failure data.

[0166] Step S2122: The optical character recognition (OCR) technology in the failure knowledge large model is used to analyze the scanned document, and the PDF text extraction technology is used to extract the text content from the specified capture range.

[0167] Step S213. The computer vision (CV) model is used to identify and extract key information from FMEA analysis tables, fault trees, product structure composition diagrams, and task reliability models.

[0168] Step S2131. The computer vision (CV) model is used to identify and extract information content in FMEA analysis tables, fault trees, product structure composition diagrams, and task reliability models in the specified capture range.

[0169] Step S2132. The semantic understanding and relationship extraction module is used to extract failure data information.

[0170] Step S2133. It is determined whether the captured text and diagram information is invalid information. If it is determined to be invalid information, it is rejected and returned for re-identification. If it is valid information, the next step is performed.

[0171] Step S214. More fault-related parameters, such as fault occurrence conditions, frequency, and severity, are extracted using a sequence labeling model.

[0172] Specifically, step S22 includes the following steps:

[0173] Step S221. The pre-built FMEA domain-specific pre-training model is used to analyze the captured unstructured failure data information, identify and label the failure modes, fault causes, and effects in the captured failure information.

[0174] Step S222. The deep learning model is used in combination with a neural network algorithm to construct possible relationship paths, such as causal relationship paths and hierarchical composition relationship paths. The labeled failure information is placed in the path for comparison to divide the possible relationships between entities.

[0175] In a preferred embodiment, the process of converting the captured product failure information into FMEA analysis information extended entity triples including weight information in step S3 utilizes the extended entity triple generation module, which includes:

[0176] Step S31. Convert the failure information into structured data in JSON format using the data structuring and dynamic weight updating module 5;

[0177] Step S311. Convert the FMEA analysis information after entity recognition and relationship extraction into structured data in JSON format;

[0178] Step S312. According to the frequency, severity, detection degree and other information in the failure information, generate entity correlation weights using neural network algorithm and add them to the structured data.

[0179] Step S32. Generate extended triples including "head entity-relation-tail entity-weight" using the FMEA analysis information extended triple generation module 6, instead of the traditional fixed format triple structure;

[0180] Step S321. According to the selected FMEA analysis specification requirements, generate a fill-in report template and custom relationship mapping rules. Bring the extracted structured failure data into the custom mapping rules, combine the calculated correlation weights, and map them into extended triples in the form of "head entity-relation-tail entity-weight", and generate candidate triple information combined with the language model.

[0181] In a preferred embodiment, the process of step S4 utilizes the data updating module to review the candidate triples and stores the data that meets the requirements in the FMEA analysis content knowledge base, which includes:

[0182] Step S41. Manually review the candidate triple information from multiple dimensions such as independence test, semantic conflict detection, cross-data source consistency verification and industry standard compliance check, to judge accuracy and scalability;

[0183] Step S42. Store the FMEA analysis triple information that meets the requirements after review into the FMEA analysis content knowledge base, and update the initial version of the FMEA analysis content knowledge base.

[0184] In a preferred embodiment, the process of step S5 utilizes the data export module to give the hazard rating of the failure mode, automatically find the weak links in the design, and generate the FMEA analysis report, which includes:

[0185] Step S51. Use the data retrieval module 9 to perform fuzzy processing on the input sentence, extract the system component name, and search for failure information and FMEA analysis information;

[0186] Step S52. Score and rank all failure modes by using the big model based weak link identification module 10, find the single or multiple failure modes with the highest severity score as the main design weak link of the system;

[0187] Step S53. The FMEA analysis report filling module 11 outputs the FMEA analysis report according to the report template requirements provided by the FMEA analysis specification module 3, and completes the automatic generation of the FMEA analysis results.

[0188] Specifically, step S51 uses the data retrieval module 9 to perform fuzzy processing on the input sentence, extracts the system component name, and the specific process of searching failure information and FMEA analysis information includes:

[0189] Step S511. Determine the product composition information according to the input product composition and analysis needs, search for the FMEA analysis information and failure information of the related product in the FMEA analysis content knowledge base;

[0190] Step S512. Fill the failure data information found into the corresponding report template under the FMEA analysis specification selected in advance.

[0191] Specifically, step S52 uses the big model based weak link identification module 10 to score and rank all failure modes, find the single or multiple failure modes with the highest severity score as the main design weak link of the system, and the specific process includes:

[0192] Step S521. Use the OCR and PDF reading technology and chart recognition technology in the already established semantic based failure knowledge big model 1 to identify the input system's component units, component structure, working principle and each component unit's function execution information, and read the expanded triple information of the failure mode retrieved;

[0193] Step S522. Pre-use standardized FMEA analysis data to train neural network algorithm model to achieve the purpose of calculating failure mode severity score;

[0194] Step S523. Establish a system work path diagram, import each failure mode as input into the system work path diagram, use the trained neural network algorithm model to judge the influence of each failure mode on system failure, and give the severity score of each failure mode;

[0195] Step S524. Rank all severity scores, find the single or multiple failure modes with the highest severity score as the main design weak link of the system, and pass the analysis results to the next module.

[0196] The embodiment provides an intelligent FMEA analysis content generation electronic device, which comprises a processor, a communication interface, a memory and a communication bus, wherein the processor is connected with the memory through the communication interface and the communication between the processor and the memory is completed through the communication bus; the memory is used for storing a computer program; and the processor is used for executing the program stored on the memory, and the intelligent FMEA analysis content generation method of the intelligent FMEA analysis content generation system is run.

[0197] The embodiment provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executed by a processor to realize the intelligent FMEA analysis content generation method of the intelligent FMEA analysis content generation system.

[0198] The embodiment provides a program product, which comprises computer program instructions, and the computer program instructions are executed to realize the intelligent FMEA analysis content generation method of the intelligent FMEA analysis content generation system.

[0199] Although the embodiments of the present application have been shown and described, it is to be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. An intelligent FMEA analysis content generation system, characterized by: include: The failure data capture module is used to pre-train a large failure knowledge model based on semantic understanding using existing data as a training set, and to extract product failure information from multiple sources; The extended entity triple generation module is used to structure and normalize the captured product failure information, convert it into structured data, and calculate the dynamic weight of the failure information. According to the requirements of the selected FMEA analysis specification, it generates custom RDF triple mapping rules that include the special requirements of the analysis specification. According to the processing rules, the processed structured data is converted into FMEA analysis information extended entity triples including weight information. The data update module is used to review the candidate triples to see whether they meet the requirements through multiple dimensions, and store the data that meet the requirements into the FMEA analysis content knowledge base; The data export module is used to search for the required analysis content from the FMEA analysis content knowledge base, read the system composition information and execution function information, combine the retrieved fault information, give the criticality score of the fault mode, search for design weaknesses based on the score, and finally generate the FMEA analysis report according to the FMEA analysis specification requirements; The information interaction module is used to input, select and display data capture content, input and select FMEA analysis specification content and requirements, display the final output report content, and complete the specified user interaction.

2. The intelligent FMEA analysis content generation system according to claim 1, characterized in that: The failure data capture module includes: A failure knowledge macromodel based on semantic understanding is used to receive specified captured objects from the data input interface of the information input interaction module, and use the knowledge macromodel to find, identify, and capture relevant fault information from multiple sources; The entity recognition and relationship extraction module is used to use the domain ontology model constructed in the FMEA analysis content knowledge base, combined with the pre-trained language model, to annotate the FMEA entity label type in the captured failure information, and identify possible relationships between entities through a deep learning model.

3. The intelligent FMEA analysis content generation system according to claim 2, characterized in that: The failure knowledge model based on semantic understanding includes: The active learning training module receives pre-trained data from the FMEA analysis content knowledge base and brings it into the training module. It then determines the data confidence level, processes the data based on the confidence level, and trains the failure knowledge model. Furthermore, it optimizes the failure knowledge model through an adversarial training mechanism and updates the training data. Multi-source heterogeneous data parsing module, including OCR and PDF parsing module and table and chart recognition module; OCR and PDF parsing module is used to parse scanned documents using AI optical character recognition technology, combined with PDF text extraction technology, to extract text content from different data sources; table and chart recognition module is used to use computer vision models to identify and extract key information from chart content, and output this information content to subsequent modules; The semantic understanding and information filtering module is used to parse unstructured text using a pre-trained model, extract contextual semantic features to form relevant fault parameters, and determine whether the parsed fault information is invalid.

4. The intelligent FMEA analysis content generation system according to claim 2, characterized in that: The extended entity triple generation module includes: The data structuring and dynamic weight updating module receives the initially processed failure information from the entity recognition and relationship extraction module, converts the failure information into structured JSON format data using Python, and performs risk scoring. Alternatively, it generates entity association weights based on contextual semantic feature information through a neural network algorithm, adds them to the structured data, and passes them together to the FMEA analysis information expansion triple generation module. The FMEA analysis information extended triple generation module is used to receive user-defined triple relationships and mapping rules from the custom triple relationship module, and automatically generate extended triples including "head entity-relationship-tail entity-weight" by combining the entity association weights generated by the data structuring and dynamic weight update module to replace the traditional fixed-format triple structure. It also passes the candidate triple information generated by combining the language model to the FMEA information multi-dimensional verification module.

5. The intelligent FMEA analysis content generation system according to claim 4, characterized in that: The data update module includes: The FMEA information multi-dimensional verification module is used to automatically and / or manually review and judge the accuracy and scalability of candidate triples from multiple dimensions. After judgment, it outputs the judgment result for each candidate triple. If the triple meets the requirements, it outputs further instructions and outputs the triple information to the FMEA analysis content update module. If the analysis content does not meet the requirements, it terminates and proceeds to the next information judgment; The FMEA analysis content update module is used to receive the FMEA analysis information extension entity triples that meet the requirements after judgment, and update and add them to the established FMEA analysis knowledge base based on the ontology model and knowledge graph according to the pre-filled module information.

6. The intelligent FMEA analysis content generation system according to claim 5, characterized in that: The data export module includes: The data retrieval module is used to read the product information to be searched through semantic understanding of the big model, search for the extended triple information of the required components from the FMEA analysis content knowledge base, and pass the found information to the weak link identification module based on the big model; The large-model-based weak link identification module is used to identify system components and functions that need to be performed through a large-scale failure knowledge model based on semantic understanding. It then expands the weight information in the triples based on the retrieved components and uses a neural network algorithm to generate a criticality score for each failure information. The criticality scores are then ranked to identify the main weak links in the system. Finally, the analysis results are transmitted to the FMEA analysis report filling module. The FMEA analysis report filling module is used to fill the FMEA analysis information found and analyzed into the corresponding report template under the FMEA analysis specification, and pass the final results to the result export interaction module.

7. The intelligent FMEA analysis content generation system according to claim 6, characterized in that: Information interaction modules include The information input interaction module is used to provide an interactive interface, display the information range that needs to be captured, select the relevant failure information content that needs to be captured, input and select the FMEA analysis specification content to be adopted, provide a window for inputting failure information, and display system errors during the generation process; The result export interactive module is used to provide an interactive interface to display the final output FMEA analysis report content.

8. The intelligent FMEA analysis content generation system according to claim 7, characterized in that: The FMEA analysis content generation system also includes: The FMEA analysis specification module is connected to the information input interaction module, the custom triple relationship module and the FMEA analysis report filling module. It is used to manually select the FMEA analysis specification standard to be adopted and output the specification requirements to the custom triple relationship module and the FMEA analysis report filling module. The custom triple relationship module is connected to the FMEA analysis specification module, the FMEA analysis content knowledge base, and the extended entity triple generation module. It is used to receive possible special entities and entity relationships from the FMEA analysis specification requirements, automatically generate and / or manually input RDF triple mapping rules for special relationships applicable to a certain FMEA analysis specification, and construct directed edges in the FMEA analysis content knowledge base to store related content. Finally, it outputs the custom RDF triple mapping rules to the extended entity triple generation module.

9. An intelligent FMEA analysis content generation method, characterized in that: The method is implemented based on the intelligent FMEA analysis content generation system according to any one of claims 1 to 8.

10. The intelligent FMEA analysis content generation method according to claim 9, characterized in that: Including steps: Step S1. Establish FMEA analysis content knowledge base; Determine the requirements of the baseline FMEA analysis specifications to be selected, extract the relevant rules and relationships in the analysis specifications, combine the collected existing FMEA analysis data, and build an initial version of the FMEA analysis content knowledge base based on the ontology model and knowledge graph, storing FMEA analysis information in modules; Step S2. Identify and extract failure information using the failure knowledge model based on semantic understanding that has been pre-trained in the failure data capture module; Through the information input interaction module, users input the source files for the failure data to be captured and define the search scope. Then, using the pre-trained failure knowledge model based on semantic understanding, the required failure information is searched within the defined search scope, and failure data that meets the requirements is preliminarily screened. Entity recognition and relationship extraction models are then used to extract failure data information. Step S3. Using the extended entity triple generation module, the captured product failure information is converted into an extended entity triple of FMEA analysis information including weight information; The extended entity triple generation module structures and normalizes the captured product failure information, converting it into structured data and calculating the associated weights of the failure information. Then, according to the requirements of the selected FMEA analysis specification, a custom RDF triple mapping rule is generated that includes the special requirements of the analysis specification. Based on the processing rules, the processed structured data is converted into extended entity triples of FMEA analysis information in the form of "head entity-relationship-tail entity-weight". Step S4. Use the data update module to review the candidate triples and store the data that meets the requirements into the FMEA analysis content knowledge base; Step S5. Use the data export module to assign a criticality score to the failure mode, automatically identify design weaknesses, and generate an FMEA analysis report. Using large model technology to read system composition information and execution function information, the data export module searches for failure analysis content of relevant components and functions from the FMEA analysis content knowledge base. Combining the fault information and system composition, the neural network algorithm is used to assign a criticality score to the fault mode. Based on the score, the system automatically searches for design weaknesses and finally automatically generates an FMEA analysis report according to the FMEA analysis specification requirements. Step S6: Use the information interaction model to input and select the FMEA analysis specification content and requirements, display the final output report content, and complete the specified user interaction.

Images