A spaceflight fault diagnosis method based on generative intelligence

Abstract

The application discloses a space fault diagnosis method based on generative intelligence, and belongs to the technical field of space fault diagnosis.The space fault diagnosis method based on generative intelligence can automatically extract fault diagnosis knowledge in documents such as FMEA, preplan disposal manual and historical zeroing report and construct a space troubleshooting knowledge graph; when an abnormality occurs, a user can quickly retrieve fault knowledge and form a fault cause analysis tree in a manner of natural language description; after the fault cause analysis tree is confirmed, the system can automatically construct an intelligent agent to execute a fault cause troubleshooting process, assist in completing fault cause screening or positioning, support a natural language interactive mode to provide fault troubleshooting suggestions to the user, and automatically complete fault diagnosis report generation and knowledge graph updating after the fault cause is finally positioned.

Description

Technical Field

[0001] This invention relates to a method for diagnosing aerospace faults based on generative intelligence, belonging to the field of aerospace fault diagnosis technology. Background Technology

[0002] Rocket launch is a crucial milestone in space missions. Problems that occur during rocket launches, if not handled quickly, correctly, and promptly, can often have catastrophic consequences. As the complexity of the launch process increases, it becomes increasingly difficult to fully predict and assess problems within each stage and with other systems. The resulting impact can directly lead to mission failure or even catastrophic consequences.

[0003] To mitigate the adverse effects of minor disturbances on space launch missions, intelligent technologies are needed to assist users in troubleshooting abnormal situations. This can improve the robustness of rocket launches and reduce economic, time, and technical costs. However, during launch and testing, the launch vehicle systems consist of numerous components and exhibit complex failure modes. Fault diagnosis heavily relies on personnel experience, and the level of intelligence and automation in system fault diagnosis and handling needs improvement. Summary of the Invention

[0004] The technical problem solved by this invention is to overcome the shortcomings of the prior art and provide a generative intelligence-based aerospace fault diagnosis method to solve the problem of on-the-spot fault handling in the aerospace launch and testing process.

[0005] The technical solution of this invention is: a space fault diagnosis method based on generative intelligence, comprising:

[0006] The large language model is used to perform semantic parsing on structured and unstructured aerospace technology documents, extracting entities including equipment names, failure modes, failure mechanisms, troubleshooting procedures, and handling strategies, and constructing an initial knowledge graph framework. Then, the similarity between entities is calculated through graph neural networks. Combined with the contextual information of the entities, the identification and fusion of entities with the same name in cross documents is realized. A verification mechanism is used to eliminate semantic ambiguity caused by different documents, forming a fusion knowledge graph of equipment topology and failure propagation path.

[0007] A hybrid retrieval model framework integrating precise retrieval and fuzzy retrieval is constructed. The hybrid retrieval model framework and the fusion knowledge graph are used to retrieve the input description of the fault phenomenon. The precise retrieval is used to find directly matching fault phenomena and related causes in the knowledge graph. The semantic similarity-based fuzzy retrieval is used to find fault phenomena that match the user description and their associated causes, and a fault cause analysis tree is generated.

[0008] Based on the association between the bottom events in the fault cause analysis tree and the corresponding troubleshooting process, the process engine drives the relevant services to automatically execute the troubleshooting process, thereby completing the screening or location of the fault cause.

[0009] Furthermore, the structured and unstructured aerospace technical documents include FMEA analysis reports, launch and test procedure manuals, historical fault contingency plans, and quality zeroing reports.

[0010] Furthermore, the processing order and priority of the aerospace technical documents are determined by sorting them according to the fault diagnosis knowledge density and information richness.

[0011] Furthermore, the process of semantically parsing structured and unstructured aerospace technology documents includes preprocessing the aerospace technology documents, including word segmentation, removal of stop words, and standardization.

[0012] Furthermore, the identification includes:

[0013] Using the identified device name and fault mode as the central node, semantic information including actions, attributes, and states is extracted by combining the syntactic dependency relations in its context, and a local semantic subgraph is constructed; the local semantic subgraph includes candidate entities corresponding to the semantic information.

[0014] We design a heterogeneous neighbor aggregation strategy based on an attention mechanism. We construct a heterogeneous graph attention network on the basis of the constructed local semantic subgraph, perform weighted aggregation on different types of relation edges and neighbor nodes, and calculate the semantic similarity between candidate entities in all local semantic subgraphs based on the entity vectors formed by aggregation, thus forming a semantic similarity matrix.

[0015] The similarity matrix is ​​divided into multiple granularities using a spectral clustering algorithm. When the entity similarity is higher than a preset threshold, entity association is achieved, and recognition is completed.

[0016] Furthermore, the fusion includes:

[0017] Entity matching based on graph embedding: Using knowledge graph embedding methods, entities and relationships from different documents or local semantic subgraphs are jointly mapped to a unified continuous vector space.

[0018] Dynamic semantic vector construction: Design a dynamic semantic vector representation mechanism that integrates multiple features, including entity type constraints, key attributes, context-dependent syntactic paths, and neighborhood structure features. Through learnable weights, the entity is converted into a context-aware dynamic semantic vector in the continuous vector space.

[0019] Threshold determination and entity merging: Combine a predefined device domain dictionary to perform weighted processing on device domain characteristic terms and calculate the cosine similarity between dynamic semantic vectors; when the similarity is higher than the set threshold, it is determined that the corresponding entities are the same entity, triggering a merging operation, including attribute fusion, relationship integration and redundancy resolution, and finally realizing the alignment of homonymous or heteronymous synonymous entities across documents and across local semantic subgraphs, thus completing knowledge fusion.

[0020] Furthermore, the process engine-driven related services include:

[0021] Define the goals, functions, and workflow of the intelligent agent;

[0022] Analyze the bottom events in the fault cause analysis tree and establish a connection with the troubleshooting process in the fusion knowledge graph;

[0023] Based on the aforementioned workflow, a fault diagnosis intelligent agent is constructed, and a workflow engine is used to automatically call relevant tools and services to execute the troubleshooting process. When encountering steps that cannot be automatically executed, the intelligent agent interacts with the troubleshooting personnel to obtain fault information or perform specific operations.

[0024] The agent compares and analyzes the collected data and information with the fault cause analysis tree in real time to filter and locate the fault causes. When the troubleshooting process is completed, the agent determines the specific cause of the fault based on the collected information. If a unique cause cannot be determined, the agent updates the fault cause analysis based on the troubleshooting results.

[0025] Furthermore, the intelligent agent compiles the preset key information and final conclusions during the troubleshooting process into a report, including fault description, troubleshooting process, troubleshooting results and recommended measures; at the same time, the intelligent agent updates the troubleshooting experience and knowledge into the fusion knowledge graph in real time.

[0026] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the aerospace fault diagnosis method based on generative intelligence.

[0027] A generative intelligence-based aerospace fault diagnosis device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: when the processor executes the computer program, it implements the steps of the generative intelligence-based aerospace fault diagnosis method.

[0028] The advantages of this invention compared to the prior art are:

[0029] The aerospace fault diagnosis technology based on generative intelligence provided in this invention can automatically extract fault diagnosis knowledge from documents such as FMEA, contingency plan manuals, and historical zeroing reports, and construct an aerospace troubleshooting knowledge graph. When an anomaly occurs, users can quickly retrieve fault knowledge through natural language descriptions and associate it to form a fault cause analysis tree. After confirming the fault cause analysis tree, the system can automatically construct an intelligent agent to execute the fault cause investigation process, assist in completing the fault cause screening or location, and support providing fault investigation suggestions to users in a natural language interaction mode. After finally locating the fault cause, the system can automatically generate a fault diagnosis report and update the knowledge graph. Attached Figure Description

[0030] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0031] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0032] To better understand the above technical solutions, the technical solutions of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solutions of the present invention, rather than limitations on the technical solutions of the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0033] The following description, in conjunction with the accompanying drawings, provides a more detailed explanation of an aerospace fault diagnosis method based on generative intelligence provided by an embodiment of the present invention. Figure 1 Specific implementation methods may include:

[0034] The large language model is used to perform semantic parsing on structured and unstructured aerospace technology documents, extracting entities including equipment names, failure modes, failure mechanisms, troubleshooting procedures, and handling strategies, and constructing an initial knowledge graph framework. Then, the similarity between entities is calculated through graph neural networks. Combined with the contextual information of the entities, the identification and fusion of entities with the same name in cross documents is realized. A verification mechanism is used to eliminate semantic ambiguity caused by different documents, forming a fusion knowledge graph of equipment topology and failure propagation path.

[0035] A hybrid retrieval model framework integrating precise retrieval and fuzzy retrieval is constructed. The hybrid retrieval model framework and the fusion knowledge graph are used to retrieve the input description of the fault phenomenon. The precise retrieval is used to find directly matching fault phenomena and related causes in the knowledge graph. The semantic similarity-based fuzzy retrieval is used to find fault phenomena that match the user description and their associated causes, and a fault cause analysis tree is generated.

[0036] Based on the association between the bottom events in the fault cause analysis tree and the corresponding troubleshooting process, the process engine drives the relevant services to automatically execute the troubleshooting process, thereby completing the screening or location of the fault cause.

[0037] The solution provided in the embodiments of the present invention specifically includes:

[0038] I. Knowledge Graph Construction Methods

[0039] Step 1: Extracting multimodal information based on a large model of aerospace vertical domain.

[0040] (1) Obtaining original design documents

[0041] The original design documents include: FMEA analysis report (structured tables), test and development process manual (semi-structured text), historical fault contingency plans (unstructured text), and quality zeroing report (hybrid document). After obtaining the design documents, they are sorted according to the fault diagnosis knowledge density and information richness to determine the processing order and priority.

[0042] (2) Document preprocessing

[0043] Document preprocessing is performed, with different preprocessing methods designed for different document types, including but not limited to word segmentation, stop word removal, and standardization, to facilitate subsequent deep semantic analysis. Specific content includes: ① Non-text data: For image information in the document, OCR technology is used to extract text from the images, or a multimodal large model is used to represent the main content of the images in text form; ② Semi-structured text data: Regular expressions are used to parse semi-structured content such as tables and numbered lists; ③ Unstructured text data: Automated tools are used to segment the text content; ④ Domain adaptation: Based on an aerospace terminology database (such as equipment models and professional abbreviations), the original text is standardized and replaced to reduce semantic ambiguity.

[0044] (3) Deep semantic parsing

[0045] A pre-trained aerospace vertical domain large model is used to perform deep semantic parsing on documents and extract key information, mainly including five types of entities: equipment name, failure mode, failure mechanism, troubleshooting process, and handling strategy, forming an aerospace troubleshooting knowledge graph. The specific steps are as follows: ① Entity recognition: The large model is used to label the five types of entities: equipment name, failure mode, failure mechanism, troubleshooting process, and handling strategy; ② Relationship extraction: Implicit relationships between entities are captured based on a multi-head attention mechanism.

[0046] Step 2: Entity disambiguation and fusion based on graph embedding

[0047] (1) Entity recognition and disambiguation

[0048] This paper utilizes graph neural networks to calculate the similarity between entities, achieving correct identification and disambiguation of entities with the same name across documents based on their contextual information. For entities with the same name but potentially representing different entities, ambiguity is eliminated through contextual information and relationships between entities. The specific process includes: ① Graph structure construction: Using device names and fault modes as core nodes, a semantic association graph is constructed by combining verb syntactic relationships in the context to accurately represent the multi-dimensional features of entities; ② Dynamic similarity calculation: A heterogeneous neighbor aggregation strategy based on an attention mechanism is designed to calculate the semantic similarity of candidate entities on the knowledge graph subgraph; ③ Disambiguation decision mechanism: A spectral clustering algorithm is used to perform multi-granularity partitioning of the similarity matrix, and entity association is achieved when the entity similarity is higher than a threshold.

[0049] (2) Entity integration

[0050] Based on ensuring correct entity identification, entity information from different documents is integrated to achieve semantic unification and conflict resolution of multi-source heterogeneous data, forming a unified knowledge graph. A verification mechanism eliminates semantic ambiguity caused by different documents, ensuring the accuracy and consistency of the knowledge graph. Entity matching based on graph embedding uses the TransE algorithm to map entities in heterogeneous documents to a unified vector space, constructing dynamic semantic vectors through entity type, attribute features, and contextual dependencies. Considering the characteristics of the device domain, a domain-dictionary-weighted cosine similarity calculation model is designed. When the similarity value exceeds a threshold, entity association is triggered, achieving cross-document entity alignment.

[0051] II. Knowledge Graph Retrieval Methods

[0052] Step 1: Construct a hybrid retrieval model framework. The purpose of this step is to integrate the advantages of precise and fuzzy retrieval to ensure a comprehensive understanding of the user's description of the fault. Constructing a hybrid retrieval model framework requires implementing the following key steps:

[0053] (1) User input preprocessing: The description of the fault provided by the user is preliminarily processed, including word segmentation, part-of-speech tagging and semantic parsing, to ensure the accuracy and comprehensiveness of subsequent retrieval.

[0054] (2) Retrieval Model Design: Design a retrieval model that integrates precise matching technology and fuzzy matching technology based on semantic similarity, so that the model can extract relevant information from the knowledge graph based on the fault description input by the user.

[0055] Step 2: Hybrid retrieval based on user fault description. This step utilizes the hybrid retrieval model framework built in Step 1 to perform the actual retrieval operation. The specific process includes:

[0056] (1) Exact matching: First, try to find the fault phenomena and related causes that are directly matched in the knowledge graph through exact matching technology.

[0057] (2) Fuzzy matching: When the results of precise matching are insufficient to meet the requirements, fuzzy matching technology based on semantic similarity is used to find fault phenomena and their associated causes that are similar to the user's description, so as to generate a comprehensive and accurate fault cause analysis tree.

[0058] (3) Results integration: The results obtained from exact matching and fuzzy matching are integrated to form a fault cause analysis tree that users can understand, helping users understand the potential causes of the fault phenomenon.

[0059] Step 3: Results Presentation and Optimization. This step aims to ensure that search results are clearly communicated to users and to continuously optimize the search model based on user feedback. Specific steps include:

[0060] (1) Results display: The fault cause analysis tree is displayed to the user in graphical or list form to help the user understand the fault phenomenon and its possible causes.

[0061] (2) User feedback collection: Evaluation of search results, including accuracy and completeness, is collected through user feedback and serves as an important basis for optimizing the search model.

[0062] (3) Model optimization: Based on user feedback, continuously adjust and optimize the retrieval model to improve the accuracy and efficiency of exact matching and fuzzy matching, so as to better serve user needs.

[0063] III. Fault Diagnosis Intelligent Agent Construction Method

[0064] Step 1: Design the infrastructure for the fault diagnosis agent. This step includes defining the agent's goals, functions, and workflow to ensure it can effectively diagnose faults.

[0065] Step 2: Parsing and Linking Base Events in the Fault Cause Analysis Tree with Troubleshooting Processes. In this stage, the agent needs to parse the base events in the fault cause analysis tree and link them with the troubleshooting processes in the knowledge graph. These processes typically include detailed troubleshooting steps, required tools, checkpoints, and other information. Through interaction with the knowledge graph, the agent can obtain all possible troubleshooting processes related to the base events, laying the foundation for subsequent automated execution.

[0066] Step 3: Intelligent Agent Construction and Troubleshooting Process Execution and Feedback. In this step, a fault diagnosis intelligent agent is built based on the workflow, and the process engine automatically calls relevant tools and services to execute the fault diagnosis process, such as equipment status monitoring, data collection and analysis. When encountering steps that cannot be automatically executed, the intelligent agent will interact with troubleshooting personnel through natural language processing technology to obtain necessary fault information or perform specific operations.

[0067] Step 4: Fault Cause Screening and Localization. As the troubleshooting process is executed step by step, the agent compares and analyzes the collected data and information with the fault cause analysis tree. Using a data-driven approach, it gradually eliminates impossible causes, thus screening and localizing the fault cause. Once the troubleshooting process is complete, the agent will determine the specific cause of the fault based on the collected information. If a unique cause cannot be determined, the agent will update the fault cause analysis tree based on the troubleshooting results, providing more precise guidance for subsequent troubleshooting.

[0068] Step 5: Report Generation and Knowledge Update. The agent compiles key information and final conclusions from the entire troubleshooting process into a report, including a fault description, troubleshooting process, results, and recommended measures. Furthermore, the agent updates the knowledge gained from this troubleshooting into the knowledge graph, providing a reference for future fault diagnosis. This step ensures the agent can continuously learn and improve, enhancing the efficiency and accuracy of fault diagnosis.

[0069] This invention provides a computer-readable storage medium storing computer instructions that, when executed on a computer, cause the computer to perform... Figure 1 The method described.

[0070] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0071] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0072] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0073] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0074] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0075] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. A method for diagnosing aerospace faults based on generative intelligence, characterized in that, include: The large language model is used to perform semantic parsing on structured and unstructured aerospace technology documents, extracting entities including equipment names, failure modes, failure mechanisms, troubleshooting procedures, and handling strategies, and constructing an initial knowledge graph framework. Then, the similarity between entities is calculated through graph neural networks. Combined with the contextual information of the entities, the identification and fusion of entities with the same name in cross documents is realized. A verification mechanism is used to eliminate semantic ambiguity caused by different documents, forming a fusion knowledge graph of equipment topology and failure propagation path. A hybrid retrieval model framework integrating precise retrieval and fuzzy retrieval is constructed. The hybrid retrieval model framework and the fusion knowledge graph are used to retrieve the input description of the fault phenomenon. The precise retrieval is used to find directly matching fault phenomena and related causes in the knowledge graph. The semantic similarity-based fuzzy retrieval is used to find fault phenomena that match the user description and their associated causes, and a fault cause analysis tree is generated. Based on the association between the bottom events in the fault cause analysis tree and the corresponding troubleshooting process, the process engine drives the relevant services to automatically execute the troubleshooting process, thereby completing the screening or location of the fault cause.

2. The aerospace fault diagnosis method based on generative intelligence according to claim 1, characterized in that, The structured and unstructured aerospace technical documents include FMEA analysis reports, launch and test procedure manuals, historical fault contingency plans, and quality zeroing reports.

3. The aerospace fault diagnosis method based on generative intelligence according to claim 2, characterized in that, The processing order and priority of the aerospace technical documents are determined by sorting them according to the knowledge density and information richness of their fault diagnosis.

4. The aerospace fault diagnosis method based on generative intelligence according to claim 1, characterized in that, The process of semantically parsing structured and unstructured aerospace technology documents includes preprocessing of the aerospace technology documents, including word segmentation, removal of stop words, and standardization.

5. The aerospace fault diagnosis method based on generative intelligence according to claim 1, characterized in that, The identification includes: Using the identified device name and fault mode as the central node, semantic information including actions, attributes, and states is extracted by combining the syntactic dependency relations in its context, and a local semantic subgraph is constructed; the local semantic subgraph includes candidate entities corresponding to the semantic information. We design a heterogeneous neighbor aggregation strategy based on an attention mechanism. We construct a heterogeneous graph attention network on the basis of the constructed local semantic subgraph, perform weighted aggregation on different types of relation edges and neighbor nodes, and calculate the semantic similarity between candidate entities in all local semantic subgraphs based on the entity vectors formed by aggregation, thus forming a semantic similarity matrix. The similarity matrix is ​​divided into multiple granularities using a spectral clustering algorithm. When the entity similarity is higher than a preset threshold, entity association is achieved, and recognition is completed.

6. The aerospace fault diagnosis method based on generative intelligence according to claim 5, characterized in that, The fusion includes: Entity matching based on graph embedding: Using knowledge graph embedding methods, entities and relationships from different documents or local semantic subgraphs are jointly mapped to a unified continuous vector space. Dynamic semantic vector construction: Design a dynamic semantic vector representation mechanism that integrates multiple features, including entity type constraints, key attributes, context-dependent syntactic paths, and neighborhood structure features. Through learnable weights, the entity is converted into a context-aware dynamic semantic vector in the continuous vector space. Threshold determination and entity merging: Combine a predefined device domain dictionary to perform weighted processing on device domain characteristic terms and calculate the cosine similarity between dynamic semantic vectors; when the similarity is higher than the set threshold, it is determined that the corresponding entities are the same entity, triggering a merging operation, including attribute fusion, relationship integration and redundancy resolution, and finally realizing the alignment of homonymous or heteronymous synonymous entities across documents and across local semantic subgraphs, thus completing knowledge fusion.

7. The aerospace fault diagnosis method based on generative intelligence according to claim 1, characterized in that, The process engine-driven services include: Define the goals, functions, and workflow of the intelligent agent; Analyze the bottom events in the fault cause analysis tree and establish a connection with the troubleshooting process in the fusion knowledge graph; Based on the aforementioned workflow, a fault diagnosis intelligent agent is constructed, and a workflow engine is used to automatically call relevant tools and services to execute the troubleshooting process. When encountering steps that cannot be automatically executed, the intelligent agent interacts with the troubleshooting personnel to obtain fault information or perform specific operations. The agent compares and analyzes the collected data and information with the fault cause analysis tree in real time to filter and locate the fault causes. When the troubleshooting process is completed, the agent determines the specific cause of the fault based on the collected information. If a unique cause cannot be determined, the agent updates the fault cause analysis based on the troubleshooting results.

8. The aerospace fault diagnosis method based on generative intelligence according to claim 7, characterized in that, The intelligent agent compiles the preset key information and final conclusions during the troubleshooting process into a report, including fault description, troubleshooting process, troubleshooting results and recommended measures; at the same time, the intelligent agent updates the troubleshooting experience and knowledge into the fusion knowledge graph in real time.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 8.

10. An aerospace fault diagnosis device based on generative intelligence, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 8.

Images