Intelligent Generation Method and Device for Equipment Fault Behavior Logic Model Based on LLM

Abstract

This application relates to the interdisciplinary field of artificial intelligence and reliability systems engineering, and provides a method and apparatus for intelligent generation of equipment fault behavior logic models based on LLM (Limited Linear Modeling). This method combines a vector database to achieve enhanced retrieval, constructs an agent-based framework for generating equipment fault behavior logic models, and then generates structural elements in the agent for each business component involved in the equipment fault behavior logic. Based on the model structure, the results are integrated and merged, ultimately achieving a complete and runnable equipment fault behavior logic simulation model. This method integrates the interactive modeling capabilities of agents with the semantic understanding and logical reasoning capabilities of large language models, enabling intelligent generation of fault behavior models for complex equipment and effectively improving the generation efficiency of equipment fault behavior models.

Description

Technical Field

[0001] This application relates to the interdisciplinary field of artificial intelligence and reliability systems engineering, and in particular to an intelligent generation method and device for equipment fault behavior logic model based on LLM. Background Technology

[0002] The equipment failure behavior logic model is a crucial step in equipment usage simulation, performance evaluation, and verification. Traditionally, this process has relied heavily on manual work by modeling experts. For example, the equipment's failure state machine or behavioral logic framework diagram can be defined in Model-Based Systems Engineering (MBSE) tools or simulation platforms, and the model code can be written and debugged. This approach suffers from inefficiency, long development cycles, and difficulty in quickly responding to rapidly changing equipment usage scenarios and equipment iteration requirements.

[0003] With the development of artificial intelligence technology, large language models (LLMs) have shown great potential in natural language understanding and logical reasoning, providing new ideas for the intelligent generation of models. However, directly using LLMs to generate equipment fault behavior logic models can easily lead to "illusion" phenomena, generating erroneous results that do not conform to the physical constraints of the equipment, and lacking a deep understanding of the equipment domain. Furthermore, the generated model components are logically loosely connected, making it difficult to integrate them into a complete and workable simulation model, and compromising the consistency and completeness of the equipment fault behavior logic.

[0004] Therefore, there is an urgent need for an intelligent model construction method that can deeply integrate knowledge of equipment domains, natural semantic understanding, and logical reasoning capabilities. Summary of the Invention

[0005] In view of this, embodiments of this application provide an intelligent generation method and apparatus for equipment fault behavior logic models based on LLM, in order to solve the problems of low generation efficiency of equipment fault behavior logic models in the prior art and the difficulty in accurately and completely matching equipment fault behaviors with the generated models.

[0006] A first aspect of this application provides an intelligent generation method for an equipment fault behavior logic model based on LLM, comprising:

[0007] A logical model of the equipment is constructed using the Model-Based Systems Engineering (MBSE) method, and retrieval-enhanced knowledge is injected into the logical model to obtain an enhanced logical model.

[0008] Using LLM, equipment-type intelligent agents, task-type intelligent agents, and maintenance support resource-type intelligent agents are constructed based on the enhanced logical model and equipment task scenarios. Attribute-layer features and behavioral-layer features are defined for each entity unit in each intelligent agent. Attribute-layer features are used to characterize the static features of the entity unit, and behavioral-layer features are used to characterize the dynamic interaction logic of the entity unit.

[0009] By utilizing LLM to construct the basic information of each entity unit in the equipment-type intelligent agent and the relationship between entity units based on the attribute layer features and behavior layer features of each entity unit, an equipment composition model is obtained.

[0010] Based on the equipment composition model, the basic information of each entity unit in the task-type intelligent agent and the relationship between entity units are constructed using the attribute layer features and behavior layer features of each entity unit in the task-type intelligent agent. The supplementary dynamic interaction logic between each equipment-type intelligent agent and the corresponding task-type intelligent agent is constructed using the behavior layer features of each entity unit in each equipment-type intelligent agent. The supplementary dynamic interaction logic between each maintenance and support resource-type intelligent agent and the corresponding task-type intelligent agent is constructed using the behavior layer features of each entity unit in each maintenance and support resource-type intelligent agent. Thus, the task scenario model is obtained.

[0011] Based on the task scenario model, supplementary static fault information and supplementary dynamic fault behavior of equipment-type intelligent agents are constructed using LLM based on the attribute layer features and behavioral layer features of each entity unit in the equipment-type intelligent agent. Similarly, supplementary static fault information and supplementary dynamic fault behavior of each maintenance and support resource-type intelligent agent are constructed using LLM based on the attribute layer features and behavioral layer features of each entity unit in the maintenance and support resource-type intelligent agent, thus obtaining the equipment fault behavior logic model.

[0012] A second aspect of this application provides an intelligent generation device for an equipment fault behavior logic model based on LLM, comprising:

[0013] The logical model building module is configured to build a logical model of the equipment using the model-based systems engineering (MBSE) method, and inject retrieval-enhanced knowledge into the logical model to obtain an enhanced logical model.

[0014] The definition module is configured to use LLM to construct equipment-type intelligent agents, task-type intelligent agents, and maintenance support resource-type intelligent agents based on the enhanced logical model and equipment task scenarios, and to define attribute layer features and behavioral layer features for each entity unit in each intelligent agent; attribute layer features are used to characterize the static features of entity units, and behavioral layer features are used to characterize the dynamic interaction logic of entity units.

[0015] The equipment composition model construction module is configured to use LLM to construct the basic information of each entity unit in the equipment-class intelligent agent and the relationship between entity units based on the attribute layer features and behavior layer features of each entity unit in the equipment-class intelligent agent, so as to obtain the equipment composition model.

[0016] The task scenario model construction module is configured to, based on the equipment composition model, use LLM to construct the basic information of each entity unit in the task-type intelligent agent and the relationship between entity units based on the attribute layer features and behavioral layer features of each entity unit in the task-type intelligent agent; use LLM to construct the supplementary dynamic interaction logic between each equipment-type intelligent agent and the corresponding task-type intelligent agent based on the behavioral layer features of entity units in each equipment-type intelligent agent; and use LLM to construct the supplementary dynamic interaction logic between each maintenance and support resource-type intelligent agent and the corresponding task-type intelligent agent based on the behavioral layer features of entity units in each maintenance and support resource-type intelligent agent, thereby obtaining the task scenario model;

[0017] The equipment failure behavior logic model construction module is configured to, based on the task scenario model, use LLM to construct supplementary static failure information and supplementary dynamic failure behavior of equipment-type intelligent agents based on the attribute layer features and behavioral layer features of each entity unit in the equipment-type intelligent agent, and use LLM to construct supplementary static failure information and supplementary dynamic failure behavior of each maintenance and support resource-type intelligent agent based on the attribute layer features and behavioral layer features of each entity unit in the maintenance and support resource-type intelligent agent, thereby obtaining the equipment failure behavior logic model.

[0018] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0019] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0020] The beneficial effects of the embodiments in this application compared with the prior art are:

[0021] This application's embodiments combine vector databases to achieve enhanced retrieval, construct an agent-based framework for generating equipment fault behavior logic models, and then generate structural elements in the agent for each business component involved in the equipment fault behavior logic. Based on the model structure, the results are integrated and merged, ultimately achieving a complete and runnable equipment fault behavior logic simulation model. This method integrates the interactive modeling capabilities of agents with the semantic understanding and logical reasoning capabilities of large language models, enabling intelligent generation of fault behavior models for complex equipment and effectively improving the generation efficiency of equipment fault behavior models. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating an intelligent generation method for an equipment fault behavior logic model based on LLM, provided in an embodiment of this application.

[0024] Figure 2 This is an example of a Markdown-formatted text illustration.

[0025] Figure 3 This is a flowchart illustrating another method for intelligent generation of equipment fault behavior logic model based on LLM provided in this application embodiment.

[0026] Figure 4 This is a schematic diagram of an intelligent generation device for equipment fault behavior logic model based on LLM provided in an embodiment of this application.

[0027] Figure 5 This is a schematic diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0028] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0029] The following describes in detail, with reference to the accompanying drawings, an intelligent generation method and apparatus for an equipment fault behavior logic model based on LLM according to an embodiment of this application.

[0030] As mentioned above, traditional equipment failure behavior logic models rely on manual work by experts, which is inefficient and has a long development cycle. If LLM is directly introduced into the construction of equipment failure behavior logic models, the resulting models usually differ significantly from the actual equipment failure logic because LLM lacks a deep understanding of equipment domain knowledge.

[0031] In view of this, this application provides an intelligent generation method for equipment fault behavior logic model based on LLM, which combines the dynamic interaction advantages of agent modeling with the logical reasoning ability of large language model to achieve efficient and intelligent generation of equipment fault behavior logic model.

[0032] Figure 1 This is a flowchart illustrating an intelligent generation method for an equipment fault behavior logic model based on LLM, provided in an embodiment of this application. Figure 1 As shown, the method includes the following steps:

[0033] In step S101, the logical model of the equipment is constructed using the MBSE method, and retrieval enhancement knowledge is injected into the logical model to obtain the enhanced logical model.

[0034] In step S102, LLM is used to construct equipment-type intelligent agents, task-type intelligent agents, and maintenance support resource-type intelligent agents based on the enhanced logical model and equipment task scenario, and attribute layer features and behavior layer features are defined for each entity unit in each intelligent agent.

[0035] Among them, attribute layer features are used to characterize the static features of entity units, and behavior layer features are used to characterize the dynamic interaction logic of entity units.

[0036] In step S103, LLM is used to construct the basic information of each entity unit in the equipment-type intelligent agent and the relationship between entity units based on the attribute layer features and behavior layer features of each entity unit in the equipment-type intelligent agent, so as to obtain the equipment composition model.

[0037] In step S104, based on the equipment composition model, the basic information of each entity unit in the task-type intelligent agent and the relationship between entity units are constructed using LLM based on the attribute layer features and behavior layer features of each entity unit in the task-type intelligent agent. The supplementary dynamic interaction logic between each equipment-type intelligent agent and the corresponding task-type intelligent agent is constructed using LLM based on the behavior layer features of entity units in each equipment-type intelligent agent. The supplementary dynamic interaction logic between each maintenance and support resource-type intelligent agent and the corresponding task-type intelligent agent is constructed using LLM based on the behavior layer features of entity units in each maintenance and support resource-type intelligent agent, thus obtaining the task scenario model.

[0038] In step S105, based on the task scenario model, supplementary static fault information and supplementary dynamic fault behavior of the equipment-type intelligent agent are constructed using LLM based on the attribute layer features and behavior layer features of each entity unit in the equipment-type intelligent agent. Similarly, supplementary static fault information and supplementary dynamic fault behavior of each maintenance and support resource-type intelligent agent are constructed using LLM based on the attribute layer features and behavior layer features of each entity unit in the maintenance and support resource-type intelligent agent, thereby obtaining the equipment fault behavior logic model.

[0039] In some embodiments of this application, the method may be executed by a server or by a terminal device with certain processing capabilities.

[0040] In some embodiments of this application, the MBSE method can be used to first construct a logical model of the analysis object, and then retrieval enhancement knowledge can be injected into the logical model to obtain an enhanced logical model.

[0041] In some embodiments of this application, LLM can also be used to construct equipment-type intelligent agents, task-type intelligent agents, and maintenance and support resource-type intelligent agents based on the enhanced logical model and equipment task scenario, and attribute layer features and behavioral layer features can be defined for each entity unit in each intelligent agent.

[0042] In some embodiments of this application, LLM can be used to construct the basic information of each entity unit in the equipment-type intelligent agent and the relationship between entity units based on the attribute layer features and behavior layer features of each entity unit in the equipment-type intelligent agent, so as to obtain the equipment composition model.

[0043] Furthermore, based on the equipment composition model, the basic information of each entity unit in the task-type intelligent agent and the relationship between entity units can be constructed using LLM based on the attribute layer features and behavioral layer features of each entity unit in the task-type intelligent agent. The supplementary dynamic interaction logic between each equipment-type intelligent agent and the corresponding task-type intelligent agent can also be constructed using LLM based on the behavioral layer features of entity units in each equipment-type intelligent agent. Finally, the supplementary dynamic interaction logic between each maintenance and support resource-type intelligent agent and the corresponding task-type intelligent agent can be constructed using LLM based on the behavioral layer features of entity units in each maintenance and support resource-type intelligent agent, thus obtaining the task scenario model.

[0044] Finally, based on the task scenario model, supplementary static fault information and supplementary dynamic fault behavior of equipment-type intelligent agents are constructed using LLM based on the attribute layer features and behavioral layer features of each entity unit in the equipment-type intelligent agent. Similarly, supplementary static fault information and supplementary dynamic fault behavior of each maintenance and support resource-type intelligent agent are constructed using LLM based on the attribute layer features and behavioral layer features of each entity unit in the maintenance and support resource-type intelligent agent, thus obtaining the equipment fault behavior logic model.

[0045] According to the technical solution provided in this application, enhanced retrieval is achieved by combining a vector database, and a framework for generating equipment fault behavior logic models based on intelligent agents is constructed. Then, for each business component involved in the equipment fault behavior logic, structural elements in the intelligent agent are generated one by one, and the results are integrated and merged based on the model structure, ultimately realizing a complete and runnable equipment fault behavior logic simulation model. This method integrates the interactive modeling capabilities of intelligent agents with the semantic understanding and logical reasoning capabilities of large language models, realizing the intelligent generation of fault behavior models for complex equipment and effectively improving the generation efficiency of equipment fault behavior models.

[0046] In some embodiments of this application, constructing the logical model of the analysis object using the MBSE method can be done by assuming the overall analysis object system is as follows: The MBSE method is used to divide the analyzed object into multiple logical levels based on its actual hardware architecture, denoted as... , of which each layer It includes a set of components that make up this layer. .in this way, It can represent the first The first in the layer Each component, also known as a physical unit, For the first The total number of components in the layer, and It is a positive integer greater than 1. It is a positive integer greater than 0.

[0047] In some embodiments of this application, injecting retrieval enhancement knowledge into the logical model can be achieved by, based on the equipment's mission scenario, converting non-handwritten format (such as Excel, Word, XML, etc.) documents (e.g., equipment detailed design reports, equipment reliability analysis reports, equipment maintainability analysis reports, etc.) into a structured intermediate text format using an Optical Character Recognition (OCR) engine; subsequently, segmenting the document by physical pages or logical chapters, and extracting semantic tags such as heading levels, paragraph boundaries, and list items based on document layout analysis, ultimately outputting a unified structured intermediate representation format—a Markdown string with semantic tags. This format preserves the original document's logical structure, facilitating subsequent semantic slicing. An exemplary Markdown format text is as follows: Figure 2 As shown.

[0048] To improve the semantic integrity and contextual coherence of knowledge fragments, a sliding window-driven dynamic aggregation slicing mechanism can be used to convert a preset format string into a semantically coherent set of knowledge slices. The specific operation is as follows:

[0049] The Markdown text above is initially segmented according to natural sentence boundaries to obtain a sentence sequence. ; Utilizing pre-trained embedding models (For example, Yinka) will process each statement Mapped to dense vectors Initialize the content of the current slice. ,vector ; for subsequent statements (in Calculate its cosine similarity to the current slice vector:

[0050] ;

[0051] like ,in If the preset similarity threshold is used, then... Merge into the current slice: Otherwise, the current slice Stored as a complete knowledge unit in the slice set, and... Start a new slice from the beginning. This strategy generates a set of semantically cohesive, length-adaptive knowledge slices. and its corresponding set of embedding vectors .

[0052] The obtained knowledge slice set and its embedding vectors can be imported into an efficient vector database (such as FAISS) to build a semantic index that can be retrieved quickly. Subsequently, for each component in the logical model... (in The following knowledge injection process will be executed:

[0053] (1) Construct structured query suggestions based on component identifiers, functional descriptions and interface features. ;

[0054] (2) Using the same embedding model right Encode the query vector;

[0055] (3) Perform an approximate nearest neighbor search in the vector database to recall and match the vectors. Most similar front A slice of knowledge;

[0056] (4) Introduce a cross-coding re-ranking model (such as bce-reranker) to perform refined relevance scoring on the recall results, and select the top scorers. Each slice serves as final contextual evidence. ;

[0057] (5) will be with Original query prompts The words are concatenated to form enhanced prompts, which are used to guide the large language model in subsequent analysis.

[0058] (6) Repeat the above process until knowledge injection of all components is completed.

[0059] Taking the first paragraph of the equipment mission support simulation document as an example, the original text reads: "The simulation scenario simulates the mission execution and fault repair of 10 aircraft over a year. Each aircraft has 4 subsystems, and each subsystem consists of 5 Line Replaceable Units (LRUs). Each LRU will randomly fail during operation according to an exponential probability distribution. If a fault occurs while the aircraft is performing its daily mission, the current mission will be interrupted, and a maintenance task will be performed. There are 10 maintenance personnel on site who can perform fault repairs on the aircraft. The working time, sorties, number of faults, and various maintenance information of each aircraft are recorded to evaluate the overall equipment availability and support effectiveness."

[0060] First, the text can be segmented into 5 sentences based on natural sentence boundaries (periods). Then, a pre-trained embedding model, Yinka, is used to map each sentence into a 768-dimensional dense vector. A similarity threshold τ = 0.65 is set, and a sliding window dynamic aggregation strategy is adopted: the current slice is initialized as sentence 1, and the cosine similarity between subsequent sentences and the average vector of the current slice is calculated sequentially. The results show that sentences 2 (0.73 similarity), 3 (0.70), and 4 (0.67) are all above the threshold and are merged into the current slice; sentence 5 has a similarity of 0.61, below the threshold, triggering the generation of a new slice. Ultimately, two semantically cohesive knowledge slices are formed:

[0061] Slice 1: "The simulation scenario simulates the mission execution and fault repair of 10 aircraft over a year. Each aircraft has 4 subsystems, and each subsystem consists of 5 LRUs. During operation, each LRU will randomly fail according to an exponential probability distribution. If a fault occurs while the aircraft is performing its daily mission, the current mission will be interrupted, and a maintenance mission will be performed. There are 10 maintenance personnel on site who can perform fault repairs on the aircraft."

[0062] Segment 2: "Record the working hours, sorties, number of malfunctions, and various maintenance information for each aircraft to assess the overall equipment availability and support effectiveness."

[0063] Each slice is assigned its corresponding embedding vector, supporting efficient semantic retrieval later.

[0064] Furthermore, taking the "LRU" component as an example, based on its identifier "LRU-A01", functional description "execution subsystem core functions", and interface characteristics "power input, signal output", a structured query prompt is constructed: "fault distribution characteristics, task interruption mechanisms, and maintenance resource configuration requirements related to LRU". This prompt, after being encoded using the Yinka embedding model, undergoes an approximate nearest neighbor search in the FAISS vector database to recall the Top-5 knowledge slices. Then, it is re-ranked using the bce-reranker cross-coding model, selecting the Top-2 slices with the highest scores, containing: "Each LRU experiences random failures during operation according to an exponential probability distribution" and "There are 10 maintenance personnel on site who can perform fault repairs on the aircraft". These slices are then concatenated with the original query to form an enhanced prompt word input large language model. Based on this, the model generates the following analysis conclusion: "According to the simulation settings, the failure rate of LRUs follows an exponential distribution, indicating that their failure rate is constant and there is no aging effect. Once any LRU fails, it will cause the failure of its subsystem, thereby interrupting the aircraft's current mission and triggering the maintenance process. Since there are only 10 maintenance personnel on site, the maintenance queuing effect under multiple aircraft concurrent failures needs to be considered."

[0065] In some embodiments of this application, the functional requirements and business logic of the simulation scenario of the equipment can be analyzed based on the mission scenario of the equipment, and an equipment-type intelligent agent can be constructed. Task-based intelligent agents and maintenance and maintenance resource-type intelligent agents This allows for the description of various objects involved in the logical model of equipment failure behavior. Among them, equipment-type intelligent agents... Used to characterize equipment entities and their constituent entity units; task-oriented intelligent agents Used to characterize the initiation, execution, and management of various tasks; intelligent agents for maintenance and support resources. Used to characterize maintenance support elements such as maintenance personnel, spare parts, and staff. Indicates the first intelligent agents, Represents the first in the equipment class of intelligent agents Each entity unit, and so on.

[0066] Taking an airport equipment mission support simulation as an example, this simulation scenario simulates the mission execution and fault maintenance of 10 aircraft over a year. Each aircraft has 4 subsystems, and each subsystem consists of 5 LRUs. Each LRU experiences a random failure during operation according to an exponential probability distribution. If a failure occurs during routine missions, the current mission is interrupted, and maintenance tasks are performed. There are 10 maintenance personnel on-site who can perform fault repairs on the aircraft. The working time, sorties, number of failures, and various maintenance information for each aircraft are recorded to evaluate the overall equipment availability and support effectiveness.

[0067] Based on the above method, three types of simulated intelligent agents were constructed according to the simulation scenario:

[0068] Equipment-type intelligent agents, Includes equipment intelligent agents j={1,2,3…,250};

[0069] Task-oriented intelligent agents Includes task-oriented intelligent agents j={1,2};

[0070] Maintenance and support resource-type intelligent agents Including intelligent entities for maintenance and support resources , j={1,2,3...,10}.

[0071] Various intelligent agents can be analyzed one by one, targeting the first... In intelligent agents Entity unit Design and define a two-layer structure as follows:

[0072] Attribute layer: Describes the static characteristics of the agent (identifier, name, parameters, variables, functions, descriptions, etc.), defined as: .

[0073] Behavioral layer: describes the dynamic interaction logic of the agent (state transition rules, event response mechanisms, relationships with other agents, etc.), defined as: Each entity unit intelligent agent has a unique identifier. ; It is a set that can be designed according to the specific needs of the equipment to perform the mission, and can include multiple specific parameters (such as failure rate, maintenance rate, etc.). This is a function of the intelligent agent, which can be implemented by writing relevant code; This is a supplementary explanation of intelligent agents; ; The relationships between agents can be represented as follows: ;in, This indicates that other intelligent agents have a relationship with this type of intelligent agent. Indicates the specific impact of the relationship (such as failure, performance degradation, etc.).

[0074] Among them, other entity units are other entity units in this intelligent agent, or other entity units in other intelligent agents.

[0075] In some embodiments of this application, the intelligent agent constructed through the above steps can be stored in the structured data format XML.

[0076] Simultaneously, based on the equipment's mission scenario and requirements, and combined with natural language, verifiable intelligent agent logic checklists and equipment fault behavior logic checklists can be generated. The former is used to verify and compensate the generated intelligent agents, while the latter is used to subsequently verify the completeness, consistency, and correctness of the simulation model implementation item by item. Each check item in the checklist can be used... express.

[0077] In other words, after constructing equipment-type intelligent agents, task-type intelligent agents, and maintenance and support resource-type intelligent agents, the equipment-type intelligent agents, task-type intelligent agents, and maintenance and support resource-type intelligent agents can be verified and compensated respectively by referring to the intelligent agent logic checklist.

[0078] Taking aircraft malfunction logic as an example, it can be further broken down into multiple checks: checking whether the aircraft has fault information reception and processing logic; checking whether the aircraft has fault state transition logic; and checking whether the aircraft has mission fault interruption logic. This generates an agent logic checklist and an equipment fault behavior logic checklist.

[0079] In some embodiments of this application, the business logic built based on the equipment failure behavior logic model is analyzed and generated one by one from three parts: equipment composition modeling, task scenario modeling, and failure behavior logic modeling, as detailed below:

[0080] First, an equipment composition model is constructed. This involves identifying target entity units within the target equipment class of intelligent agents; the target equipment class of intelligent agents can be any type of intelligent agent, and the target entity unit can be any entity unit within that class. Static prompts for the equipment composition model are constructed based on the attribute-level features of the target entity units, guiding the LLM to generate static attributes of the target entity units. Dynamic prompts for the equipment composition model are then constructed based on the behavioral-level features of the target entity units, guiding the LLM to generate dynamic interaction logic for the target entity units. Finally, the static attributes and dynamic interaction logic of each entity unit are embedded into its attribute and behavioral structures to obtain the equipment composition model.

[0081] In other words, targeting equipment-type intelligent agents Each entity unit in Based on its attribute layer characteristics, it can construct guiding prompts using identifiers, names (such as equipment component names), and descriptions as clues. LLM then generates static attribute codes for the intelligent agent that conform to the equipment background and mission requirements, thereby realizing the construction of basic information for the entity unit.

[0082] And, for equipment-type intelligent agents Each entity unit in Based on its behavioral characteristics, and using the relationships with other intelligent agents (such as the relationships between equipment components and the subordinate relationships between units) as clues, guiding prompts are constructed. Based on this, LLM generates dynamic interaction logic code for intelligent agents that conforms to the equipment background and mission requirements, thereby realizing the construction of relationships between entity units.

[0083] Finally, the static attributes and dynamic interaction logic code of the agent generated by LLM are embedded into the attribute structure and behavior structure of each entity unit of the equipment-type agent.

[0084] In one example, using an aircraft plane as an equipment-type intelligent agent, a description is constructed: "Please analyze the simulation scenario, consider the hierarchy and compositional relationships between equipment, and based on the static attributes of the entity unit 'intelligent agent Plane,' construct XML code to implement its static attributes." This code is then concatenated with the static attributes of the intelligent agent Plane to form a prompt, from which the larger model provides the implementation code. Additionally, based on the behavioral logic description of the intelligent agent Plane, prompts are constructed to generate the implementation code for the behavioral logic. Finally, the two parts of the implementation code are extracted into one... <activeobjectclass>Under the tag, merge them into a single XML fragment, which is the implementation code of the intelligent agent Plane.

[0085] Then, a task scenario model can be constructed. The target task unit within the target task class agent can be identified; the target task class agent can be any task class agent, and the target task unit can be any entity unit within that target task class agent. Static prompts for the task scenario model are constructed using the attribute layer features of the target task unit in the equipment composition model as clues, guiding the LLM to generate static attributes of the target task unit based on these static prompts. Dynamic prompts for the task scenario model are constructed using the behavioral layer features of the target task unit in the equipment composition model as clues, guiding the LLM to generate dynamic interaction logic for the target task unit based on these dynamic prompts.

[0086] Simultaneously, the target entity unit in the target equipment class intelligent agent can be identified; the target equipment class intelligent agent is any equipment class intelligent agent, and the target entity unit is any entity unit in the target equipment class intelligent agent; the behavioral layer features of the target entity unit in the equipment composition model are used as clues to construct supplementary prompt words for the first task scenario model, and guide LLM to generate supplementary dynamic interaction logic between the target equipment class intelligent agent and its corresponding task class intelligent agent based on the supplementary prompt words of the first task scenario model.

[0087] And determine the target resource entity in the target maintenance and support resource class intelligent agent; the target maintenance and support resource class intelligent agent is any maintenance and support resource class intelligent agent, and the target resource entity is any entity unit in the target maintenance and support resource class intelligent agent; construct supplementary prompt words for the second task scenario model based on the behavioral layer features of the target resource entity in the equipment composition model, and guide LLM to generate supplementary dynamic interaction logic between the target maintenance and support resource class intelligent agent and its corresponding task class intelligent agent based on the supplementary prompt words of the second task scenario model.

[0088] Finally, the static attributes and dynamic interaction logic of each task-type intelligent agent are embedded into the attribute structure and behavior structure of this task unit. The supplementary dynamic interaction logic of each equipment-type intelligent agent and its corresponding task-type intelligent agent is embedded into the attribute structure and behavior structure of this entity unit. The supplementary dynamic interaction logic of each maintenance and support resource-type intelligent agent and its corresponding task-type intelligent agent is embedded into the attribute structure and behavior structure of this target resource entity, thus obtaining the task scenario model.

[0089] In other words, for task-oriented intelligent agents Each task unit in Based on its attribute-level characteristics, and using identifiers, names (e.g., training, inspection, maintenance), parameters (e.g., task type, task start / end time, task duration), variables (e.g., constraints), functions, and descriptions as clues, guiding prompts can be constructed. Based on this, LLM generates static attribute codes for the intelligent agent that meet the equipment's task requirements, thus realizing the construction of basic information for various task units. Furthermore, based on its behavioral-level characteristics, and using state transition rules (e.g., task start rules, task end rules), and relationships with other intelligent agents (e.g., task and equipment-related intelligent agents), it can also... Maintenance and support resource-based intelligent agents Using relationships (such as connections) as clues, guiding prompts are constructed. Based on these, the LLM generates dynamic interaction logic code for intelligent agents that meets the equipment's mission requirements, thus realizing the construction of relationships between various task units. Then, the static attributes and dynamic interaction logic code of the intelligent agents generated by the LLM are embedded into the attribute structure and behavior structure of the equipment-type intelligent agents, respectively.

[0090] Taking the Task agent as an example, first, its ID is set to 2001, its type to "training", the required number of aircraft to be 5, the task start time to "10:00", and the end time to "12:00". The agent description is "Task agent, assigns aircraft to perform the task". Its behavioral logic "At the task start time, assigns aircraft to perform the task" is defined as an event and saved in XML format. The description "Please analyze the task scenario, based on the static attributes and behavioral logic of the entity unit 'Agent Task', and considering its interaction logic with 'Agent Plane', construct and implement the implementation code of Task" is constructed. After concatenating the description with the attributes and behavioral characteristics of Task into a prompt, the large model first provides the implementation code for the static attributes of Task, then provides the implementation code for the behavioral logic. Finally, the two parts of the implementation code are extracted into one. <activeobjectclass>Under the tag, merge them into an XML fragment, which is the implementation code of the agent Task.

[0091] Meanwhile, for equipment-type intelligent agents Each entity unit in Based on its behavioral characteristics, guided prompts can be constructed using state transition rules (such as standby to task execution, standby to maintenance, etc.), event response mechanisms (such as executing tasks after receiving task instructions, etc.), and relationships with other intelligent agents. LLM can then generate equipment-type intelligent agents based on these prompts. Task-oriented intelligent agents The dynamic interaction behavior logic supplementary code is then merged into the entity unit. In the structure.

[0092] In one example, the aircraft plane is an equipment-type intelligent agent. The construction description "Please analyze the mission scenario and, based on the existing static attributes and behavioral logic of the entity unit 'intelligent agent Pane', construct supplementary interaction logic between it and the mission intelligent agent" serves as a prompt. The large model provides the implementation code for the supplementary logic, and this implementation code is then merged into the intelligent agent plane structure.

[0093] And, for intelligent agents related to maintenance and support resources Each resource entity in Based on its behavioral characteristics, guided prompts can be constructed using state transition rules (such as maintenance personnel going from idle to working), event response mechanisms (such as executing maintenance tasks after receiving task instructions), and relationships with other intelligent agents. LLM can then generate maintenance support resource-type intelligent agents based on these prompts. Task-oriented intelligent agents The dynamic interaction behavior logic supplementary code is then merged into the entity unit. In the structure.

[0094] In one example, the maintenance personnel staff is a maintenance support resource-type intelligent agent. The description is "Please analyze the task scenario and construct supplementary logic code related to task execution based on the existing static attributes and behavioral logic of the entity unit 'maintenance personnel intelligent agent staff'". Combining the attribute layer features and behavioral layer features of the maintenance personnel intelligent agent staff, the large model provides the implementation code of the supplementary logic, and then the implementation code is merged into the maintenance personnel intelligent agent staff structure.

[0095] Finally, a fault behavior logic model can be constructed. The target entity unit within the target equipment class intelligent agent can be identified; the target equipment class intelligent agent can be any equipment class intelligent agent, and the target entity unit can be any entity unit within the target equipment class intelligent agent. Static prompts for the fault behavior logic model are constructed using the attribute layer features of the target entity unit in the task scenario model as clues, guiding the LLM to generate static fault attributes of the target entity unit based on these static prompts. Dynamic prompts for the task fault behavior logic model are constructed using the behavioral layer features of the target entity unit in the task scenario model as clues, guiding the LLM to generate dynamic fault interaction logic of the target entity unit based on these dynamic prompts.

[0096] Simultaneously, it can identify the target resource entity within the target maintenance and support resource class intelligent agent; the target maintenance and support resource class intelligent agent is any maintenance and support resource class intelligent agent, and the target resource entity is any entity unit within the target maintenance and support resource class intelligent agent; using the attribute layer features of the target resource entity in the task scenario model as clues, static prompt words for the fault behavior logic model are constructed to guide the LLM to generate static attributes of the target resource entity based on the static prompt words of the fault behavior logic model; using the behavior layer features of the target resource entity in the task scenario model as clues, dynamic prompt words for the fault behavior logic model are constructed to guide the LLM to generate dynamic interaction logic of the target resource entity based on the dynamic prompt words of the fault behavior logic model.

[0097] Finally, the static fault attributes and dynamic fault interaction logic of each entity unit are embedded into the attribute structure and behavior structure of the entity unit, and the static attributes and dynamic interaction logic of each resource entity are embedded into the attribute structure and behavior structure of the resource entity, thus obtaining the fault behavior logic model.

[0098] In other words, when constructing the fault behavior logic model, for equipment-type intelligent agents... Each entity unit in Based on its attribute layer characteristics, guided prompts can be constructed using parameters (such as mean time between failures (MTBF), failure rate, etc.), variables (such as remaining time after failure, etc.), and functions (such as failure probability distribution function, etc.). The LLM then generates static attribute codes of the intelligent agent that conform to the equipment failure generation logic, thereby realizing the construction of failure information for the entity unit.

[0099] Furthermore, based on its behavioral layer characteristics, guided prompts can be constructed using state transition rules (such as transitioning from a normal state to a fault state), event response mechanisms (from executing a task to terminating a task), and relationships with other intelligent agents as clues. The LLM then generates dynamic interaction logic code for intelligent agents that conforms to the equipment fault generation logic, thereby realizing the construction of dynamic fault behavior of the entity unit.

[0100] Then, the static fault information code and the dynamic fault behavior code of the equipment-type intelligent agent are supplemented and merged into the entity unit. In the structure.

[0101] In one example, the aircraft plane is an equipment-type intelligent agent. A prompt message, "Please analyze the fault behavior logic. Based on the existing static attributes and behavioral logic of the entity unit 'intelligent agent plane,' supplement the static attribute code that conforms to the fault generation logic," is provided by the larger model. Simultaneously, the larger model provides the supplementary behavioral logic implementation code that conforms to the fault logic interaction. Finally, the two parts of the implementation code are combined and merged into the structure of the aircraft intelligent agent plane.

[0102] Intelligent agents for maintenance and support resources Each resource entity in Based on its attribute layer characteristics, it can construct guiding prompts using identifiers, names (such as maintenance support resource names), parameters (such as the number of maintenance personnel), variables (such as the number of idle maintenance personnel), functions, and descriptions as clues. The LLM then generates static attribute codes for the intelligent agent that conform to the logic of equipment fault maintenance behavior, thereby realizing the construction of basic fault maintenance information for resource entities.

[0103] Furthermore, based on its behavioral layer characteristics, guided prompts can be constructed using state transition rules (such as transitioning from idle state to maintenance state), event response mechanisms (such as maintenance personnel being occupied), and relationships with other intelligent agents as clues. Based on these prompts, LLM can generate dynamic interaction logic code for intelligent agents that conforms to the logic of equipment fault maintenance behavior, thereby realizing the construction of dynamic fault maintenance behavior for resource entities.

[0104] Finally, the static fault repair information code and the dynamic fault repair behavior code of the maintenance and support resource-type intelligent agent are supplemented and merged into the resource entity. In the structure.

[0105] In one example, the maintenance staff is a maintenance support resource-type intelligent agent. The description "Please analyze the fault behavior logic and supplement the static attribute code that conforms to the fault generation logic based on the existing static attributes and behavior logic of the entity unit 'intelligent agent Plane'" is constructed. The attribute layer features and behavior layer features of the intelligent agent Plane are concatenated as prompt words. The large model first provides the implementation code of the static attributes and generates the implementation code of the behavior logic. Finally, the two parts of the implementation code are supplemented and merged into the structure of the maintenance staff.

[0106] In some embodiments of this application, logical verification can also be performed based on a logical checklist determined by the equipment failure behavior logic model, and compensation can be made to the equipment failure behavior logic model based on the verification results.

[0107] In some examples, performing logical verification and compensation on the equipment failure behavior logic model against a logical checklist can be as follows: In response to determining that the target business logic in the equipment failure behavior logic model is not implemented based on the logical checklist, locate the target agent associated with the target business logic; based on the existing attributes and behavioral context of the target agent, guide the LLM to generate missing attribute layer features and missing behavior layer features; add the missing attribute layer features and missing behavior layer features to the target agent; iteratively execute the steps of performing logical verification on the equipment failure behavior logic model against the logical checklist, generating missing attribute layer features and missing behavior layer features again when it is determined that the target business logic is not implemented, and adding the missing attribute layer features and missing behavior layer features to the target agent, until the logical verification result indicates that the target business logic has been implemented.

[0108] In other examples, performing logical verification and compensation on the equipment failure behavior logic model against a logical checklist can be as follows: In response to determining that the target business logic part of the equipment failure behavior logic model is implemented or functionally incomplete based on the logical checklist, locate the target agent associated with the target business logic, as well as the code implementation fragments in the target agent; while retaining the existing effective logic of the target agent, guide the LLM to generate supplementary attribute layer features and supplementary behavior layer features, and extend or modify the existing behavior logic; add the supplementary attribute layer features and supplementary behavior layer features to the target agent; iteratively execute the steps of performing logical verification on the equipment failure behavior logic model against the logical checklist, generating supplementary attribute layer features and supplementary behavior layer features again when it is determined that the target business logic part is implemented or functionally incomplete, and adding the supplementary attribute layer features and supplementary behavior layer features to the target agent, until the logical verification result indicates that the target business logic has been fully implemented.

[0109] In other examples, performing logical verification and compensation on the equipment failure behavior logic model by referring to the logical checklist can also involve: in response to determining that the target business logic error in the equipment failure behavior logic model is incorrect based on the logical checklist, locating the target agent associated with the target business logic; performing compliance verification on the target agent based on retrieval-enhanced knowledge, and guiding the LLM to update the target agent based on the verification result at least based on retrieval-enhanced knowledge; iteratively executing the steps of performing logical verification on the equipment failure behavior logic model by referring to the logical checklist, and performing compliance verification and updating on the target agent again when a target business logic error is determined, until the logical verification result indicates that the target business logic is executed correctly.

[0110] In other words, LLM is used to logically review the generated equipment failure behavior logic model, and categorize and compensate for any problems found to ensure that the model meets business logic requirements. For each inspection item... The logical checks and compensations are as follows:

[0111] Step 1: Logical Gap Compensation. When the LLM determines that a certain business logic is completely unimplemented based on the logic checklist, it performs the following operations in sequence: Locate the agent associated with the logic; Based on the existing attributes and behaviors of the agent, guide the LLM to generate the missing attribute and behavior elements; Add the newly generated elements to the corresponding agent structure; Re-execute the logic check until it passes.

[0112] For example, if the check item is sent as "Check if the aircraft has fault state transitions," the LLM is specified to judge and return one of three errors based on the review results: "Logical blank, logical missing, logical error." When the LLM returns "Logical blank," it further locates the specific agent and, based on the existing code, supplements and generates the missing attributes and behaviors. The LLM locates the aircraft agent Plane, finds that "fault" and state transitions are missing, and provides the implementation code for the "fault" state and state transitions. Then, the implementation code is added to the structure of the aircraft agent Plane, and the check item is re-validated.

[0113] Step 2: Logic Gap Compensation. When the LLM determines that a certain business logic is partially implemented or functionally incomplete, it performs the following operations in sequence: Locate the relevant intelligent agent and specific elements; While retaining the existing valid logic, guide the LLM to generate supplementary attributes or behavioral elements, and expand or modify the existing behavioral logic; Replace the original implementation with the updated elements; Re-verify the integrity of the logic.

[0114] For example, when the LLM returns "Logic Missing," it further locates the specific agent and generates corrected and missing attributes and behaviors based on existing code. The LLM locates the aircraft agent Plane, finds that only the "Fault" state is missing, and provides the implementation code for the "Fault" state. This implementation code is then added to the structure of the aircraft agent Plane; if a "Fault" state already exists, it is directly replaced. Finally, the check item is re-validated.

[0115] Step 3: Logical Error Compensation. When the LLM determines that a certain business logic contains a factual or structural error, it performs the following operations in sequence: Locate the element in the specific erroneous intelligent body; Perform compliance verification on the erroneous element based on the original knowledge definitions in the injected retrieval enhancement knowledge (such as attribute tuple specifications, behavioral semantic constraints) and the attribute and behavior definitions in Step 2; If the element does not conform to the definition, guide the LLM to regenerate the correct version based on the original knowledge and perform replacement or give a deletion instruction; After all relevant elements are corrected, re-execute the logic check.

[0116] For example, if the LLM returns a "logic error," it further locates the specific agent and element that erred. Based on the existing code, it performs compliance checks on the erroneous element according to the original knowledge definition (such as attribute tuple specifications and behavioral semantic constraints) used when injecting and retrieving augmented knowledge. The LLM locates the aircraft agent Plane and finds that the transition from the "standby" state to the "fault" state is incorrect. The LLM then modifies this to transition from the "working" state to the "fault" state and provides the implementation code. This implementation code replaces the original code, and finally, the logic check is re-executed.

[0117] Figure 3 This is a flowchart illustrating another method for intelligently generating an equipment fault behavior logic model based on LLM, provided in an embodiment of this application. Figure 3 As shown, the method includes the following steps:

[0118] In step one, enhanced knowledge injection can be performed to introduce simulation data information into the equipment's logical model, resulting in an enhanced logical model.

[0119] In step two, an agent model can be constructed, and a logic checklist can be generated. This logic checklist can include a verifiable agent logic checklist and an equipment malfunction behavior logic checklist.

[0120] In step three, an LLM model of equipment failure behavior can be constructed based on the built agent and checklist. In one example, the agent can first be verified and compensated using a verifiable agent logic checklist, and then the verified and compensated agent can be used to construct the equipment failure behavior logic model.

[0121] In step four, the LLM and equipment failure behavior logic checklist can be used to verify and compensate for the equipment failure behavior logic model.

[0122] By adopting the technical solution provided in the embodiments of this application, the intelligent generation of fault behavior models for complex equipment is realized by integrating the interactive modeling capabilities of intelligent agents with the semantic understanding and logical reasoning capabilities of large language models, thereby effectively improving the generation efficiency of equipment fault behavior models.

[0123] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0124] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0125] Figure 4 This is a schematic diagram of an intelligent generation device for an equipment fault behavior logic model based on LLM, provided in an embodiment of this application. Figure 4 As shown, the device includes:

[0126] The logical model building module 401 is configured to build a logical model of the equipment using the model-based systems engineering (MBSE) method and inject retrieval enhancement knowledge into the logical model to obtain an enhanced logical model.

[0127] The definition module 402 is configured to use LLM to construct equipment-type intelligent agents, task-type intelligent agents, and maintenance and support resource-type intelligent agents based on the enhanced logical model and equipment task scenarios, and to define attribute layer features and behavioral layer features for each entity unit in each intelligent agent; attribute layer features are used to characterize the static features of the entity unit, and behavioral layer features are used to characterize the dynamic interaction logic of the entity unit.

[0128] The equipment composition model construction module 403 is configured to use LLM to construct the basic information of each entity unit in the equipment class intelligent agent and the relationship between entity units based on the attribute layer features and behavior layer features of each entity unit in the equipment class intelligent agent, so as to obtain the equipment composition model.

[0129] The task scenario model construction module 404 is configured to, based on the equipment composition model, use LLM to construct the basic information and relationships between the entity units in the task-type intelligent agent based on the attribute layer features and behavioral layer features of each entity unit in the task-type intelligent agent; use LLM to construct the supplementary dynamic interaction logic between each equipment-type intelligent agent and the corresponding task-type intelligent agent based on the behavioral layer features of the entity units in each equipment-type intelligent agent; and use LLM to construct the supplementary dynamic interaction logic between each maintenance and support resource-type intelligent agent and the corresponding task-type intelligent agent based on the behavioral layer features of the entity units in each maintenance and support resource-type intelligent agent, thereby obtaining the task scenario model.

[0130] The equipment failure behavior logic model construction module 405 is configured to, based on the task scenario model, use LLM to construct supplementary static failure information and supplementary dynamic failure behavior of equipment-type intelligent agents based on the attribute layer features and behavioral layer features of each entity unit in the equipment-type intelligent agent, and use LLM to construct supplementary static failure information and supplementary dynamic failure behavior of each maintenance and support resource-type intelligent agent based on the attribute layer features and behavioral layer features of each entity unit in the maintenance and support resource-type intelligent agent, thereby obtaining the equipment failure behavior logic model.

[0131] According to the technical solution provided in this application, enhanced retrieval is achieved by combining a vector database, and a framework for generating equipment fault behavior logic models based on intelligent agents is constructed. Then, for each business component involved in the equipment fault behavior logic, structural elements in the intelligent agent are generated one by one, and the results are integrated and merged based on the model structure, ultimately realizing a complete and runnable equipment fault behavior logic simulation model. This method integrates the interactive modeling capabilities of intelligent agents with the semantic understanding and logical reasoning capabilities of large language models, realizing the intelligent generation of fault behavior models for complex equipment and effectively improving the generation efficiency of equipment fault behavior models.

[0132] In some implementations, the equipment composition model is constructed as follows: Target entity units within the target equipment class intelligent agent are identified; the target equipment class intelligent agent is any equipment class intelligent agent, and the target entity unit is any entity unit within the target equipment class intelligent agent; static prompts for the equipment composition model are constructed using the attribute layer features of the target entity unit as clues, guiding the LLM to generate static attributes of the target entity unit based on these static prompts; dynamic prompts for the equipment composition model are constructed using the behavioral layer features of the target entity unit as clues, guiding the LLM to generate dynamic interaction logic of the target entity unit based on these dynamic prompts; the static attributes and dynamic interaction logic of each entity unit are embedded into the attribute structure and behavioral structure of that entity unit to obtain the equipment composition model.

[0133] In some implementations, the task scenario model is constructed as follows: The target task unit within the target task class intelligent agent is determined; the target task class intelligent agent can be any task class intelligent agent, and the target task unit is any entity unit within the target task class intelligent agent; static prompts for the task scenario model are constructed using the attribute layer features of the target task unit in the equipment composition model as clues, guiding the LLM to generate static attributes of the target task unit based on the static prompts; dynamic prompts for the task scenario model are constructed using the behavioral layer features of the target task unit in the equipment composition model as clues, guiding the LLM to generate dynamic interaction logic of the target task unit based on the dynamic prompts; the target entity unit within the target equipment class intelligent agent is determined; the target equipment class intelligent agent can be any equipment class intelligent agent, and the target entity unit is any entity unit within the target equipment class intelligent agent; supplementary prompts for the first task scenario model are constructed using the behavioral layer features of the target entity unit in the equipment composition model as clues, guiding the LLM to generate... The process involves: supplementing the dynamic interaction logic between the target equipment-type intelligent agent and its corresponding task-type intelligent agent; identifying the target resource entity within the target maintenance and support resource-type intelligent agent; defining the target maintenance and support resource-type intelligent agent as any maintenance and support resource-type intelligent agent and the target resource entity as any entity unit within the target maintenance and support resource-type intelligent agent; constructing supplementary prompts for the second task scenario model based on the behavioral layer features of the target resource entity in the equipment composition model, guiding the LLM to generate supplementary dynamic interaction logic between the target maintenance and support resource-type intelligent agent and its corresponding task-type intelligent agent based on the supplementary prompts of the second task scenario model; embedding the static attributes and dynamic interaction logic of each task-type intelligent agent into the attribute structure and behavioral structure of this task unit, embedding the supplementary dynamic interaction logic of each equipment-type intelligent agent and its corresponding task-type intelligent agent into the attribute structure and behavioral structure of this entity unit, and embedding the supplementary dynamic interaction logic of each maintenance and support resource-type intelligent agent and its corresponding task-type intelligent agent into the attribute structure and behavioral structure of this target resource entity, thus obtaining the task scenario model.

[0134] In some implementations, the fault behavior logic model is constructed as follows: The target entity unit in the target equipment class intelligent agent is determined; the target equipment class intelligent agent is any equipment class intelligent agent, and the target entity unit is any entity unit within the target equipment class intelligent agent; static prompts for the fault behavior logic model are constructed using the attribute layer features of the target entity unit in the task scenario model as clues, guiding the LLM to generate static fault attributes of the target entity unit based on the static prompts; dynamic prompts for the task fault behavior logic model are constructed using the behavior layer features of the target entity unit in the task scenario model as clues, guiding the LLM to generate dynamic fault interaction logic of the target entity unit based on the dynamic prompts; the target resource entity in the target maintenance and support resource class intelligent agent is determined; the target maintenance and support resource class intelligent agent is... For any maintenance and support resource-type intelligent agent, the target resource entity is any entity unit within the target maintenance and support resource-type intelligent agent; using the attribute layer features of the target resource entity in the task scenario model as clues, static prompt words for the fault behavior logic model are constructed to guide the LLM to generate static attributes of the target resource entity based on the static prompt words of the fault behavior logic model; using the behavior layer features of the target resource entity in the task scenario model as clues, dynamic prompt words for the fault behavior logic model are constructed to guide the LLM to generate dynamic interaction logic of the target resource entity based on the dynamic prompt words of the fault behavior logic model; the fault static attributes and fault dynamic interaction logic of each entity unit are embedded in the attribute structure and behavior structure of the entity unit, and the static attributes and dynamic interaction logic of each resource entity are embedded in the attribute structure and behavior structure of the resource entity, thus obtaining the fault behavior logic model.

[0135] In some implementations, after obtaining the enhanced logic model, the method further includes: determining a verifiable checklist of equipment failure behavior logic based on the equipment's mission scenario and requirements; and after obtaining the equipment failure behavior logic model, the method further includes: performing logic verification based on the equipment failure behavior logic model against the logic checklist, and compensating the equipment failure behavior logic model based on the verification results.

[0136] In some implementations, logical verification and compensation of the equipment failure behavior logic model is performed against a logical checklist, including: in response to determining, based on the logical checklist, that the target business logic in the equipment failure behavior logic model is not implemented, locating the target agent associated with the target business logic; based on the existing attributes and behavioral context of the target agent, guiding the LLM to generate missing attribute layer features and missing behavioral layer features; adding the missing attribute layer features and missing behavioral layer features to the target agent; iteratively executing the steps of logically verifying the equipment failure behavior logic model against the logical checklist, regenerating the missing attribute layer features and missing behavioral layer features when it is determined that the target business logic is not implemented, and adding the missing attribute layer features and missing behavioral layer features to the target agent, until the logical verification result indicates that the target business logic has been implemented; and, in response to determining, based on the logical checklist, that the target business logic in the equipment failure behavior logic model is partially implemented or functionally incomplete, locating the target agent associated with the target business logic, and the code implementation fragments in the target agent; and, while retaining the existing valid logic of the target agent, guiding the LLM to generate missing attribute layer features and missing behavioral layer features to the target agent. M generates supplementary attribute layer features and supplementary behavior layer features, and expands or modifies the existing behavioral logic; adds the supplementary attribute layer features and supplementary behavior layer features to the target intelligent agent; iteratively executes the logic verification of the equipment failure behavior logic model against the logic checklist, and when it is determined that the target business logic is partially implemented or functionally incomplete, generates supplementary attribute layer features and supplementary behavior layer features again, and adds the supplementary attribute layer features and supplementary behavior layer features to the target intelligent agent, until the logic verification result indicates that the target business logic has been fully implemented; and, in response to determining that the target business logic in the equipment failure behavior logic model is incorrect based on the logic checklist, locates the target intelligent agent associated with the target business logic; performs compliance verification on the target intelligent agent based on retrieval augmented knowledge, and guides LLM to update the target intelligent agent at least based on retrieval augmented knowledge based on the verification result; iteratively executes the logic verification of the equipment failure behavior logic model against the logic checklist, and when it is determined that the target business logic is incorrect, performs compliance verification and updates the target intelligent agent again, until the logic verification result indicates that the target business logic is correctly executed.

[0137] In some implementations, the attribute layer features include at least one of the identifier, name, parameter, variable, function, or description of each entity unit; the behavioral layer features include at least one of the state transition rules, event response mechanism, or association relationship with other entity units of each entity unit; wherein, other entity units are other entity units in this intelligent agent, or other entity units in other intelligent agents.

[0138] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0139] Figure 5 This is a schematic diagram of the electronic device provided in an embodiment of this application. For example... Figure 5 As shown, the electronic device 5 of this embodiment includes: a processor 501, a memory 502, and a computer program 503 stored in the memory 502 and executable on the processor 501. When the processor 501 executes the computer program 503, it implements the steps in the various method embodiments described above. Alternatively, when the processor 501 executes the computer program 503, it implements the functions of each module / unit in the various device embodiments described above.

[0140] Electronic device 5 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 5 may include, but is not limited to, processor 501 and memory 502. Those skilled in the art will understand that... Figure 5 This is merely an example of electronic device 5 and does not constitute a limitation on electronic device 5. It may include more or fewer components than shown, or different components.

[0141] The processor 501 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0142] The memory 502 can be an internal storage unit of the electronic device 5, such as a hard disk or RAM of the electronic device 5. The memory 502 can also be an external storage device of the electronic device 5, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc., equipped on the electronic device 5. The memory 502 can also include both internal and external storage units of the electronic device 5. The memory 502 is used to store computer programs and other programs and data required by the electronic device.

[0143] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0144] If an integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable medium may include: any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0145] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.< / activeobjectclass> < / activeobjectclass>

Claims

1. A method for intelligently generating a logic model of equipment fault behavior based on LLM, characterized in that, include: A logical model of the equipment is constructed using the Model-Based Systems Engineering (MBSE) method, and retrieval-enhanced knowledge is injected into the logical model to obtain an enhanced logical model. Using the Large Language Model (LLM), based on the enhanced logical model and the equipment's task scenario, equipment-type intelligent agents, task-type intelligent agents, and maintenance and support resource-type intelligent agents are constructed, and attribute layer features and behavioral layer features are defined for each entity unit in each intelligent agent. The attribute layer features are used to characterize the static features of the entity unit, and the behavior layer features are used to characterize the dynamic interaction logic of the entity unit. Using LLM, the basic information of each entity unit in the equipment-type intelligent agent and the relationship between entity units are constructed based on the attribute layer features and behavior layer features of each entity unit in the equipment-type intelligent agent, thus obtaining the equipment composition model; Based on the equipment composition model, the LLM is used to construct the basic information of each entity unit in the task-type intelligent agent and the relationship between entity units based on the attribute layer features and behavior layer features of each entity unit in the task-type intelligent agent. The LLM is also used to construct the supplementary dynamic interaction logic between each equipment-type intelligent agent and the corresponding task-type intelligent agent based on the behavior layer features of the entity units in each equipment-type intelligent agent. Finally, the LLM is used to construct the supplementary dynamic interaction logic between each maintenance and support resource-type intelligent agent and the corresponding task-type intelligent agent based on the behavior layer features of the entity units in each maintenance and support resource-type intelligent agent, thus obtaining the task scenario model. Based on the task scenario model, supplementary static fault information and supplementary dynamic fault behavior of the equipment-type intelligent agent are constructed using LLM based on the attribute layer features and behavior layer features of each entity unit in the equipment-type intelligent agent. In addition, supplementary static fault information and supplementary dynamic fault behavior of each maintenance and support resource-type intelligent agent are constructed using LLM based on the attribute layer features and behavior layer features of each entity unit in the maintenance and support resource-type intelligent agent, thus obtaining the equipment fault behavior logic model. The supplementary dynamic interaction logic between equipment-type intelligent agents and their corresponding task-type intelligent agents is constructed in the following manner: Identify the target entity unit within the target equipment class intelligent agent; the target equipment class intelligent agent can be any equipment class intelligent agent, and the target entity unit is any entity unit within the target equipment class intelligent agent; Using the behavioral layer features of the target entity unit in the equipment composition model as clues, a first task scenario model supplementary prompt words are constructed to guide the LLM to generate supplementary dynamic interaction logic between the target equipment class intelligent agent and its corresponding task class intelligent agent based on the first task scenario model supplementary prompt words. The supplementary dynamic interaction logic between maintenance and support resource-type intelligent agents and their corresponding task-type intelligent agents is constructed in the following manner: Identify the target resource entity within the target maintenance and support resource class intelligent agent; the target maintenance and support resource class intelligent agent is any maintenance and support resource class intelligent agent, and the target resource entity is any entity unit within the target maintenance and support resource class intelligent agent; Using the behavioral features of the target resource entity in the equipment composition model as clues, a second task scenario model supplementary prompt words are constructed to guide the LLM to generate supplementary dynamic interaction logic between the target maintenance and support resource class intelligent agent and its corresponding task class intelligent agent based on the second task scenario model supplementary prompt words.

2. The method according to claim 1, characterized in that, The equipment composition model was constructed in the following manner: Identify the target entity unit within the target equipment class intelligent agent; the target equipment class intelligent agent can be any equipment class intelligent agent, and the target entity unit is any entity unit within the target equipment class intelligent agent; Using the attribute layer features of the target entity unit as clues, static prompt words for the equipment composition model are constructed to guide the LLM to generate static attributes of the target entity unit based on the static prompt words of the equipment composition model; Using the behavioral features of the target entity unit as clues, dynamic prompt words for the equipment composition model are constructed to guide the LLM to generate dynamic interaction logic for the target entity unit based on the dynamic prompt words of the equipment composition model. The static attributes and dynamic interaction logic of each entity unit are embedded into the attribute structure and behavior structure of the entity unit to obtain the equipment composition model.

3. The method according to claim 1, characterized in that, The task scenario model is constructed in the following manner: Identify the target task unit within the target task class intelligent agent; the target task class intelligent agent can be any task class intelligent agent, and the target task unit can be any entity unit within the target task class intelligent agent; Using the attribute layer features of the target task unit in the equipment composition model as clues, static prompt words for the task scenario model are constructed to guide the LLM to generate static attributes of the target task unit based on the static prompt words for the task scenario model. Using the behavioral features of the target task unit in the equipment composition model as clues, dynamic prompt words for the task scenario model are constructed to guide the LLM to generate the dynamic interaction logic of the target task unit based on the dynamic prompt words of the task scenario model. The static attributes and dynamic interaction logic of each task-type intelligent agent are embedded into the attribute structure and behavior structure of this task unit. The supplementary dynamic interaction logic of each equipment-type intelligent agent and its corresponding task-type intelligent agent is embedded into the attribute structure and behavior structure of this entity unit. The supplementary dynamic interaction logic of each maintenance and support resource-type intelligent agent and its corresponding task-type intelligent agent is embedded into the attribute structure and behavior structure of this target resource entity to obtain the task scenario model.

4. The method according to claim 1, characterized in that, The fault behavior logic model is constructed in the following manner: Identify the target entity unit within the target equipment class intelligent agent; the target equipment class intelligent agent can be any equipment class intelligent agent, and the target entity unit is any entity unit within the target equipment class intelligent agent; Using the attribute layer features of the target entity unit in the task scenario model as clues, static prompt words for the fault behavior logic model are constructed to guide the LLM to generate the fault static attributes of the target entity unit based on the static prompt words for the fault behavior logic model. Using the behavioral features of the target entity unit in the task scenario model as clues, construct dynamic prompt words for the task fault behavior logic model, and guide LLM to generate the fault dynamic interaction logic of the target entity unit based on the dynamic prompt words of the fault behavior logic scenario model; Identify the target resource entity within the target maintenance and support resource class intelligent agent; the target maintenance and support resource class intelligent agent is any maintenance and support resource class intelligent agent, and the target resource entity is any entity unit within the target maintenance and support resource class intelligent agent; Using the attribute layer features of the target resource entity in the task scenario model as clues, a static prompt word for the fault behavior logic model is constructed to guide the LLM to generate the static attributes of the target resource entity based on the static prompt word of the fault behavior logic model; Using the behavioral features of the target resource entity in the task scenario model as clues, construct dynamic prompt words for the fault behavior logic model, and guide the LLM to generate dynamic interaction logic for the target resource entity based on the dynamic prompt words of the fault behavior logic model. The fault behavior logic model is obtained by embedding the static fault attributes and dynamic fault interaction logic of each entity unit into the attribute structure and behavior structure of the entity unit, and the static attributes and dynamic interaction logic of each resource entity into the attribute structure and behavior structure of the resource entity.

5. The method according to claim 1, characterized in that, After obtaining the enhanced logical model, the method further includes: Based on the mission scenarios and requirements of the equipment, a verifiable checklist of intelligent agent logic and a checklist of equipment fault behavior logic are determined. After constructing equipment-type intelligent agents, task-type intelligent agents, and maintenance support resource-type intelligent agents, the equipment-type intelligent agents, task-type intelligent agents, and maintenance support resource-type intelligent agents are respectively verified and compensated according to the intelligent agent logic checklist; Furthermore, after obtaining the equipment failure behavior logic model, the method further includes: The equipment failure behavior logic model is logically verified against the equipment failure behavior logic checklist, and compensation is performed on the equipment failure behavior logic model based on the verification results.

6. The method according to claim 5, characterized in that, The equipment failure behavior logic model is logically verified and compensated against the equipment failure behavior logic checklist, including: In response to the determination that the target business logic in the equipment failure behavior logic model is not implemented based on the equipment failure behavior logic checklist, the target intelligent agent associated with the target business logic is located. Based on the existing attributes and behavioral context of the target agent, guide the LLM to generate features at the missing attribute layer and features at the missing behavioral layer. Add the missing attribute layer features and missing behavior layer features to the target agent; The process involves iteratively executing a checklist of equipment failure behavior logic to perform logical verification on the equipment failure behavior logic model. If it is determined that the target business logic has not been implemented, the missing attribute layer features and missing behavior layer features are generated again and added to the target intelligent agent. This process continues until the logical verification result indicates that the target business logic has been implemented. In addition, in response to determining that the target business logic part of the equipment failure behavior logic model is implemented or functionally incomplete based on the equipment failure behavior logic checklist, the target intelligent agent associated with the target business logic and the code implementation fragment in the target intelligent agent are located. While retaining the existing effective logic of the target agent, the LLM is guided to generate supplementary attribute layer features and supplementary behavior layer features, and the existing behavior logic is expanded or modified. The supplementary attribute layer features and supplementary behavior layer features are added to the target agent; The process involves iteratively executing a checklist of equipment failure behavior logic to perform logical verification on the equipment failure behavior logic model. When it is determined that the target business logic is partially implemented or the function is incomplete, supplementary attribute layer features and supplementary behavior layer features are generated again and added to the target intelligent agent. This process continues until the logical verification result indicates that the target business logic has been fully implemented. In addition, in response to determining the target business logic error in the equipment failure behavior logic model based on the equipment failure behavior logic checklist, the target intelligent agent associated with the target business logic is located. Based on the enhanced retrieval knowledge, the target agent is subjected to compliance verification, and based on the verification result, the LLM is guided to update the target agent at least based on the enhanced retrieval knowledge. The process involves iteratively executing a checklist of equipment failure behavior logic to perform logical verification on the equipment failure behavior logic model. If an error is found in the target business logic, the target intelligent agent is updated for compliance verification again until the logical verification result indicates that the target business logic is executed correctly.

7. The method according to any one of claims 1 to 6, characterized in that, The attribute layer features include at least one of the following for each entity unit: identifier, name, parameter, variable, function, or description; The behavioral layer features include at least one of the following: state transition rules, event response mechanisms, or association relationships with other entity units; wherein, other entity units are other entity units in this agent, or other entity units in other agents.

8. An intelligent generation device for equipment fault behavior logic model based on LLM, characterized in that, include: The logical model building module is configured to build a logical model of the equipment using the model-based systems engineering (MBSE) method, and inject retrieval enhancement knowledge into the logical model to obtain an enhanced logical model. The definition module is configured to use the Large Language Model (LLM) to construct equipment-type intelligent agents, task-type intelligent agents, and maintenance and support resource-type intelligent agents based on the enhanced logical model and the equipment's task scenario, and to define attribute layer features and behavioral layer features for each entity unit in each intelligent agent. The attribute layer features are used to characterize the static features of the entity unit, and the behavior layer features are used to characterize the dynamic interaction logic of the entity unit. The equipment composition model construction module is configured to use LLM to construct the basic information of each entity unit in the equipment class intelligent agent and the relationship between entity units based on the attribute layer features and behavior layer features of each entity unit in the equipment class intelligent agent, so as to obtain the equipment composition model. The task scenario model construction module is configured to, based on the equipment composition model, use LLM to construct the basic information of each entity unit in the task-type intelligent agent and the relationship between entity units based on the attribute layer features and behavior layer features of each entity unit in the task-type intelligent agent; use LLM to construct the supplementary dynamic interaction logic between each equipment-type intelligent agent and the corresponding task-type intelligent agent based on the behavior layer features of entity units in each equipment-type intelligent agent; and use LLM to construct the supplementary dynamic interaction logic between each maintenance and support resource-type intelligent agent and the corresponding task-type intelligent agent based on the behavior layer features of entity units in each maintenance and support resource-type intelligent agent, thereby obtaining the task scenario model. The equipment failure behavior logic model construction module is configured to, based on the task scenario model, use LLM to construct supplementary static failure information and supplementary dynamic failure behavior of the equipment-type intelligent agent based on the attribute layer features and behavior layer features of each entity unit in the equipment-type intelligent agent, and use LLM to construct supplementary static failure information and supplementary dynamic failure behavior of each maintenance and support resource-type intelligent agent based on the attribute layer features and behavior layer features of each entity unit in each maintenance and support resource-type intelligent agent, thereby obtaining the equipment failure behavior logic model. The supplementary dynamic interaction logic between equipment-type intelligent agents and their corresponding task-type intelligent agents is constructed in the following manner: Identify the target entity unit within the target equipment class intelligent agent; the target equipment class intelligent agent can be any equipment class intelligent agent, and the target entity unit is any entity unit within the target equipment class intelligent agent; Using the behavioral layer features of the target entity unit in the equipment composition model as clues, a first task scenario model supplementary prompt words are constructed to guide the LLM to generate supplementary dynamic interaction logic between the target equipment class intelligent agent and its corresponding task class intelligent agent based on the first task scenario model supplementary prompt words. The supplementary dynamic interaction logic between maintenance and support resource-type intelligent agents and their corresponding task-type intelligent agents is constructed in the following manner: Identify the target resource entity within the target maintenance and support resource class intelligent agent; the target maintenance and support resource class intelligent agent is any maintenance and support resource class intelligent agent, and the target resource entity is any entity unit within the target maintenance and support resource class intelligent agent; Using the behavioral features of the target resource entity in the equipment composition model as clues, a second task scenario model supplementary prompt words are constructed to guide the LLM to generate supplementary dynamic interaction logic between the target maintenance and support resource class intelligent agent and its corresponding task class intelligent agent based on the second task scenario model supplementary prompt words.

9. An electronic device 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 7.

10. 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 7.

Images