Intelligent review system and method integrating multi-agent cooperative driving

Abstract

The invention provides an intelligent review system and method fusing multi-agent cooperative driving, and the system comprises a task receiving and planning module which is used for receiving a research and development project task, analyzing a task type, planning an execution path, and generating a task instruction containing an agent type and a task feature; the rule calling module is used for selecting an examination rule matched with the task type from a rule base based on the task instruction; the multi-scene review agent module comprises at least one agent, each agent executes a review process, and the review process comprises the steps of analyzing a review rule and determining a review target and a review standard; obtaining to-be-examined data from the research and development project data based on the examination target; matching the to-be-examined data with an examination standard to generate an examination result; and the report generation module is used for summarizing the review results of all the agents and generating a review report containing a problem list, a compliance conclusion and a modification suggestion. According to the invention, automatic execution and high result integration of the collaborative intelligent review task in multiple scenes are realized.

Description

Technical Field

[0001] This application relates to the field of intelligent review technology, and in particular to an intelligent review system and method that integrates multi-agent collaborative driving. Background Technology

[0002] The R&D system for high-end equipment, complex systems, and industrial products typically encompasses multiple R&D processes, including requirements management, 2D drawing design, 2D and 3D general design, 3D modeling, process and manufacturing, design verification, design release, and circuit and hardware design. These processes are broad in scope and highly interconnected, placing extremely high demands on the standardization, compliance, and consistency of design deliverables.

[0003] In actual R&D processes, after completing tasks at each stage, designers need to undergo multi-level technical reviews to ensure that the design deliverables comply with internal company specifications, industry standards, national standards, and international standards. However, the existing review model mainly relies on manual review, with experienced engineers checking each item against the standards. This approach suffers from low efficiency, error-proneness, fragmentation across different scenarios, and difficulties in cross-team collaboration, making it difficult to meet the modern demands for efficient and precise R&D quality assurance.

[0004] With the development of artificial intelligence, knowledge graphs, and intelligent agent technologies, existing research has attempted to achieve partial automation of review through rule engines, expert systems, and CAD plugins. However, these R&D review systems are mostly based on single intelligent agents or independent rule execution, lacking a unified collaborative mechanism for multiple scenarios and tasks. This hinders the effective integration and sharing of knowledge, resulting in review results that are often one-sided and difficult to adapt to changing R&D needs. Furthermore, the maintenance and updating of rule bases are costly, and the review process cannot achieve a fully automated closed loop. Therefore, there is an urgent need for a highly efficient and intelligent collaborative review solution to improve the quality and efficiency of R&D review. Summary of the Invention

[0005] This application provides an intelligent review system and method that integrates multi-agent collaborative driving to solve the technical problems of efficient collaboration and multi-scenario adaptation in the prior art. It realizes intelligent decomposition of R&D project tasks, automatic invocation of rules, collaborative review of multiple agents, and unified generation of results, which significantly improves the automation level and accuracy of R&D review.

[0006] Firstly, this application provides an intelligent review system that integrates multi-agent collaborative driving, comprising: The task receiving and planning module is used to receive R&D project tasks, parse the task type and plan the execution path, and generate task instructions containing agent type and task characteristics. The rule invocation module is used to select review rules that match the task type from the rule base based on task instructions; The multi-scenario review agent module includes at least one agent, each agent executing a review process. The review process includes: parsing review rules, determining review objectives and review standards; obtaining data to be reviewed from R&D project data based on the review objectives; matching the data to be reviewed with the review standards, and generating review results. The report generation module is used to summarize the review results of all agents and generate a review report that includes a list of issues, compliance conclusions, and modification suggestions.

[0007] Secondly, this application also provides an intelligent review method that integrates multi-agent collaborative driving, including: Receive R&D project tasks, parse task types and plan execution paths, and generate task instructions containing agent types and task characteristics; Based on the task instructions, select the review rules that match the task type from the rule base; The review process is executed by at least one intelligent agent. The review process includes: parsing review rules, determining review objectives and review standards; obtaining data to be reviewed from R&D project data based on the review objectives; matching the data to be reviewed with the review standards, and generating review results. The review results of all agents are summarized to generate a review report that includes a list of issues, compliance conclusions, and modification suggestions.

[0008] Thirdly, this application also 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 any of the above-described intelligent review methods that integrate multi-agent collaborative driving.

[0009] Fourthly, this application also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described intelligent review methods that integrate multi-agent collaborative driving.

[0010] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements any of the above-described intelligent review methods that integrate multi-agent collaborative driving.

[0011] This application provides an intelligent review system and method that integrates multi-agent collaborative drive. Through automatic task parsing and intelligent planning, it improves the accuracy of task allocation, enables parallel collaboration among multiple agents, and increases review processing speed. Multi-scenario review agents perform professional reviews for different R&D stages, achieving cross-domain knowledge integration and sharing, ensuring comprehensive and reliable review results. The system supports various R&D scenarios and can quickly expand review capabilities through modular design to meet complex and ever-changing R&D needs. It automatically summarizes agent review results, generates a unified report, and supports a user feedback mechanism to promote continuous optimization of rules and models and system self-learning. Intelligent collaboration reduces human intervention, achieves a standardized review process, and improves the collaborative efficiency of R&D teams. This application realizes the automated execution and high integration of results for collaborative intelligent review tasks in multiple scenarios. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is one of the structural schematic diagrams of the intelligent review system that integrates multi-agent collaborative driving provided in the embodiments of this application; Figure 2 This is the second schematic diagram of the structure of the intelligent review system that integrates multi-agent collaborative driving provided in the embodiments of this application; Figure 3 This is a flowchart illustrating the intelligent review method that integrates multi-agent collaborative driving provided in the embodiments of this application; Figure 4 This is a schematic diagram of the rule generation process provided in the embodiments of this application; Figure 5 This is a schematic diagram of the intelligent review process provided in an embodiment of this application; Figure 6 This is a schematic diagram of the collaborative review process provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] Figure 1 This is one of the structural schematic diagrams of the intelligent review system that integrates multi-agent collaborative driving provided in the embodiments of this application, such as... Figure 1 As shown, the intelligent review system mainly includes the following functional modules: The task receiving and planning module 101 is used to receive R&D project tasks, parse the task type and plan the execution path, and generate task instructions containing agent type and task characteristics.

[0016] The rule invocation module 102 is used to select review rules that match the task type from the rule base based on the task instruction.

[0017] The multi-scenario review agent module 103 includes at least one agent, each agent executes a review process, which includes: parsing review rules, determining review objectives and review standards; obtaining data to be reviewed from R&D project data based on the review objectives; matching the data to be reviewed with the review standards, and generating review results.

[0018] The report generation module 104 is used to summarize the review results of all intelligent agents and generate a review report that includes a list of issues, compliance conclusions, and modification suggestions.

[0019] The intelligent review system provided in this application, which integrates multi-agent collaborative driving, improves the accuracy of task allocation through automatic task parsing and intelligent planning, enables parallel collaboration of multiple agents, and increases the review processing speed. Multi-scenario review agents perform professional reviews for different R&D stages, achieving cross-domain knowledge integration and sharing, ensuring comprehensive and reliable review results. The system supports various R&D scenarios and can quickly expand review capabilities through modular design to meet complex and ever-changing R&D needs. It automatically summarizes the review results of agents, generates a unified report, and supports a user feedback mechanism to promote continuous optimization of rules and models and system self-learning. Intelligent collaboration reduces human intervention, achieves a standardized review process, and improves the collaborative efficiency of R&D teams.

[0020] The task receiving and planning module 101 receives various R&D project tasks submitted by users, including design documents, two-dimensional / three-dimensional drawings, process flow, circuit and hardware design, etc.; it parses the task type, identifies the R&D scenario, intelligently plans the execution path, and generates task instructions containing agent type and task characteristics, providing guidance for subsequent agent task allocation.

[0021] In some embodiments, the task receiving and planning module 101 uses Natural Language Processing (NLP) technology to parse the task type of the R&D project task and dynamically adjusts the execution path based on the task type and system resources.

[0022] Specifically, the task receiving and planning module 101 uses NLP technology to automatically parse the text of the R&D project tasks input by the user, extract the task type and key information, and dynamically adjust the execution path according to the current system resources to realize intelligent allocation and scheduling of tasks.

[0023] The embodiments of this application improve the automation and intelligence of task parsing, enhance the accuracy of task instructions, strengthen the system's flexible scheduling capability under multi-task and multi-scenario conditions, and improve the overall review efficiency.

[0024] The rule invocation module 102 selects review rules corresponding to the task type from the built-in rule base based on the task type contained in the task instruction. The rule base can include domain knowledge, expert experience, and historical review cases to ensure the comprehensiveness and accuracy of the review rules; alternatively, it can call external rules to generate review rules for the intelligent agent. The review rules are executed by the intelligent agent, eliminating the need for rule generation within the system.

[0025] The multi-scenario review agent module 103 includes multiple functional agents, such as drawing review agents, structural review agents, process review agents, and circuit and hardware review agents. Under the scheduling of the planning agent, multiple agents execute the review process in parallel or collaboratively. Each agent specifically executes the review process based on the invoked review rules. The review process specifically includes: Step a: Analyze the review rules and clarify the review objectives and standards; Step b: Obtain the data to be reviewed from the multimodal data of the original R&D project tasks based on the review objectives; Step c: Match the acquired data to be reviewed with the review criteria, identify potential problems, and generate review results.

[0026] In some embodiments, the intelligent agent includes at least one of a drawing review intelligent agent, a structure review intelligent agent, a process review intelligent agent, and a circuit and hardware review intelligent agent.

[0027] Specifically, the intelligent agent for drawing review automatically reviews dimensions, annotations, and symbol specifications in CAD drawings; the intelligent agent for structural review performs mechanical analysis and verifies the structural rationality of product structural designs, confirming compliance with engineering standards and constraints; the intelligent agent for process review examines the rationality of manufacturing process flows and parameters, assessing the rationality, manufacturability, and compliance of the process flows; and the intelligent agent for circuit and hardware review checks the compliance of circuit designs and hardware interfaces, verifying whether circuit diagrams and hardware designs meet electrical specifications and safety standards. Other expandable review agents, such as verification agents and result review agents, can also be included to adapt to different R&D scenarios.

[0028] This application provides a system that can configure different types of intelligent agents to work in parallel or collaboratively, supports cross-domain information sharing and comprehensive judgment, and can cover R&D project tasks in multiple fields and stages. It supports review in multiple R&D scenarios such as requirements, 2D / 3D design, process, circuit and hardware, and realizes comprehensive review across scenarios and disciplines, thereby improving the coverage and professionalism of the review.

[0029] The report generation module 104 is responsible for collecting and integrating the review results of all intelligent agents. Through conflict resolution and information fusion, it ultimately generates a unified review report. The review report includes a list of issues, compliance conclusions, and modification suggestions for the issues. It supports structured and visual presentation of the report to facilitate user understanding and use.

[0030] In some embodiments, the system further includes: The feedback and knowledge optimization module is used to receive user feedback on the review report, associate the feedback with the review results, and optimize the rule calling logic of the rule calling module 102 and the agent collaboration mechanism of the multi-scenario review agent module 103 based on the feedback.

[0031] Specifically, the feedback and knowledge optimization module receives and processes user feedback on the review report, feeding back the review results, user feedback, and manual confirmation results to the system. This module correlates the feedback with previously generated review results, analyzes the impact of the feedback on the rule invocation logic of the rule invocation module 102 and the agent collaboration mechanism of the multi-scenario review agent module 103, and automatically adjusts and optimizes these mechanisms to enhance the system's self-learning and evolution capabilities.

[0032] The embodiments of this application implement a closed-loop feedback mechanism, which can feed back review results, user feedback and manual confirmation information to the system to optimize the rule calling logic and the intelligent agent collaboration mechanism, thereby improving the system's self-learning ability.

[0033] In some embodiments, the system further includes: The multi-agent communication module is used to enable communication and collaboration between agents using the Multi-agent Communication Protocol (MCP), supporting modular deployment and scalable services.

[0034] Specifically, the multi-intelligent communication module enables efficient communication and collaboration between intelligent agents based on the MCP protocol, supports asynchronous message passing and state synchronization, and ensures modular deployment while achieving system scalability and high reliability.

[0035] The embodiments of this application ensure real-time interaction capabilities between intelligent agents, promote collaborative task execution, improve the scalability and maintenance convenience of the system, and support complex and ever-changing review business needs.

[0036] In some embodiments, the interaction process between intelligent agents is stored and managed through a memory tree structure, which includes nodes that record historical task information, parameter preferences and review results, and supports dynamic data access based on the memory tree structure.

[0037] Specifically, all agent interactions are stored and managed through a memory tree structure. This structure records historical task information, parameter preferences, and review results, which are then dynamically recalled in subsequent interactions, enabling agents to learn and reinforce their knowledge. Through continuous memory recall and feedback updates, the system can continuously optimize task planning strategies and response accuracy, enhancing the personalization of user interactions and the intelligence of responses.

[0038] The embodiments of this application enhance the contextual understanding and continuous learning capabilities of the intelligent agent, enabling the system to continuously improve task planning and response accuracy based on historical data, thereby increasing the personalization and intelligence of user interaction.

[0039] In some embodiments, the system employs a layered architecture design, including: The MCP client layer includes a planning agent, a multi-server management module, a model management module, a file set management module, a user management module, and a data management module, which are used to uniformly manage and schedule multi-agent services. The MCP server layer includes an agent server cluster and a file management component. The agent server cluster is equipped with a file execution service module, a rule generation service module, a multi-scenario review service module, and a report generation service module. It uses the MCP protocol to realize message interaction and data transmission between agents and servers. The intelligent agent layer includes modules for document content intelligent agents, rule generation intelligent agents, report generation intelligent agents, and multi-scenario review intelligent agents, which implement specific review tasks. The data storage layer includes file storage, input file recognition, file data fusion, structured file management, and R&D system interface. It is used to store and manage various types of R&D data and supports the storage and management of multiple types of R&D data. The tool layer, including the rule engine, multiple programming language runtime environments, and CAD / PLM interfaces, supports the operation of intelligent agents and data processing.

[0040] The layered architecture design provided in this application clearly defines the functions of each module, improves the flexibility and maintainability of the system, supports independent updates and expansions of modules, and meets the integration needs of complex and heterogeneous R&D environments.

[0041] Figure 2 This is the second structural schematic diagram of the intelligent review system that integrates multi-agent collaborative driving provided in the embodiments of this application, as shown below. Figure 2 As shown, the system adopts an outside-in, layered, and decoupled architecture design, consisting of a front-end application layer, an MCP client layer, an MCP server layer, an intelligent agent layer, a data storage layer, and a tool layer, from top to bottom. The layered architecture aims to achieve multi-agent collaboration, modular deployment, and scalable services, allowing for flexible expansion with new intelligent agents and review rules to adapt to different enterprise and industry R&D scenarios.

[0042] At the front-end application layer, the system supports multiple interaction methods, accessible via client programs, web pages, or by embedding or integrating with enterprise office automation (OA) platforms using Uniform Resource Locators (URLs). The front-end interface provides users with functions such as intelligent search, rule generation, model review, result viewing, and statistical analysis, achieving a unified business entry point and interactive experience.

[0043] At the MCP Client layer, the system employs a communication mechanism compliant with the MCP protocol. The client uses a planning agent as its core scheduling unit, responsible for task reception, decomposition, scheduling, and execution control. Based on task type and complexity, the planning agent can invoke connected MCP Servers layer modules to automatically complete tasks such as task orchestration, model parameter configuration, and file and data management. As the system core, the planning agent automatically receives various types of tasks uploaded by users, performs task parsing, classification, and allocation, enabling dynamic scheduling across different scenarios. Simultaneously, the Client layer also handles user permission management, task status monitoring, and data security management, serving as the unified interface for interaction between the system and the external environment.

[0044] At the MCP Servers layer, the system constructs a multi-agent capability cluster, which serves as the core computing and execution center of the entire architecture. Each capability is deployed as an independent server, forming a decoupled but collaboratively schedulable agent cluster. This layer includes three core services: Intelligent Agent Servers Cluster: Responsible for core tasks such as file fragmentation, rule generation, multi-scenario review and report generation. Each server consists of one or more intelligent agents and can be independently expanded and dynamically loaded. Artificial Intelligence (AI) Model Service Layer: Provides intelligent algorithm support based on Large Language Model (LLM), vLLM, or Transformer networks, and performs model inference, task decomposition, result generation, and multimodal data analysis; File Management Layer: Responsible for the isolated storage, type identification, workflow editing, and secure uploading of files, ensuring data isolation and secure transmission between different tasks, users, and enterprises.

[0045] In this layer of architecture, the capabilities of the intelligent agent server can come from various implementation methods: it can be composed of traditional software workflows, or it can integrate tool modules developed in languages ​​such as Python, C++, and Java, or it can be combined with a finely tuned AI model to build an intelligent working system. Each server realizes message passing and data flow interaction through the MCP protocol, thereby supporting system-level distributed collaboration.

[0046] At the agent layer, the system constructs multiple types of agents according to their functional divisions: the file fragmentation agent is used to fragment and preprocess design documents, responsible for the standardized decomposition and organization of input data; the rule generation agent is used to generate the rules required for review based on existing specifications, or to call external review rules generated by models and rule generation algorithms; the multi-scenario review agent performs compliance reviews for different areas such as drawings, structures, processes, and circuits; and the report generation agent completes statistical analysis and report output. Agents collaborate across modules through the scheduling of planning agents and can be flexibly combined and executed according to task characteristics, achieving multi-task parallelism and dynamic allocation. Different agents are provided with interface capabilities by the MCP Servers layer, supporting independent operation and collaborative invocation.

[0047] At the data storage layer, the system constructs a unified database architecture to support the operation and result storage of intelligent agents. The data layer includes a file system, graph database, structured relational database, and historical record database, used to store graph model data, review rules, review records, statistical information, and historical comparison results. Simultaneously, a unified data interface and transmission standard are defined to ensure data compatibility and traceability between different modules.

[0048] At the tool support layer, the system encapsulates various types of development and runtime tools, including a rule engine (Drools), programming language environments (Python, Java, C#, C++), an AI search engine (Milvus), CAD / CAE integrated applications (CATIA, CREO), and an enterprise collaboration platform (TC / PLM). These tools provide upper-layer intelligent agents with algorithm execution, workflow-driven, and system integration capabilities, forming the foundational support layer of the entire architecture.

[0049] This system architecture, driven by the MCP protocol, employs a multi-agent cluster design, model service-oriented architecture, and layered management of data and tools. This achieves high cohesion, low coupling, and flexible collaboration across different levels, enabling seamless integration between traditional programming tools and large AI models. It facilitates multi-scenario review and intelligent task orchestration, along with excellent scalability and cross-platform adaptability, allowing for rapid configuration and deployment tailored to different enterprise R&D systems.

[0050] The system provided in this application embodiment has a high degree of multi-scenario scalability in its design. Based on the MCP (ModelContext Protocol) architecture mechanism, each intelligent agent module, toolkit, and service interface of the system is deployed in a decoupled manner as an independent node, and different review scenarios can be quickly expanded through "descriptive access".

[0051] When new business review scenarios need to be added (such as new process compliance checks, electronic hardware layout reviews, etc.), developers only need to add the corresponding text descriptions and task definition information to the system. The MCP client can automatically identify the new scenario and complete capability registration and tool mapping during the multi-agent task planning phase. The system automatically loads the corresponding review agents and toolsets through standardized interfaces, enabling dynamic integration and operation of new review scenarios without modifying the original code logic.

[0052] Figure 3 This is a flowchart illustrating the intelligent review method integrating multi-agent collaborative driving provided in the embodiments of this application, as shown below. Figure 3 As shown, this method, applied to the aforementioned intelligent review system driven by multi-agent collaboration, includes the following steps: S301. Receive R&D project tasks, parse task types and plan execution paths, and generate task instructions containing agent types and task characteristics. S302. Select review rules that match the task type from the rule base based on the task instructions; S303. At least one intelligent agent executes the review process, which includes: parsing review rules, determining review objectives and review standards; obtaining data to be reviewed from R&D project data based on the review objectives; matching the data to be reviewed with the review standards to generate review results; S304. Summarize the review results of all agents and generate a review report that includes a list of issues, compliance conclusions, and modification suggestions.

[0053] Specifically, the system receives R&D project tasks, analyzes the task type using technologies such as Natural Language Processing (NLP), plans the optimal execution path based on task characteristics and system resources, and generates task instructions containing the corresponding agent type and task characteristics.

[0054] Based on the task instructions, the system matches and selects review rules suitable for the current task type from the rule base, which includes standard specifications, expert experience, and historical review data.

[0055] At least one intelligent agent receives the review rules, and sequentially completes the following steps: parsing the review rules, determining the review objectives and standards, obtaining the data to be reviewed, and performing rule matching to generate the review results.

[0056] The system aggregates the review results submitted by each intelligent agent and integrates them into a complete review report that includes a list of issues, compliance conclusions, and rectification suggestions. It supports structured and visual presentation to facilitate user understanding.

[0057] The embodiments of this application realize the automated decomposition of R&D project tasks and cross-agent collaborative review, ensuring the comprehensiveness of the review and the efficiency of system processing, and improving the level of intelligence in quality assurance.

[0058] In some embodiments, the review process in S303 is performed by at least one agent, including: Different intelligent agents are selected to perform the review process based on the research and development scenario. The research and development scenario includes at least one of the following: drawing review, structural review, process review, and circuit and hardware review.

[0059] Specifically, during the process of an intelligent agent executing a review process, the system dynamically selects the appropriate intelligent agent to execute the review process based on the specific R&D scenario. These R&D scenarios include, but are not limited to: Drawing review: Automatically reviews dimensions, annotations, symbol specifications, etc. in CAD drawings; Structural review: Verifying whether the structural design complies with engineering standards and constraints; Process review: Assessing the feasibility, manufacturability, and compliance of the process flow; Circuit and hardware review: Check whether the circuit diagrams and hardware designs meet electrical specifications and safety standards.

[0060] This application's embodiments improve the targeting and professionalism of the review process by selecting intelligent agents tailored to R&D scenarios, and enhance the system's ability to adapt to diverse project needs and complex R&D environments.

[0061] In some embodiments, the method further includes: It receives user feedback on the review report, associates the feedback with the review results, and optimizes the rule invocation logic and agent collaboration mechanism based on the feedback.

[0062] Specifically, the system receives feedback information from the user regarding the review report, performs correlation analysis with the review results, and dynamically optimizes the rule invocation logic and agent collaboration mechanism based on the feedback information.

[0063] The embodiments of this application realize closed-loop optimization and adaptive adjustment of the intelligent review system, continuously improve the accuracy of rule matching and the efficiency of intelligent agent collaboration, enhance the system's ability to respond to changing needs, and improve user satisfaction.

[0064] It should be noted that the intelligent review method with multi-agent collaborative driving provided in the embodiments of this application can be referred to in correspondence with the intelligent review system with multi-agent collaborative driving described above.

[0065] The following specific example further illustrates the intelligent review method based on the fusion of multi-agent collaborative drive provided in the embodiments of this application.

[0066] This intelligent review method, which integrates multi-agent collaborative drive, can be divided into five stages: user interaction, task planning, rule generation, agent collaborative execution, and result generation. Through the multi-agent collaborative mechanism, the system achieves fully automated processing from natural language input to rule generation, intelligent review, and report output.

[0067] User interaction phase: Users input commands or answer basic questions in natural language through the front-end interface, and can directly submit requests such as rule generation, drawing review, and report generation. The system front-end supports both text input and file upload, and users can attach semantic descriptions to clarify task objectives or constraints.

[0068] Task Planning Phase: After a user uploads a file or submits a complex task, the planning agent receives and parses the user's intent, automatically decomposing the task and planning the process. Based on the capabilities of the server cluster connected to the MCP client, the planning agent searches for available toolsets (such as rule generation services, the Drools engine, and file parsing modules), determines the optimal execution path and steps, and assigns specific sub-tasks to subsequent execution agents.

[0069] Rule generation phase: Figure 4 This is a schematic diagram of the rule generation process provided in the embodiments of this application, such as... Figure 4 As shown, taking rule generation as an example, when a user uploads a JSON-formatted rule file and requests its conversion into an executable DRL rule, the system automatically identifies the adaptation tool and determines the execution order. The planning agent, based on task dependencies, sequentially calls the file structure parsing tool, the rule mapping conversion tool, and the Drools compilation interface to automate rule conversion and deployment. The entire process is completed automatically by the agent, handling task scheduling and data preparation without manual intervention.

[0070] Agent collaborative execution and result generation phase: Figure 5 This is a schematic diagram of the intelligent review process provided in the embodiments of this application, such as... Figure 5 As shown, when a user uploads drawings, models, or process documents, the system first uses a planning agent to analyze the document type and review objectives, automatically matching the required rule documents and assigning the task to the corresponding intelligent review agent. Once the review is complete, the system automatically calls upon the report generation agent to format and output the review results, generating a visual review report and archiving it.

[0071] For complex review tasks, the system can simultaneously invoke multiple tools or agents to achieve collaborative inspection. Figure 6 This is a schematic diagram of the collaborative review process provided in an embodiment of this application, such as... Figure 6 As shown, when a user uploads an NX 3D drawing, the system first calls the integrated NX inspection software to perform a basic review, and then the dimensional chain inspection agent completes the geometric association and dimensional logic verification, thereby achieving a comprehensive review at different levels and in different dimensions.

[0072] This method can flexibly adapt to the review needs of different R&D fields, including mechanical design, electrical design, process manufacturing, structural verification, and other scenarios, achieving seamless expansion from single applications to multiple business scenarios, and significantly improving the system's versatility, maintainability, and continuous evolution capabilities. Simultaneously, this method makes the integration and collaboration between artificial intelligence technology and traditional programming logic more intelligent, achieving efficient connection between model reasoning and engineering implementation, significantly reducing repetitive workload in collaborative development and system integration, significantly shortening the R&D cycle, and improving overall development efficiency. The integration of multiple agents retains the review capabilities of traditional programming to ensure the interpretability and stability of the review process, while leveraging large language models to challenge technical boundaries that are difficult to achieve with traditional programming.

[0073] The intelligent review system and method that integrates multi-agent collaborative driving provided in this application have the following beneficial effects: Strong multi-scenario coverage capability: Supports unified review of multiple R&D scenarios such as structural design, process design, circuit and hardware design, improving system applicability; Intelligent scheduling: By planning intelligent agents to analyze and allocate tasks, flexible scheduling across tasks and scenarios can be achieved; Multi-agent collaboration: Different functional agents can cooperate and share information to form comprehensive review conclusions across domains; Fully automated process: From task upload to review report generation, everything is done automatically by the system, greatly reducing manual intervention; Knowledge loop and optimization: Review data and feedback flow back to the system to support continuous learning and optimization, so that performance can be continuously improved with use; Excellent scalability and versatility: The system adopts a modular design, which can quickly expand new intelligent agents according to different R&D processes and standards, adapt to the collaborative R&D needs of multiple teams and multiple scenarios, and reduce the code integration work of multiple teams.

[0074] Figure 7 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application, such as... Figure 7 As shown, the electronic device may include: a processor 701, a communications interface 702, a memory 703, and a communication bus 704, wherein the processor 701, the communications interface 702, and the memory 703 communicate with each other via the communication bus 704. The processor 701 can call logical instructions in the memory 703 to execute an intelligent review method that integrates multi-agent collaborative driving, the method including: Receive R&D project tasks, parse task types and plan execution paths, and generate task instructions containing agent types and task characteristics; Based on the task instructions, select the review rules that match the task type from the rule base; The review process is executed by at least one intelligent agent. The review process includes: parsing review rules, determining review objectives and review standards; obtaining data to be reviewed from R&D project data based on the review objectives; matching the data to be reviewed with the review standards, and generating review results. The review results of all agents are summarized to generate a review report that includes a list of issues, compliance conclusions, and modification suggestions.

[0075] Furthermore, the logical instructions in the aforementioned memory 703 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0076] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the intelligent review method that integrates multi-agent cooperative driving provided by the above methods, the method including: Receive R&D project tasks, parse task types and plan execution paths, and generate task instructions containing agent types and task characteristics; Based on the task instructions, select the review rules that match the task type from the rule base; The review process is executed by at least one intelligent agent. The review process includes: parsing review rules, determining review objectives and review standards; obtaining data to be reviewed from R&D project data based on the review objectives; matching the data to be reviewed with the review standards, and generating review results. The review results of all agents are summarized to generate a review report that includes a list of issues, compliance conclusions, and modification suggestions.

[0077] Furthermore, this application also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the intelligent review method fused with multi-agent cooperative drive provided by the methods described above, the method comprising: Receive R&D project tasks, parse task types and plan execution paths, and generate task instructions containing agent types and task characteristics; Based on the task instructions, select the review rules that match the task type from the rule base; The review process is executed by at least one intelligent agent. The review process includes: parsing review rules, determining review objectives and review standards; obtaining data to be reviewed from R&D project data based on the review objectives; matching the data to be reviewed with the review standards, and generating review results. The review results of all agents are summarized to generate a review report that includes a list of issues, compliance conclusions, and modification suggestions.

[0078] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0079] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0080] Finally, it should be noted that 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.

Claims

1. A fusion multi-agent collaborative driving intelligent review system, characterized in that, Comprise: a task receiving and planning module for receiving a research and development project task, analyzing a task type, and planning an execution path, and generating a task instruction containing an agent type and a task feature; a rule calling module for selecting an examination rule matching the task type from a rule library based on the task instruction; a multi-scene examination agent module comprising at least one agent, each agent executing an examination process, the examination process comprising: analyzing the examination rule, determining an examination target and an examination standard; obtaining to-be-examined data from research and development project data based on the examination target; matching the to-be-examined data with the examination standard to generate an examination result; a report generation module for summarizing the examination results of all agents and generating an examination report containing a problem list, a compliance conclusion, and a modification suggestion. 2.The intelligent review system of claim 1, wherein, The agent comprises at least one of a drawing examination agent, a structure examination agent, a process examination agent, and a circuit and hardware examination agent. 3.The intelligent review system of claim 1, wherein, The system further comprises: a feedback and knowledge optimization module for receiving feedback information of the examination report from a user, associating the feedback information with the examination result, and optimizing a rule calling logic of the rule calling module and an agent cooperation mechanism of the multi-scene examination agent module based on the feedback information. 4.The intelligent review system of claim 1, wherein, The task receiving and planning module analyzes the task type of the research and development project task by using a natural language processing (NLP) technology, and dynamically adjusts the execution path based on the task type and system resources.

5. The intelligent review system of claim 1, wherein, The system further comprises: a multi-agent communication module for realizing communication and cooperation between agents by using a MCP protocol, supporting modular deployment and scalable services. 6.The intelligent review system of claim 1, wherein, An interaction process between the agents is stored and managed by a memory tree structure, the memory tree structure comprising nodes recording historical task information, parameter preferences, and examination results, and supporting dynamic data access based on the memory tree structure. 7.The intelligent review system of claim 1, wherein, The system adopts a hierarchical architecture design, comprising: an MCP client layer comprising a planning agent, a multi-server management module, a model management module, a file set management module, a user management module, and a data management module, for uniformly managing and scheduling multi-agent services; an MCP server layer comprising an agent server cluster and a file management component, the agent server cluster being provided with a file execution service module, a rule generation service module, a multi-scene examination service module, and a report generation service module, and realizing message interaction and data transmission between agents and servers by using a MCP protocol; an agent layer containing a file content agent, a rule generation agent, a report generation agent, and the multi-scene examination agent module; a data storage layer comprising a file storage, an input file identification, a file data fusion, a structured file management, and a research and development system interface, for storing and managing various research and development data; a tool layer comprising a rule engine, a plurality of programming language running environments, and a CAD / PLM interface, for supporting running and data processing of the agents.

8. An intelligent review method driven by fusion of multi-agent collaboration, characterized in that, Comprise: receiving a research and development project task, analyzing a task type, and planning an execution path, and generating a task instruction containing an agent type and a task feature; selecting, based on the task instruction, an examination rule matching the task type from a rule library; performing, by at least one agent, an examination process, the examination process comprising: analyzing the examination rule, determining an examination target and an examination standard; obtaining, based on the examination target, to-be-examined data from the R&D project data; matching the to-be-examined data with the examination standard, and generating an examination result; aggregating the examination results of all the agents to generate an examination report containing a problem list, a compliance conclusion and a modification suggestion. 9.The intelligent review method of claim 8, wherein, The performing, by at least one agent, an examination process comprises: selecting different agents to perform the examination process based on different R&D scenarios, the R&D scenarios comprising at least one of a drawing examination, a structure examination, a process examination and a circuit and hardware examination.

10. The intelligent review method of claim 8 or 9, wherein, The method further comprises: receiving feedback information of the examination report from a user, associating the feedback information with the examination result, and optimizing a rule calling logic and an agent cooperation mechanism based on the feedback information.

Images