A multi-agent collaborative manufacturing method

Abstract

The application discloses a kind of multi-agent collaborative manufacturing method, comprising: intelligent management and control system, intelligent process system and digital twin system, by the 2D drawing / 3D model and demand submitted by user received by intelligent management and control system, issue to intelligent process system and start process generation;Intelligent process system generates initial scheme containing process route and NC code, and sends to digital twin system;Digital twin system verifies the scheme, and feedback structured optimization suggestion;Intelligent process system is iterated and optimized accordingly, until satisfying precision and efficiency index, form final executable process scheme and return to intelligent management and control system;Intelligent management and control system generates production planning and contract draft in combination with the scheme and real-time resource state;Task is issued, and the execution data collected are synchronously fed back to digital twin system to carry out deviation detection, and running result and correction record are returned to intelligent process system, drive intelligent process system to continue evolution.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, specifically a multi-agent collaborative manufacturing method. Background Technology

[0002] As the manufacturing industry moves towards higher quality, greater flexibility, and faster response, traditional manufacturing models face severe challenges in complex scenarios involving multiple varieties and small batches. Current manufacturing systems generally suffer from the following prominent problems, which seriously restrict the level of intelligence and lean operation capabilities of production lines.

[0003] First, at the production planning and scheduling level, existing methods mostly rely on static demand forecasting and manual experience to formulate plans, which are easily affected by subjective judgment biases, resulting in low planning accuracy. The production scheduling process is usually done manually, which is time-consuming and labor-intensive, and it is difficult to comprehensively consider multi-dimensional constraints such as real-time equipment status, worker skill matching, and material supply dynamics, resulting in unreasonable resource allocation, frequent scheduling conflicts, or idle and wasted capacity. Production scheduling relies on manual inspection and delayed reports, which cannot detect abnormal events in a timely manner, resulting in delayed scheduling decisions and low response efficiency.

[0004] Secondly, in terms of process and manufacturing collaboration, structured and standardized process data is difficult to efficiently transfer from process systems such as CAPP and CAM to the MES production management system. The lack of a unified data model and interface specifications between systems leads to a gap in process information. At the same time, the actual execution data and optimization suggestions generated during the production process are also difficult to effectively feed back to the process design end, forming a missing "design-manufacturing" closed loop, which hinders the continuous iteration and optimization of process knowledge.

[0005] Third, in the production preparation and logistics process, multiple elements are involved, such as materials, tooling, cutting tools, and CNC programs. However, warehousing and logistics management is still mainly based on offline paper circulation or semi-automation. The physical flow, manufacturing flow and information flow are seriously disconnected. There is a lack of coordinated scheduling of intelligent logistics equipment such as AGVs and smart shelves. The material distribution efficiency is low and the error rate is high, making it difficult to support the rapid changeover needs of flexible production lines.

[0006] Fourth, in terms of equipment interconnection and data acquisition, the overall CNC rate of production equipment is relatively low, and the brands and models are complex and the communication protocols are heterogeneous. The equipment networking rate is insufficient, and a large number of equipment become "information islands". Key indicators such as their operating status, processing parameters, alarm information, and energy consumption data cannot be collected and aggregated in real time. Data analysis relies on manual statistics, and the timeliness, completeness and accuracy of the data are difficult to guarantee, which restricts data-based intelligent decision-making.

[0007] Fifth, in terms of quality management, quality inspection relies heavily on manual measurement and recording by inspectors. Test data is entered into the information system afterward, which is prone to errors, tampering, and serious delays. Quality information is weakly linked to key elements such as production batches, equipment numbers, and process parameters. Once a quality problem occurs, the traceability chain breaks, making it difficult to quickly locate the root cause. The quality control model is still at the "post-event inspection" stage, lacking the ability to provide "pre-event warnings" and "in-event interventions" for potential risks.

[0008] Finally, regarding system integration and collaboration, enterprises generally face a situation where multiple business systems such as ERP, PLM, MES, and CAD / CAE coexist. These systems were built at different times, have different technical architectures, and inconsistent data standards, resulting in low integration between systems. Discontinuities and ambiguities occur during the transmission of design BOM, manufacturing BOM, and actual manufacturing data, preventing data from flowing smoothly in the business chain. Due to the lack of a unified data platform and integration architecture, intelligent agents, virtual agents, and physical entities in the virtual space cannot achieve mutual mapping and closed-loop control. The system struggles to make rapid assessments and dynamic strategy adjustments based on global real-time data, severely restricting the overall intelligence level and flexible adaptability of the production line.

[0009] In summary, existing manufacturing systems have systemic shortcomings in terms of planning accuracy, resource coordination, data connectivity, quality foresight, and system intelligence. There is an urgent need for a new manufacturing architecture that can break down information silos, integrate multi-source intelligence, and achieve seamless integration of the entire chain from design to process to production to logistics to quality, as well as a closed loop from perception to analysis to decision-making to execution. Summary of the Invention

[0010] The purpose of this invention is to solve the technical problems existing in the current manufacturing system, such as unintelligent production planning and scheduling, disconnect between process and manufacturing, low logistics efficiency, isolated equipment information, lagging quality control and low system integration, and to provide a multi-agent collaborative manufacturing method.

[0011] This invention is achieved through the following technical solution: A multi-agent collaborative manufacturing method includes: an intelligent control system, an intelligent process system, and a digital twin system. The intelligent control system receives 2D drawings / 3D models and requirements submitted by users and sends them to the intelligent process system to initiate process generation. The intelligent process system generates an initial solution containing process routes and NC code and sends it to the digital twin system. The digital twin system verifies the solution and provides structured optimization suggestions. The intelligent process system iteratively optimizes the solution until it meets accuracy and efficiency indicators, forming a final executable process solution and sending it back to the intelligent control system. The intelligent control system combines this solution with real-time resource status to generate a production schedule and draft contract. After the task is issued, the collected execution data is synchronously fed back to the digital twin system for deviation detection, and the running results and correction records are sent back to the intelligent process system, driving the continuous evolution of the intelligent process system.

[0012] In the above technical solution, the intelligent control system adopts a cloud-edge collaborative architecture. The edge layer deploys lightweight microservice processes in the workshop to run real-time control modules, ensuring that the response latency of control commands meets the interaction requirements. The cloud platform layer builds a private cloud platform based on virtualization clusters and a centralized microservice governance framework.

[0013] In the above technical solution, the intelligent control system integrates the following functional modules: ①The planning management module is used to generate production scheduling plans based on the process plans generated by the intelligent process system and resource data such as real-time equipment status and material inventory; ② Production execution control module, used to issue work orders and monitor process execution, track the work progress of each manufacturing unit in real time, and automatically collect work reporting data and abnormal events; ③ The process management module is used to receive the structured process specifications generated by the intelligent process system and distribute them to the corresponding manufacturing units according to the production schedule. ④ The material management module is used to monitor the entire material tracking process from raw material warehousing and production to finished product delivery; ⑤ The equipment management module is used to monitor the real-time operating status of the equipment, including processing parameters, alarm information, etc., and to record the equipment utilization rate and the reasons for downtime; ⑥ The Kanban management module provides a visual monitoring interface to display macro indicators such as order delivery and capacity load, and presents real-time information such as workstation tasks and work-in-process status to the workshop level, so as to achieve transparent management of the entire manufacturing process; ⑦ Production quality management module, used to statistically analyze quality data based on production and inspection data and establish a traceability chain between quality data and production batches, equipment, and processes, supporting rapid location and root cause analysis of quality problems; ⑧ Supply chain management module, used to select and assign manufacturers based on task characteristics, monitor cross-factory collaboration progress, and use de-identified task packages to ensure the confidentiality of order sources; ⑨ The virtual simulation module, integrated with the digital twin system, is used to call virtual models to simulate production plans and detect conflicts, identifying potential risks before tasks are issued.

[0014] In the above technical solution, the intelligent process system, based on a large language model and knowledge graph intelligent parsing module, as well as a CAPP knowledge database module and a CAM knowledge database module, is used to realize intelligent process and intelligent programming; among them, intelligent process completes the automatic generation of CAPP process specifications, and intelligent programming completes the intelligent generation of CAM NC code.

[0015] In the above technical solution, the intelligent parsing module is constructed using a three-layer architecture: the bottom layer deploys an AI computing hardware cluster and acceleration nodes as the computing power foundation, the middle layer runs a large language model, and the top layer constructs a process knowledge graph; among them, the large language model continuously learns from historical process documents and engineer experience to achieve semantic extraction of unstructured knowledge; the knowledge graph layer establishes multi-dimensional associations of materials, equipment, parameters, and quality, providing a unified knowledge source for the CAPP knowledge database module and the CAM knowledge database module.

[0016] In the above technical solution, the intelligent process system, based on the intelligent parsing module of large language model and knowledge graph, automatically identifies geometric features such as holes, slots, and curved surfaces in the drawings, as well as process constraints such as dimensional tolerances and surface roughness, and decomposes complex parts into several process units that can be processed independently. Subsequently, the intelligent process system invokes the CAPP knowledge database module to execute a dual-path decision-making mechanism: ① Rule-based path: Based on a pre-built process knowledge graph, retrieve standardized process rules that match the current features and materials to generate a preliminary process sequence; ② AI-based path: Analyze process cases of similar parts in history through the LLM engine, and dynamically optimize cutting parameters and process arrangement by combining real-time constraints such as current equipment status, tool inventory and capacity load; After the results of the two paths are merged, a structured process specification containing process sequence, equipment type requirements, tool specifications and quality control points is output. This process specification, along with the original drawing data, is transferred to the CAM knowledge database module. The CAM knowledge database module starts the NC code generation engine guided by the knowledge graph, automatically matches the optimal machining strategy based on the feature type, and calls LLM to semantically describe the tool path, generating a CNC executable NC code file with embedded process annotations.

[0017] In the above technical solution, the digital twin system uses Unity3D to build a digital twin model covering the entire life cycle of the production line, realizing real-time mapping and dynamic interaction between the physical production line and the virtual system.

[0018] In the above technical solution, the digital twin system comprises three parts: production line process simulation, layout simulation, and manufacturing execution process simulation. The production line process simulation constructs a virtual processing environment based on equipment parameters and process constraints, dynamically simulating the entire process execution from raw materials to finished products. The layout simulation verifies the rationality of the physical layout of the workshop through spatial topology analysis and logistics path optimization algorithms, identifies bottleneck areas, and provides layout adjustment suggestions. The manufacturing execution process simulation focuses on real-time production scenarios, injecting multi-source data such as equipment status, material flow, and quality inspection into the virtual model to achieve a high-fidelity reproduction of the actual manufacturing process.

[0019] In the above technical solution, the digital twin system constructs five simulation virtual entities, including: ① a dynamic scheduling simulation virtual entity, used to receive and issue production tasks, and predict delivery cycles and resource conflicts by combining real-time equipment load, worker skills, and material inventory data; ② a logistics planning simulation virtual entity, used to monitor logistics status; ③ a cutting performance simulation virtual entity, used to predict tool wear trends and machining vibration amplitude using physical mechanism models and historical cutting data, and identify quality risks in advance; ④ a geometric compensation simulation virtual entity, used to automatically calculate compensation amounts and feed them back to the machining equipment by analyzing temperature deformation and clamping errors; and ⑤ a machining process simulation virtual entity, used to virtually verify NC code, simulate tool trajectories and interference, and ensure process feasibility.

[0020] Another aspect of the present invention provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein, when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the methods described above.

[0021] The advantages and beneficial effects of this invention are as follows: The multi-agent collaborative manufacturing method designed in this invention includes: an intelligent control system, an intelligent process system, and a digital twin system. These three systems interact deeply through an event-driven architecture. The intelligent control system serves as the central control and decision-making hub of the multi-agent collaborative manufacturing architecture, used for command, scheduling, and resource optimization throughout the entire manufacturing process. The intelligent process system is the central hub for the intelligent application of process knowledge, used to automatically convert 2D drawings / 3D models into executable processes. The digital twin system acts as a mapping bridge between the physical manufacturing production line and the virtual space, used for mapping, verifying, and optimizing the physical manufacturing process.

[0022] Through mutual synergy and data interaction with users and manufacturers, the three systems have achieved the following outstanding technical effects: 1. The three major systems form a complete manufacturing closed loop through deep information collaboration: The intelligent control system, as the unified scheduling hub of the manufacturing process, first receives 2D drawings / 3D models and preliminary requirements submitted by users, and then issues a request to the intelligent process system to initiate process generation. The intelligent process system generates an initial process plan containing process routes and NC code, and sends it to the digital twin system. The digital twin system verifies the process plan and provides structured optimization suggestions. The intelligent process system iteratively optimizes the plan until it meets accuracy and efficiency indicators, forming a final executable process plan and sending it back to the intelligent control system. The intelligent control system combines the process plan with real-time resource status to generate a production schedule and draft contract. After the task is issued, the collected execution data is synchronously fed back to the digital twin system for deviation detection, and the results and correction records are sent back to the intelligent process system, driving its continuous evolution. This closed loop of "process generation—simulation verification—execution feedback—knowledge evolution" achieves a paradigm shift from passive response to pre-warning and in-process intervention, significantly improving flexibility and resource utilization efficiency.

[0023] 2. Collaborative Decision-Making and Knowledge Evolution of the Three Systems: During the manufacturing process, the three systems share equipment, process, and quality data to form a unified decision-making view. When deviations occur on the production line, the digital twin system triggers intervention commands, which are then verified by the control system and the execution strategy is dynamically adjusted. Key operational data is structured and fed back to the intelligent process system, where the Large Language Model (LLM) analyzes the root causes and automatically updates the intelligent process system's process knowledge graph, CAPP knowledge database module, and CAM knowledge database module. This enables the continuous evolution of the intelligent process system and achieves self-learning and self-optimization of manufacturing capabilities. Attached Figure Description

[0024] 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 only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a schematic diagram of the architecture of the multi-agent collaborative manufacturing method of the present invention. Detailed Implementation

[0026] To enable those skilled in the art to better understand the present invention, the technical solution of the present invention will be further described below with reference to specific embodiments.

[0027] See appendix Figure 1This invention provides a multi-agent collaborative manufacturing method, comprising: an intelligent control system, an intelligent process system, and a digital twin system. These three systems interact deeply through an event-driven architecture. The intelligent control system serves as the central control and decision-making hub of the multi-agent collaborative manufacturing architecture, enabling command, scheduling, and resource optimization throughout the entire manufacturing process. The intelligent process system acts as the intelligent application hub for process knowledge, facilitating the automatic conversion from 2D drawings / 3D models to executable processes. The digital twin system serves as a mapping bridge between the physical manufacturing production line and the virtual space, enabling the mapping, verification, and optimization of the physical manufacturing process.

[0028] The pairwise interactions between the three systems are as follows: 1. The interaction between the intelligent control system and the intelligent process system is as follows: After receiving 2D drawings / 3D models and preliminary requirements from users, the intelligent control system sends them to the intelligent process system to generate an executable process plan. The intelligent control system then combines real-time equipment status, material inventory, and other resource data to dynamically schedule production based on this process plan, calculate manufacturing costs and delivery cycles, and generate a structured draft contract that includes the process route, cost composition, and delivery timeline. Users can adjust their requirements in real time through the interactive interface, and the intelligent control system provides immediate feedback on resource matching assessments and impact analyses, thus achieving a precise and rapid response closed loop.

[0029] After receiving the drawings, the intelligent process system activates an intelligent parsing module based on a large language model and knowledge graph to decompose complex parts into machinable process units. These process units are then used to generate a structured process plan containing process sequences, equipment requirements, and quality control points. Once the process plan is fed back to the intelligent management and control system, it provides accurate input for dynamic production scheduling, manufacturing cost calculation, and delivery cycle prediction, supporting the rapid generation and user confirmation of draft contracts.

[0030] 2. The interaction between the intelligent control system and the digital twin system is as follows: The intelligent management and control system will generate a production schedule and push it to the digital twin system to simulate and verify the plan, including predicting risks such as resource conflicts, equipment overload, and logistics bottlenecks. The verification results will be fed back to the intelligent management and control system in the form of structured optimization suggestions for iterative optimization, and finally generate a highly feasible production schedule.

[0031] The intelligent control system officially issues production tasks, optimized through scheduling and virtual verification, to the physical production line. The digital twin system simultaneously collects real-world operational information such as equipment status, processing parameters, and quality data, continuously comparing it with the predicted values ​​from the virtual model to identify execution deviations in real time. Once an anomaly is detected, the digital twin system automatically generates tiered intervention instructions based on an AI model. These instructions are verified by the intelligent control system and then issued to the physical production line for execution. Through this closed-loop process of "task issuance—data collection—deviation identification—intervention decision—instruction execution," a closed-loop control system for the production process is achieved.

[0032] 3. The interaction between the intelligent process system and the digital twin system is as follows: After generating an initial process plan, the intelligent process system does not directly issue it to production. Instead, it pushes it to a digital twin system for virtual verification. The digital twin system performs a comprehensive simulation of the feasibility, safety, and efficiency of the process plan, identifies potential risks, and generates structured optimization suggestions that are fed back to the intelligent process system. Upon receiving the feedback, the intelligent process system adjusts and optimizes the NC code and verifies it again. This process is repeated until the plan meets efficiency indicators, forming a final executable process plan, thus achieving a closed loop of process verification and optimization: "process—simulation—optimization".

[0033] The process of collaboration among the three systems is as follows: 1. The three major systems form a complete manufacturing closed loop through deep information collaboration: The intelligent control system, as the unified scheduling hub of the manufacturing process, first receives 2D drawings / 3D models and preliminary requirements submitted by users, and then issues a request to the intelligent process system to initiate process generation. The intelligent process system generates an initial process plan containing process routes and NC code, and sends it to the digital twin system. The digital twin system verifies the process plan and provides structured optimization suggestions. The intelligent process system iteratively optimizes the plan until it meets accuracy and efficiency indicators, forming a final executable process plan and sending it back to the intelligent control system. The intelligent control system combines the process plan with real-time resource status to generate a production schedule and draft contract. After the task is issued, the collected execution data is synchronously fed back to the digital twin system for deviation detection, and the results and correction records are sent back to the intelligent process system, driving its continuous evolution. This closed loop of "process generation—simulation verification—execution feedback—knowledge evolution" achieves a paradigm shift from passive response to pre-warning and in-process intervention, significantly improving flexibility and resource utilization efficiency.

[0034] 2. Collaborative Decision-Making and Knowledge Evolution of the Three Systems: During the manufacturing process, the three systems share equipment, process, and quality data to form a unified decision-making view. When deviations occur on the production line, the digital twin system triggers intervention commands, which are then verified by the control system and the execution strategy is dynamically adjusted. Key operational data is structured and fed back to the intelligent process system, where the Large Language Model (LLM) analyzes the root causes and automatically updates the intelligent process system's process knowledge graph, CAPP knowledge database module, and CAM knowledge database module. This enables the continuous evolution of the intelligent process system and achieves self-learning and self-optimization of manufacturing capabilities.

[0035] Furthermore, in one embodiment, the intelligent control system adopts a cloud-edge collaborative architecture. The edge layer deploys lightweight microservice processes in the workshop to run real-time control modules, ensuring that the response latency of control commands meets interaction requirements. The cloud platform layer builds a private cloud platform based on a virtualization cluster and a centralized microservice governance framework, integrating the following functional modules: ①The planning management module is used to generate production scheduling plans based on the process schemes generated by the intelligent process system and resource data such as real-time equipment status and material inventory.

[0036] ② Production execution control module, used to issue work orders and monitor process execution, track the work progress of each manufacturing unit in real time, and automatically collect work reporting data and abnormal events.

[0037] ③ The process management module is used to receive the structured process specifications generated by the intelligent process system and distribute them to the corresponding manufacturing units according to the production schedule.

[0038] ④ The material management module is used to monitor the entire material tracking process from raw material warehousing and production to finished product delivery.

[0039] ⑤ The equipment management module is used to monitor the real-time operating status of the equipment, including processing parameters, alarm information, etc., and to record the equipment utilization rate and the reasons for downtime.

[0040] ⑥ The Kanban management module provides a visual monitoring interface to display macro indicators such as order delivery and capacity load, and presents real-time information such as workstation tasks and work-in-process status to the workshop level, realizing transparent management of the entire manufacturing process.

[0041] ⑦ Production quality management module, used to statistically analyze quality data based on production and inspection data and establish a traceability chain between quality data and production batches, equipment, and processes, supporting rapid location and root cause analysis of quality problems.

[0042] ⑧ Supply chain management module, used to select and assign manufacturers based on task characteristics, monitor cross-factory collaboration progress, and use de-identified task packages to ensure the confidentiality of order sources.

[0043] ⑨ The virtual simulation module, integrated with the digital twin system, is used to call virtual models to simulate production plans and detect conflicts, identifying potential risks before tasks are issued.

[0044] Furthermore, in one embodiment, the intelligent process system, based on the intelligent parsing module of the large language model and knowledge graph, as well as the CAPP knowledge database module and the CAM knowledge database module, realizes two core functions: intelligent process and intelligent programming; wherein, intelligent process completes the automatic generation of CAPP process specifications, and intelligent programming completes the intelligent generation of CAM NC code.

[0045] Furthermore, the intelligent parsing module is constructed using a three-layer architecture: the bottom layer deploys an AI computing hardware cluster and acceleration nodes as the computing power foundation, the middle layer runs a large language model (LLM engine), and the top layer constructs a process knowledge graph; among them, the LLM engine continuously learns from historical process documents and engineer experience to achieve semantic extraction of unstructured knowledge; the knowledge graph layer establishes multi-dimensional associations of materials, equipment, parameters, and quality, providing a unified knowledge source for the CAPP knowledge database module and the CAM knowledge database module.

[0046] The intelligent parsing module based on large language models and knowledge graphs integrates multi-scale geometric feature extraction technology and process knowledge graph reasoning capabilities. It can automatically identify geometric features such as holes, slots, and curved surfaces in drawings, as well as process constraints such as dimensional tolerances and surface roughness, and decompose complex parts into several independently machinable process units.

[0047] Subsequently, the intelligent process system invokes the CAPP knowledge database module to execute a dual-path decision-making mechanism: ① Rule-based path: Based on a pre-built process knowledge graph, retrieve standardized process rules that match the current features and materials to generate a preliminary process sequence; ② AI-based path: Analyze process cases of similar parts in history through the LLM engine, and dynamically optimize cutting parameters and process arrangement by combining real-time constraints such as current equipment status, tool inventory and capacity load.

[0048] The results from the two paths are merged to output a structured process specification that includes the process sequence, equipment type requirements, tool specifications, and quality control points. This process specification, along with the original drawing data, is then transferred to the CAM knowledge database module. The CAM knowledge database module activates the NC code generation engine guided by the knowledge graph, automatically matches the optimal machining strategy based on feature types, and calls LLM to semantically describe the toolpath, generating a CNC executable NC code file with embedded process annotations.

[0049] Finally, the complete process solution package, including structured process specifications and NC code, is pushed to the digital twin system to trigger the virtual machining verification process to verify the feasibility of the process and the quality of the machining.

[0050] Furthermore, in one embodiment, the digital twin system, relying on a unified data foundation, uses Unity3D (a 3D creation engine used to develop digital twin visualization scenarios) to construct a digital twin model covering the entire lifecycle of the production line, achieving real-time mapping and dynamic interaction between the physical production line and the virtual system. The system integrates three main parts: production line process simulation, layout simulation, and manufacturing execution process simulation, forming a complete virtual verification capability. Specifically, the production line process simulation constructs a virtual processing environment based on equipment parameters and process constraints, dynamically simulating the entire process execution from raw materials to finished products; the layout simulation verifies the rationality of the workshop's physical layout through spatial topology analysis and logistics path optimization algorithms, identifies bottleneck areas, and provides layout adjustment suggestions; the manufacturing execution process simulation focuses on real-time production scenarios, injecting multi-source data such as equipment status, material flow, and quality inspection into the virtual model to achieve a high-fidelity reproduction of the actual manufacturing process.

[0051] Furthermore, the digital twin system constructs five virtual simulation entities to achieve refined management and control, including: ① a dynamic scheduling virtual simulation entity, used to receive and issue production tasks, and predict delivery cycles and resource conflicts by combining real-time equipment load, worker skills, and material inventory data; ② a logistics planning virtual simulation entity, which monitors logistics status based on UWB (Ultra-Wideband, used for high-precision indoor positioning and ranging) positioning data and AGV operating status; ③ a cutting performance virtual simulation entity, which uses physical mechanism models and historical cutting data to predict tool wear trends and machining vibration amplitude, and identify quality risks in advance; ④ a geometric compensation virtual simulation entity, which automatically calculates compensation amounts and feeds them back to the machining equipment by analyzing temperature deformation and clamping errors; and ⑤ a machining process virtual simulation entity, used to virtually verify NC code, simulate tool trajectories and interference, and ensure process feasibility.

[0052] The above description is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A multi-agent collaborative manufacturing method, characterized in that: include: Intelligent control system, intelligent process system, and digital twin system; The intelligent control system is used to receive 2D drawings / 3D models and requirements submitted by users, and then send them to the intelligent process system to start process generation. The intelligent process system generates an initial plan containing process routes and NC code, and sends it to the digital twin system; The digital twin system validates the solution and provides structured optimization suggestions. The intelligent process system iterates and optimizes accordingly until it meets the accuracy and efficiency indicators, forming a final executable process plan and sending it back to the intelligent control system; the intelligent control system combines the plan with real-time resource status to generate a production schedule and a draft contract. After the task is issued, the collected execution data is synchronously fed back to the digital twin system for deviation detection, and the operation results and correction records are sent back to the intelligent process system to drive the continuous evolution of the intelligent process system.

2. The multi-agent collaborative manufacturing method according to claim 1, characterized in that: The intelligent management and control system adopts a cloud-edge collaborative architecture. The edge layer deploys lightweight microservice processes in the workshop; the cloud platform layer builds a private cloud platform based on virtualization clusters and a centralized microservice governance framework.

3. The multi-agent collaborative manufacturing method according to claim 1, characterized in that: The intelligent control system integrates the following functional modules: ①The planning management module is used to generate production scheduling plans based on the process plans generated by the intelligent process system, as well as the real-time status of equipment and material inventory data; ② Production execution control module, used to issue work orders and monitor process execution, track the work progress of each manufacturing unit in real time, and automatically collect work reporting data and abnormal events; ③ The process management module is used to receive the structured process specifications generated by the intelligent process system and distribute them to the corresponding manufacturing units according to the production schedule. ④ The material management module is used to monitor the entire material tracking process from raw material warehousing and production to finished product delivery; ⑤ The equipment management module is used to monitor the real-time operating status of the equipment, including processing parameters, alarm information, and record equipment utilization and downtime reasons; ⑥ The Kanban management module provides a visual monitoring interface to display order delivery and capacity load, presenting real-time information on workstation tasks and work-in-process status to the workshop level, thereby achieving transparent management of the entire manufacturing process; ⑦ Production quality management module, used to statistically analyze quality data based on production and inspection data and establish a traceability chain between quality data and production batches, equipment, and processes, supporting rapid location and root cause analysis of quality problems; ⑧ Supply chain management module, used to select and assign manufacturers based on task characteristics, monitor cross-factory collaboration progress, and use de-identified task packages to ensure the confidentiality of order sources; ⑨ The virtual simulation module, integrated with the digital twin system, is used to call virtual models to simulate production plans and detect conflicts, identifying potential risks before tasks are issued.

4. The multi-agent collaborative manufacturing method according to claim 1, characterized in that: The intelligent process system, based on a large language model and knowledge graph intelligent parsing module, as well as a CAPP knowledge database module and a CAM knowledge database module, is used to realize intelligent process and intelligent programming. Among them, intelligent process completes the automatic generation of CAPP process specifications, and intelligent programming completes the intelligent generation of NC code for CAM.

5. The multi-agent collaborative manufacturing method according to claim 4, characterized in that: The intelligent parsing module is built with a three-layer architecture: the bottom layer deploys an AI computing hardware cluster and acceleration nodes as the computing power base, the middle layer runs a large language model, and the top layer builds a process knowledge graph. Among them, the large language model continuously learns from historical process documents and engineers' experience to achieve semantic extraction of unstructured knowledge; the knowledge graph layer establishes multi-dimensional associations of materials, equipment, parameters, and quality, providing a unified knowledge source for the CAPP knowledge database module and the CAM knowledge database module.

6. The multi-agent collaborative manufacturing method according to claim 4, characterized in that: The intelligent process system, based on the intelligent parsing module of large language model and knowledge graph, automatically identifies the geometric features of holes, slots and curved surfaces in drawings, as well as the process constraints of dimensional tolerances and surface roughness, and decomposes complex parts into several process units that can be processed independently. Subsequently, the intelligent process system invokes the CAPP knowledge database module to execute a dual-path decision-making mechanism: ① Rule-based path: Based on a pre-built process knowledge graph, retrieve standardized process rules that match the current features and materials to generate a preliminary process sequence; ② AI-based path: Analyze process cases of similar parts in history through the LLM engine, and dynamically optimize cutting parameters and process arrangement by combining the current equipment status, tool inventory and real-time constraints of capacity load; After the results of the two paths are merged, a structured process specification containing process sequence, equipment type requirements, tool specifications and quality control points is output. This process specification, along with the original drawing data, is transferred to the CAM knowledge database module. The CAM knowledge database module starts the NC code generation engine guided by the knowledge graph, automatically matches the optimal machining strategy based on the feature type, and calls LLM to semantically describe the tool path, generating a CNC executable NC code file with embedded process annotations.

7. The multi-agent collaborative manufacturing method according to claim 1, characterized in that: The digital twin system uses Unity3D to build a digital twin model covering the entire lifecycle of the production line, enabling real-time mapping and dynamic interaction between the physical production line and the virtual system.

8. The multi-agent collaborative manufacturing method according to claim 1, characterized in that: The digital twin system comprises three parts: production line process simulation, layout simulation, and manufacturing execution process simulation. The production line process simulation constructs a virtual processing environment based on equipment parameters and process constraints, dynamically simulating the entire process execution from raw materials to finished products. The layout simulation verifies the rationality of the physical layout of the workshop through spatial topology analysis and logistics path optimization algorithms, identifies bottleneck areas, and provides layout adjustment suggestions. The manufacturing execution process simulation focuses on real-time production scenarios, injecting multi-source data on equipment status, material flow, and quality inspection into the virtual model to achieve a high-fidelity reproduction of the actual manufacturing process.

9. The multi-agent collaborative manufacturing method according to claim 1, characterized in that: The digital twin system comprises five virtual simulation entities: ① a dynamic scheduling simulation entity, used to receive and issue production tasks, and predict delivery cycles and resource conflicts by combining real-time equipment load, worker skills, and material inventory data; ② a logistics planning simulation entity, used to monitor logistics status; ③ a cutting performance simulation entity, used to predict tool wear trends and machining vibration amplitude using physical mechanism models and historical cutting data, and to identify quality risks in advance; ④ a geometric compensation simulation entity, used to automatically calculate compensation amounts and feed them back to the machining equipment by analyzing temperature deformation and clamping errors; and ⑤ a machining process simulation entity, used to virtually verify NC code, simulate tool trajectories and interference, and ensure process feasibility.

10. An electronic device, characterized in that, include: One or more processors; A memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the multi-agent cooperative manufacturing method as described in any one of claims 1-9.

Images