CN-121980926-A - Intelligent planning and simulation verification system for processing technology based on LLM multiple agents

CN121980926ACN 121980926 ACN121980926 ACN 121980926ACN-121980926-A

Abstract

The invention relates to the technical field of intelligent manufacturing, in particular to a processing technology intelligent planning and simulation verification system based on LLM multiple agents, which adopts a layered architecture design and comprises a core coordination layer, a professional execution layer and an infrastructure layer, wherein the layers cooperatively realize the intelligent planning and simulation verification of the processing technology, and the core coordination layer is a central LLM agent. The invention adopts a layered architecture, takes a central LLM intelligent body as a core coordination layer, matches with professional execution layer intelligent bodies such as manufacturability analysis, processing feature recognition and the like and an infrastructure layer, solves the problem of short plate of LLM space geometric analysis by a ray method, complements manufacturing expertise by relying on a structural technology knowledge base, combines NX platform simulation verification and manual feedback closed loop, realizes automatic and accurate generation from CAD model to technological scheme, shortens planning period, reduces processing risk and expert dependence, and adapts to the processing requirements of complex parts in multiple fields.

Inventors

CAI MAOLIN
TONG XIAOMENG
NIU SHUAI
NING FANGWEI
SHI YAN

Assignees

北京航空航天大学

Dates

Publication Date: 20260505
Application Date: 20251231

Claims (8)

1. The intelligent planning and simulation verification system for the processing technology based on the LLM multi-agent is characterized in that a layered architecture design is adopted, and the intelligent planning and simulation verification system comprises a core coordination layer, a professional execution layer and an infrastructure layer, wherein the intelligent planning and simulation verification of the processing technology are realized by cooperation of the layers; The core coordination layer is a central LLM agent and is used for receiving CAD models and processing scheme requirements input by users and completing task understanding and intelligent decomposition, professional agent scheduling, decision integration and natural language interaction; The professional execution layer comprises four independent professional agents, namely manufacturability analysis agents, processing characteristic identification agents, processing technology planning agents and simulation and verification agents, wherein each professional agent executes specific field tasks based on modularized design; The infrastructure layer comprises a memory system and a feedback system, wherein the memory system is used for storing and managing process knowledge, and the feedback system is used for realizing continuous optimization of the system by combining manual expert feedback; the central LLM agent integrates design strategies through a formula P=f (F, C, H), wherein F is a processing function requirement input by a user, C is a manufacturing constraint preset by a system, H is manual expert feedback transmitted by a feedback system, and the realization logic of the formula is that F, C, H is quantized into decision parameters through a weighted summation algorithm to generate a process planning optimization instruction, so that a closed-loop workflow of task understanding, decomposition, execution, verification and optimization is formed.
2. The intelligent planning and simulation verification system for the processing technology based on the LLM multi-agent according to claim 1, wherein the manufacturability analysis agent adopts a manufacturability analysis algorithm based on a ray method, and specifically comprises the steps of adopting six standard processing directions of +X, -X, +Y, -Y, +Z and-Z to carry out ray projection on all geometric surfaces of a CAD model; Determining the manufacturability of the model by judging whether each geometric surface has an effective machining direction or not, wherein if the geometric surface has a surface without an effective machining direction, the model is judged to be non-manufacturable, otherwise, the model is judged to be manufacturable; And feeding back the manufacturability evaluation result to the central LLM intelligent agent, and providing basis for subsequent task allocation.
3. The intelligent planning and simulation verification system for the processing technology based on the LLM multi-agent is characterized in that the processing characteristic recognition agent realizes manufacturing characteristic extraction based on an improved ray shielding algorithm, and specifically comprises the steps of taking an effective processing direction verified by a manufacturability analysis agent as a core constraint, removing ray projection data of an ineffective processing direction, only preserving ray shielding information in the effective direction, and carrying out characteristic recognition by combining a topological structure of a geometric surface; the processing feature recognition intelligent body is used for automatically recognizing processing feature types including small holes, large holes, open type through hole pockets, closed type through hole pockets, open type blind hole pockets and closed type blind hole pockets; And the structural conversion from the CAD model to the manufacturing feature is completed by judging the matching relation between the ray shielding position in the effective machining direction and the radius of the cutter, and the feature data is transmitted to the machining process planning intelligent body.
4. The intelligent planning and simulation verification system for machining processes based on LLM multi-agent according to claim 3, wherein the intelligent planning and simulation verification system for machining processes is characterized in that the intelligent planning and simulation verification system for machining processes is based on CAM modules, the secondary development has the core content that a feature-process mapping interface is developed, feature data output by machining feature identification agents are automatically matched with parameters of a structural process knowledge base, input parameters comprise machining directions, machining strategies, machining feature surface sets, cutters and cutting parameters, the machining directions and the machining feature surface sets are provided by the machining feature identification agents, the machining strategies and the cutting parameters are provided by the structural process knowledge base, and the machining process planning agents generate detailed machining process routes through automatic mapping of features to processes, and the machining process routes comprise rough machining, semi-finishing and finishing sequences.
5. The intelligent planning and simulation verification system for the processing technology based on the LLM multi-agent system is characterized in that a structural technology knowledge base adopts JSON format organization data and is designed for a layered structure, the intelligent planning and simulation verification system comprises a feature classification module, a special feature module and a general parameter module, the feature classification module divides a processing object into six major categories including small hole features, large hole features, open cavity features, closed cavity features, open bottom features and closed bottom features, each category of features corresponds to a preset processing strategy combination, the processing strategy comprises drilling, cavity milling, plane milling and contour milling, and the general parameter module comprises cutting parameters, allowance setting, a feed strategy and a cutter selection rule, and the cutting parameters comprise feed speed, spindle rotation speed, axial cutting depth and radial cutting depth.
6. The intelligent planning and simulation verification system for the processing technology based on the LLM multi-agent according to claim 5, wherein the simulation and verification agent is developed based on a CAM platform and an OpenAPI, and the core content of the development is that a virtual manufacturing interface is called through the API, and a tool path verification module, a collision detection module, an over-cut detection module and a residual quantity evaluation module are integrated; The virtual manufacturing technology is used for verifying whether a cutter path accords with machining constraint, whether collision exists between a cutter and a model/equipment, whether a cutting path exceeds a part boundary and whether the residual quantity after machining is less than or equal to 0.1mm; and generating a comprehensive verification report containing simulation data, risk assessment results and optimization suggestions, visually presenting the comprehensive verification report to a process engineer, and feeding back the verification results to a central LLM agent to form closed-loop optimization of the process scheme.
7. The intelligent LLM multi-agent based process planning and simulation verification system according to claim 6, wherein the infrastructure layer is: the memory system is used for storing a process knowledge base, a historical process scheme, intelligent agent interaction data and manual feedback records; The feedback system guides the intelligent agent to select the optimal scheme when the process decision is ambiguous through manual expert input, and achieves knowledge enhancement when the system reaches the capability boundary, so that the decision accuracy of the intelligent agent is continuously optimized.
8. An intelligent planning and simulation verification method for a machining process based on the system of any one of claims 1 to 7, which is characterized by comprising the following steps: S1, inputting a CAD model (single or multiple) and machining scheme requirements (including machining precision, material types and equipment types) by a user, analyzing the requirements by a central LLM intelligent agent through a task understanding module, and splitting the requirements into four subtasks through a task decomposition module, wherein the four subtasks comprise manufacturability analysis, machining feature identification, process planning and simulation verification; S2, the manufacturability analysis agent invokes a ray method algorithm to evaluate the manufacturability of the CAD model, outputs a 'manufacturable' or 'non-manufacturable' result, terminates the flow of the non-manufacturable model, feeds back a user through a natural language interaction module, and enables the manufacturable model to enter the next step; S3, extracting manufacturing characteristics by the processing characteristic identification agent through an improved ray shielding algorithm and FeatureRecognizer modules (the core function is characteristic topological relation modeling) based on the effective processing direction, and establishing characteristic topological relation; S4, the processing technology planning agent searches a structural technology knowledge base through a feature-technology mapping interface, and matches processing strategies and parameters according to the extracted feature types to generate detailed technology routes and cutter paths; S5, performing virtual simulation verification on the process scheme by a simulation and verification intelligent body, identifying potential defects (collision, over-cutting and residual exceeding standard) and generating an optimization suggestion; And S6, if the artificial feedback system inputs feedback information, the central LLM intelligent agent integration simulation verification result adjusts the process parameters according to the formula P=f (F, C and H), and if no feedback is performed, the executable NC codes and the process report are directly output.

Description

Intelligent planning and simulation verification system for processing technology based on LLM multiple agents Technical Field The invention relates to the technical field of intelligent manufacturing, in particular to a processing technology intelligent planning and simulation verification system based on LLM multiple agents. Background Along with the transformation of the manufacturing industry to intellectualization and digitalization, the process planning is a core link for connecting the design and the manufacture of products, and directly determines the production quality and the production efficiency. At present, a computer-aided process planning (CAPP) system is mainly adopted in the industry to solve the process planning problem, but the existing CAPP system still has the remarkable defects of limited intelligence degree, difficulty in realizing comprehensive automation of process decision, poor flexibility, insufficient suitability for complex geometric feature parts, low reliability, dependence on manual intervention to correct the process defect, high use threshold and abundant expertise and practical experience of process engineers. In recent years, large Language Models (LLMs) such as the GPT family of OpenAI exhibit excellent capabilities in terms of natural language understanding, logical reasoning and content generation, providing a new technological path for intelligent upgrades of manufacturing industries. However, LLM faces the following inherent bottlenecks when actually used in the field of mechanical fabrication: because of insufficient space reasoning and geometric cognition capability, LLM is difficult to directly process tasks such as manufacturability analysis of CAD models, processing feature recognition and the like which depend on real-time space visualization; lacking an intuitive understanding of manufacturing process constraints, material properties, and process feasibility, it is difficult to meet the professional needs of process planning with text training alone. Disclosure of Invention The invention aims to provide an intelligent planning and simulation verification system for a processing technology based on LLM multi-agent, so as to solve the problems in the background technology. In order to achieve the aim, the intelligent planning and simulation verification system for the processing technology based on the LLM multi-agent adopts a layered architecture design, and comprises a core coordination layer, a professional execution layer and an infrastructure layer, wherein the intelligent planning and simulation verification of the processing technology are cooperatively achieved by the layers; The core coordination layer is a central LLM agent and is used for receiving CAD models and processing scheme requirements input by users and completing task understanding and intelligent decomposition, professional agent scheduling, decision integration and natural language interaction; The professional execution layer comprises four independent professional agents, namely manufacturability analysis agents, processing characteristic identification agents, processing technology planning agents and simulation and verification agents, wherein each professional agent executes specific field tasks based on modularized design; The infrastructure layer comprises a memory system and a feedback system, wherein the memory system is used for storing and managing process knowledge, and the feedback system is used for realizing continuous optimization of the system by combining manual expert feedback; The central LLM agent integrates design strategies through a formula P=f (F, C, H), wherein F is a processing function requirement input by a user, C is a manufacturing constraint preset by a system (comprising equipment model, material property and processing precision requirement), H is manual expert feedback transmitted by a feedback system, and the realization logic of the formula is that F, C, H is quantized into decision parameters through a weighted summation algorithm to generate a process planning optimization instruction, so that a closed-loop workflow of task understanding, decomposing, executing, verifying and optimizing is formed. Preferably, the manufacturability analysis agent adopts a manufacturability analysis algorithm based on a ray method, and specifically comprises the steps of adopting six standard machining directions of +X, -X, +Y, -Y, +Z and-Z to carry out ray projection on all geometric surfaces of a CAD model; Determining the manufacturability of the model by judging whether each geometric surface has an effective machining direction or not, wherein if the geometric surface has a surface without an effective machining direction, the model is judged to be non-manufacturable, otherwise, the model is judged to be manufacturable; And feeding back the manufacturability evaluation result to the central LLM intelligent agent, and providing basis for subsequent task allocation.