Multi-agent-based truss structure autonomous design system and method

Abstract

The invention provides a truss structure autonomous design system and method based on multiple agents. A layered multi-agent architecture is adopted, a general control agent is taken as a core, and a high-level design intention is analyzed and a plurality of downstream special agents are arranged for cooperative work; the whole design process realizes automatic closed loop through a graph-based workflow engine; the core innovation of the method is a hybrid intelligent optimization algorithm of LLM heuristic exploration and genetic algorithm accurate fine tuning, and the algorithm is driven by an original and quantifiable design quality scoring system to efficiently carry out analysis-diagnosis-optimization iteration so as to cooperatively improve the safety and economy of the design. The design time can be shortened from several hours to several minutes, about 7.0% of material consumption is saved on average, it is ensured that the scheme completely conforms to industry specifications, the design efficiency, economy and reliability are greatly improved, and meanwhile the use threshold of professional design software is remarkably reduced through natural language interaction.

Description

Technical Field

[0001] This invention relates to the field of intelligent design and automation of building structures. Specifically, it relates to an end-to-end autonomous design framework, method and system for truss structures from conceptual design to construction drawing output by utilizing artificial intelligence technology, particularly by constructing a hierarchical multi-agent collaborative system and combining it with hybrid intelligent optimization algorithms. Background Technology

[0002] Truss structures are widely used in the construction industry, but their traditional design methods have significant technical limitations. Firstly, the design process heavily relies on manual operation, requiring manual model building, data transfer, and repeated adjustments between multiple independent software programs (such as CAD and finite element analysis programs). This results in lengthy design cycles, low efficiency, and high demands on engineers' professional skills. To improve efficiency, academia and industry have introduced traditional optimization methods such as genetic algorithms (GA). However, these methods are highly dependent on initial parameters, exhibit unstable convergence performance, have high computational costs, and are difficult to integrate with complex engineering specifications (such as construction regulations), limiting their application in practical engineering. In recent years, artificial intelligence (AI) technology, especially Large Language Models (LLM), has opened up new avenues for addressing these problems. Existing research has attempted to utilize LLM for "conversational design," transforming natural language requirements into geometric models, demonstrating the potential to reduce the difficulty of software operation. Some studies have also proposed multi-agent frameworks based on LLM, using a master agent to schedule multiple specialized tools, initially demonstrating the feasibility of AI-driven complex tasks. However, current engineering design applications based on AI and LLM still have significant shortcomings: (1) Existing applications are mostly "sporadic" solutions, focusing on geometric modeling or single task optimization, failing to integrate the steps of requirement analysis, mechanical calculation, specification verification, iterative optimization, and construction drawing into a continuous automated closed loop. (2) Many methods can generate the "form" of a structure, but it is difficult to ensure its "performance" and "efficiency". That is, they cannot automatically perform rigorous mechanical property analysis and specification compliance checks, and there is a lack of reliable optimization mechanisms to ensure the safety and economy of the solution. Simply relying on LLM for optimization has the problem of "illusion", which may produce solutions that violate physical laws or engineering principles. (3) Existing research has failed to effectively integrate the heuristic and macro-reasoning advantages of LLM with the accuracy and reliability of traditional optimization algorithms to build a synergistic and complementary hybrid intelligent model.

[0003] In summary, existing technologies have significant shortcomings in achieving end-to-end, full-process, and highly automated autonomous design of truss structures. There is an urgent need for an autonomous design framework that tightly integrates the decision-making capabilities of large language models, the analytical capabilities of specialized engineering software, and reliable optimization algorithms. This framework would fundamentally address the difficulties of traditional design methods and provide a new technological path for the deep application of AI technology in traditional engineering fields. Summary of the Invention

[0004] The purpose of this invention is to provide an autonomous design system and method for truss structures based on multi-agent intelligence, in order to solve the problems mentioned in the background art, such as the low efficiency of traditional design methods, reliance on human experience, and limited optimization capabilities, as well as the shortcomings of existing AI-aided design technologies in terms of full-process integration, depth of engineering analysis, and reliability of optimization.

[0005] This invention first proposes an autonomous design system for truss structures based on multi-agent technology, comprising: The central control agent employs a large language model to receive and parse the user's natural language design requirements, decompose tasks, and plan and decide the execution flow of downstream specialized agents. Multiple specialized intelligent agents, communicatively connected to the central control intelligent agent, are used to execute specific professional and technical tasks. These specialized intelligent agents include: Natural language understanding agents are used to transform users' natural language design requirements into structured design parameters; A verification agent is used to automatically verify the rationality of the design parameters based on a built-in engineering specification database. The finite element analysis agent, based on the design parameters verified by the verification agent, is used to automatically generate analysis scripts and call the analysis engine to perform structural mechanics calculations; The diagnostic agent is used to perform compliance verification and design quality scoring on the analysis results of the finite element analysis agent. An optimized agent is used to execute a hybrid intelligent optimization algorithm to adjust design parameters; Computer-aided design intelligence agents are used to automatically generate standardized engineering drawings based on final design parameters.

[0006] Furthermore, it also includes a result verification agent, which performs a final consistency and integrity check on the output of the entire process before the final results are delivered.

[0007] The present invention further proposes a multi-agent-based autonomous design method for truss structures, employing the aforementioned multi-agent-based autonomous design system for truss structures, comprising the following steps: The central control agent receives and parses the user's natural language design requirements, decomposes the user's natural language design requirements into sub-tasks, and plans and decides the execution process of downstream specialized agents. Natural language understanding agents transform users' natural language design requirements into structured design parameters; Then it enters a closed-loop iterative process of "analysis-diagnosis-optimization" until the preset termination conditions are met; The closed-loop iterative process specifically includes: a. The finite element analysis agent automatically generates analysis scripts and calls the analysis engine to perform structural mechanics calculations based on the design parameters generated by the agent through natural language understanding; b. The diagnostic agent performs compliance verification on the analysis results of the finite element analysis agent and uses a design quality scoring system to quantitatively score the current design scheme; c. The overall control agent makes decisions based on the verification and scoring results of the diagnostic agent: If the termination condition is not met, the optimization agent will optimize and adjust the design parameters. Based on the optimized and adjusted design parameters, the process will return to step a to continue iterating. If the termination conditions are met, the computer-aided design agent will automatically generate standardized engineering drawings based on the final determined design parameters.

[0008] Furthermore, the final step involves the overall control agent calling the result verification agent to confirm the completeness of the drawings and the accuracy of the parameters. Once confirmed, the system presents the final design report and CAD drawing files to the user, and the task is completed.

[0009] Furthermore, the optimized agent employs a hybrid intelligent optimization algorithm, which includes: The overall control agent provides macro-level, global heuristic optimization guidance based on the diagnostic report returned by the diagnostic agent. The optimization agent invokes a genetic algorithm to perform a refined parameter search in the optimization direction given by the overall control agent.

[0010] Furthermore, the design quality score includes a safety score and an economic score; Safety rating is used to assess the structural safety of a design based on indicators including strength, stiffness, and construction specifications. Specifically, the strength verification follows the formula: ; in, σ For maximum stress, f This is the strength design value. Stiffness verification checks whether the structural deflection meets the code limits. For truss structures, the vertical deflection limit is 1 / 250 of the span, and the maximum deflection is less than or equal to L / 250, where L is the calculated span (for 2D trusses, take the span; for 3D trusses, take the span). When deflection exceeds the limit, the percentage exceeding the limit is calculated and the severity is classified. Recommended measures are "increasing the truss height" or "increasing the cross-sectional area"; stability is controlled by the slenderness ratio λ. ; in To calculate the length,i Where is the radius of gyration; according to the specifications, the slenderness ratio λ of a compression member is ≤150, and the slenderness ratio λ of a tension member is ≤300. The economic efficiency score is used to evaluate the material economy of the design scheme based on material utilization efficiency and unit steel consumption; the overall material utilization rate is calculated by averaging the stress ratio and deflection ratio. , where xs≤1.0 and xd≤1.0; The unit steel consumption is defined as the ratio of the total steel consumption W of the structure to its span L: Steel consumption per unit = W / L; The design quality score is calculated by weighting the safety score and the economic score to obtain a comprehensive quantitative score, which serves as the objective function for optimization iteration. The total score is calculated using the following formula: ; Wherein, Ss is the strength score, Sd is the stiffness score, Sc is the structural score, Su is the material utilization score, and Sm is the steel consumption efficiency score. When the strength or stiffness does not meet the specification requirements (i.e., Ss<0 or Sd<0), the DQS total score is directly judged as unqualified, to ensure that structural safety is always the primary premise in the optimization process.

[0011] Furthermore, the method uses a directed graph-based workflow engine for process orchestration and task scheduling, where the action of each dedicated agent is defined as a node, and the transition logic between nodes is dynamically decided and conditionally routed by the overall control agent based on a central state object shared and transmitted among all nodes.

[0012] Compared with the prior art, the beneficial effects of the present invention include: 1. High efficiency: This invention seamlessly integrates all aspects from understanding requirements to generating construction drawings, achieving end-to-end full-process automation. This invention can reduce design time from several hours to several minutes, save an average of about 7.0% of material usage, and ensure that the design scheme is 100% compliant with industry standards, greatly improving design efficiency, economy and reliability. At the same time, natural language interaction significantly reduces the threshold for using professional design software.

[0013] 2. High ease of use: This invention uses natural language interaction as the entry point. Users do not need to master complex and professional CAD and finite element software operations. They can drive the huge professional toolchain behind it simply through intuitive language descriptions, which significantly reduces the application threshold of professional design technology.

[0014] 3. High economy and reliability: The invention’s unique hybrid intelligent optimization strategy combines the macro-creativity of AI with the mathematical rigor of traditional algorithms. It can systematically find better design solutions while ensuring 100% compliance with specifications, achieving significant material savings, while avoiding the “illusion” risk of pure large language model (LLM) optimization.

[0015] 4. Scientific and efficient optimization process: This invention establishes a quantitative design quality assessment system (DQS), which encapsulates abstract and multi-dimensional engineering design knowledge into a single objective function that can be calculated and quantified. This provides clear guidance and termination conditions for the automated optimization process, ensuring continuous improvement in design quality. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is an architecture diagram of an autonomous truss structure design system based on multi-agent technology according to the present invention. Figure 2 This is a flowchart of the directed graph-based workflow engine in Embodiment 2 of the present invention; Figure 3 This is a diagram illustrating the design of the quality scoring system in this invention. Figure 4 This is the CAD drawing folder automatically generated by this invention; Figure 5 This is an isometric CAD drawing containing a truss structure automatically generated by the present invention. Figure 6 This is a CAD front view of the truss structure automatically generated by the present invention; Figure 7 This is a CAD side view of the truss structure automatically generated by the present invention; Figure 8 This is a CAD upper chord plan view of the truss structure automatically generated by the present invention; Figure 9 This is a CAD lower chord plan view of the truss structure automatically generated by the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1

[0019] Reference Figure 1 An autonomous design system for truss structures based on multi-agent technology, comprising: The central control agent employs a large language model to receive and parse the user's natural language design requirements, decompose tasks, and plan and decide the execution flow of downstream specialized agents. Multiple specialized intelligent agents, communicatively connected to the central control intelligent agent, are used to execute specific professional and technical tasks. These specialized intelligent agents include: Natural language understanding agents are used to transform users' natural language design requirements into structured design parameters; A verification agent is used to automatically verify the rationality of the design parameters based on a built-in engineering specification database. The finite element analysis agent, based on the design parameters verified by the verification agent, is used to automatically generate analysis scripts and call the analysis engine to perform structural mechanics calculations; The diagnostic agent is used to perform compliance verification and design quality scoring on the analysis results of the finite element analysis agent. An optimized agent is used to execute a hybrid intelligent optimization algorithm to adjust design parameters; Computer-aided design intelligence agents are used to automatically generate standardized engineering drawings based on final design parameters.

[0020] As a preferred implementation, it also includes a result verification agent for performing a final consistency and integrity check on the output of the entire process before the final result is delivered. Example 2

[0021] Reference Figure 1This paper presents an autonomous design system for truss structures based on a multi-agent architecture. The system is built upon a hierarchical multi-agent collaborative architecture. This architecture includes a central control agent acting as the system's "brain," and multiple specialized agents responsible for executing specific technical tasks. In this embodiment, the central control agent preferably employs a Large Language Model (LLM) with powerful natural language understanding, code generation, and logical reasoning capabilities, such as the DeepSeek-V2 model. The core responsibility of the central control agent is to understand the user's high-level design intent, decompose complex design tasks into a series of ordered sub-tasks, and plan, orchestrate, and decide the execution flow of downstream specialized agents.

[0022] The dedicated intelligent agent includes: Natural Language Understanding (NLU) agent: Responsible for receiving natural language input from the user (e.g., "Please design a roof truss for an indoor basketball court, spanning 24 meters, with a load of 5.0 kN / m"). 2 ", and extract structured design parameters from it, such as span, load, boundary conditions, truss form, etc.

[0023] The verification agent incorporates key structural requirements and empirical parameters from national design standards (such as GB50017-2017 "Standard for Steel Structure Design"). It is responsible for verifying the rationality of parameters extracted by the Natural Language Understanding (NLU) agent, such as checking whether the truss height-to-span ratio is within the recommended range, and correcting unreasonable parameters or providing suggestions to the user.

[0024] Finite Element Analysis (FEA) Agent: As a wrapper for a professional tool, it automatically generates parameterized finite element analysis scripts (APDL language scripts in this embodiment) based on the current design parameters, and calls the kernel of commercial finite element analysis software (such as Ansys) to perform structural mechanics calculations, thereby obtaining stress, displacement and other response data of the structure under a given load.

[0025] The diagnostic agent is responsible for interpreting and evaluating the analysis results of the finite element analysis (FEA) agent. It performs two core tasks: First, it conducts a comprehensive code compliance check, examining whether the strength of all members, overall and local stability, and displacement of key nodes meet the code limits; second, it uses the Design Quality Score (DQS) system, which will be detailed later, to provide a comprehensive quantitative score for the current design scheme.

[0026] Optimization Agent: When the evaluation results of the diagnostic agent indicate that the current solution has room for optimization or does not meet the specification requirements, the agent is activated. It is responsible for executing the core hybrid intelligent optimization algorithm of this invention, adjusting design variables such as the geometric dimensions of the truss or the cross-section of the members to generate a better new solution.

[0027] Computer-Aided Design (CAD) Agent: In the final stage of the design process, once an optimal solution satisfying all requirements is determined, this agent is responsible for converting the final design parameters into standard engineering drawings. In this embodiment, it drives the AutoCAD software by automatically generating AutoLISP scripts to create complete construction drawings including floor plans, elevations, detailed drawings, and bills of materials. Please refer to [link / reference]. Figure 4-9 , Figure 4-9 The exhibition showcases complete construction drawings automatically generated by a computer-aided design (CAD) agent. These drawings include multiple views of the truss structure, such as isometric views, front views, side views, top chord plan, and bottom chord plan, as well as a detailed component list (including cross-sectional specifications, materials, and lengths) and complete dimensioning. They comply with GB / T drafting standards and can be directly used for engineering construction.

[0028] The result verification agent performs a final consistency and integrity check on all outputs of the entire process, especially the final CAD drawings and design reports, before the final deliverables are delivered, ensuring that everything is in perfect order.

[0029] Please see Figure 2 In this embodiment of the invention, a workflow engine based on a directed graph is used for process orchestration and task scheduling. In this embodiment, the engine is preferably implemented using the LangGraph framework. The entire design process is constructed as a directed acyclic graph (DAG), where the action of each agent is defined as a node, such as the "requirement parsing" node, the "parameter verification" node, etc. All agents share a central state object, which exists in the form of a strongly typed dictionary containing all key information such as design parameters, analysis results, and design quality score (DQS). The central control agent performs conditional judgments and decisions based on the state updates returned after each node's execution, thereby determining the next direction of the workflow and realizing a closed-loop iteration of "analysis-diagnosis-optimization". For example, when the DQS score returned by the diagnosis node is lower than a preset threshold, the central control agent will route to the optimization node; when the score meets the threshold, it will route to the CAD drawing node, ultimately ending the process.

[0030] Please see Figure 3To achieve continuous quantitative evaluation of design quality, this invention proposes a Design Quality Score (DQS) system. This system decomposes abstract design quality into two dimensions: safety and economy. The safety score (out of 60 points) is a weighted average of strength, stiffness, and structural scores, directly reflecting the degree to which the design meets mandatory code requirements. The economy score (out of 40 points) is composed of material utilization and steel efficiency, used to measure the cost-effectiveness of the solution. Through this system, any design solution can be assigned a comprehensive score from 0 to 100, thus transforming the optimization problem into an optimization process with a clearly defined objective function. Figure 3 The verification results shown indicate that the Design Quality Score (DQS) has a good negative correlation with the total steel consumption of the structure, proving the effectiveness of the scoring system as an optimization objective. Example 3

[0031] Reference Figure 1 and Figure 2 A method for autonomous design of truss structures based on multi-agent intelligence, employing the autonomous design system for truss structures based on multi-agent intelligence described in Embodiment 1 or 2 above, includes the following steps: The central control agent receives and parses the user's natural language design requirements, decomposes the user's natural language design requirements into sub-tasks, and plans and decides the execution process of downstream specialized agents. Natural language understanding agents transform users' natural language design requirements into structured design parameters; Then it enters a closed-loop iterative process of "analysis-diagnosis-optimization" until the preset termination conditions are met; The closed-loop iterative process specifically includes: a. The finite element analysis agent automatically generates analysis scripts and calls the analysis engine to perform structural mechanics calculations based on the design parameters generated by the agent through natural language understanding; b. The diagnostic agent performs compliance verification on the analysis results of the finite element analysis agent and uses a design quality scoring system to quantitatively score the current design scheme; c. The overall control agent makes decisions based on the verification and scoring results of the diagnostic agent: If the termination condition is not met, the optimization agent will optimize and adjust the design parameters. Based on the optimized and adjusted design parameters, the process will return to step a to continue iterating. If the termination conditions are met, the computer-aided design (CAD) agent will automatically generate standardized engineering drawings based on the final determined design parameters.

[0032] As a preferred implementation, the system also includes a final step whereby the central control agent calls the result verification agent to make a final confirmation of the completeness of the drawings and the accuracy of the parameters. Once the confirmation is complete, the system presents the final design report and CAD drawing files to the user, and the task is finished.

[0033] In a preferred embodiment, the optimizing agent employs a hybrid intelligent optimization algorithm, which includes: The overall control agent provides macro-level, global heuristic optimization guidance based on the diagnostic report returned by the diagnostic agent. The optimization agent invokes a genetic algorithm to perform a refined parameter search in the optimization direction given by the overall control agent.

[0034] In a preferred embodiment, the design quality score includes a safety score and an economic score; Safety rating is used to assess the structural safety of a design based on indicators including strength, stiffness, and construction specifications. Specifically, the strength verification follows the formula: ; in, σ For maximum stress, f This is the strength design value. Stiffness verification checks whether the structural deflection meets the code limits. For truss structures, the vertical deflection limit is 1 / 250 of the span, and the maximum deflection is less than or equal to L / 250, where L is the calculated span (for 2D trusses, take the span; for 3D trusses, take the span). When deflection exceeds the limit, the percentage exceeding the limit is calculated and the severity is classified. Recommended measures are "increasing the truss height" or "increasing the cross-sectional area"; stability is controlled by the slenderness ratio λ. ; in To calculate the length, i Where is the radius of gyration; according to the specifications, the slenderness ratio λ of a compression member is ≤150, and the slenderness ratio λ of a tension member is ≤300. The economic efficiency score is used to evaluate the material economy of the design scheme based on material utilization efficiency and unit steel consumption; the overall material utilization rate is calculated by averaging the stress ratio and deflection ratio. , where xs≤1.0 and xd≤1.0; The unit steel consumption is defined as the ratio of the total steel consumption W of the structure to its span L: Steel consumption per unit = W / L; The design quality score is calculated by weighting the safety score and the economic score to obtain a comprehensive quantitative score, which serves as the objective function for optimization iteration. The total score is calculated using the following formula: ; Wherein, Ss is the strength score, Sd is the stiffness score, Sc is the structural score, Su is the material utilization score, and Sm is the steel consumption efficiency score. When the strength or stiffness does not meet the specification requirements (i.e., Ss<0 or Sd<0), the DQS total score is directly judged as unqualified, to ensure that structural safety is always the primary premise in the optimization process.

[0035] In one preferred embodiment, the method uses a directed graph-based workflow engine for process orchestration and task scheduling, where the action of each dedicated agent is defined as a node, and the transition logic between nodes is dynamically decided and conditionally routed by the central control agent based on a central state object shared and transmitted among all nodes. Example 4

[0036] Reference Figure 1 and Figure 2 A method for autonomous design of truss structures based on multi-agent intelligence, employing the autonomous design system for truss structures based on multi-agent intelligence described in Embodiment 1 or 2 above, includes the following steps: Step S1: The user inputs design requirements into the system via natural language: "Please design a roof for an industrial plant used for material storage, using planar trusses with a span of 30 meters, a spacing of 6 meters, and a roof dead load of 0.5 kN / m." 2 The live load is 0.4 kN / m. 2 。

[0037] Step S2: The central control agent receives the request and initiates the workflow. First, it calls NLU agent 101, which successfully extracts the following structured parameters: {Structure type: planar truss, span: 30m, spacing: 6m, dead load: 0.5kN / m} 2 Live load: 0.4 kN / m 2}

[0038] Step S3: The Natural Language Understanding (NLU) agent extracts the generated structured parameters and passes them to the validation agent. Based on its built-in normative knowledge, the validation agent determines that the reasonable height-to-span ratio of the truss is between 1 / 8 and 1 / 12, and recommends an initial height of 3.0m accordingly. Simultaneously, it selects an initial truss type (e.g., Pratt type) and an initial library of member sections based on experience. These parameters are then updated in the central state object.

[0039] Step S4: The main control agent invokes the Finite Element Analysis (FEA) agent. This agent reads the parameters from the state object and automatically generates an Ansys Parametric Design Language (APDL) script. This script defines the nodes, elements, material properties, load cases, and solution commands. Subsequently, the agent silently executes Ansys in the background to perform mechanical analysis and writes the calculated stresses of each member, nodal displacements, and other results back to the state object.

[0040] Step S5: The main control agent calls the diagnostic agent. The diagnostic agent reads the analysis results and performs standard verification one by one. The verification reveals that the stress of some upper chord members exceeds the material strength design value, and the maximum deflection also exceeds the standard limit L / 240. Subsequently, the diagnostic agent calls the Design Quality Score (DQS) system for scoring. Due to the existence of non-compliance items, the safety score is D (unqualified), and the final comprehensive design quality score (DQS) is 45 points. The diagnostic report ("Upper chord instability, deflection exceeding limit") and the design quality score (DQS) are updated to the state object.

[0041] Step S6: The overall control agent checks the status object and finds that the Design Quality Score (DQS) (45 points) is far below the preset pass threshold (e.g., 75 points). Therefore, it decides to enter the optimization process and calls the optimization agent.

[0042] Step S7: The optimization agent initiates the hybrid intelligent optimization algorithm. The master control agent first uses heuristic reasoning based on the diagnostic report: "The instability and excessive deflection of the upper chord indicate insufficient overall structural stiffness. The truss height should be increased or the cross-sections of the upper and lower chords should be enlarged." Based on this, the optimization agent adjusts the search space and initial population of the genetic algorithm, and then begins iterative optimization. In each generation, steps S4 and S5 are repeated until a new scheme with a significantly improved Design Quality Score (DQS) is found.

[0043] Step S8: After several rounds of closed-loop iteration of "analysis-diagnosis-optimization", the optimization agent found a solution with a Design Quality Score (DQS) of 88 (Grade A). All indicators of this solution meet the specification requirements, and the steel consumption is reduced by 8.2% compared to the initial solution. The overall control agent determines that the termination condition has been met and decides to end the optimization loop.

[0044] Step S9: The central control agent invokes the computer-aided design (CAD) agent. This agent reads the design parameters of the final scheme, automatically generates an AutoLISP script, and calls AutoCAD to execute the script. Within seconds, it generates a complete set of truss construction drawings that conform to drafting standards, including multiple views such as isometric views, front views, side views, upper chord plan views, and lower chord plan views, as well as a detailed component list (including cross-sectional specifications, materials, and lengths) and complete dimensioning.

[0045] Step S10: The central control agent calls the result verification agent to perform final confirmation of the completeness of the drawings and the accuracy of the parameters. After confirmation, the system presents the final design report and CAD drawing files to the user, and the task is completed.

[0046] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-agent-based autonomous design system for truss structures, characterized in that: include: The central control agent employs a large language model to receive and parse the user's natural language design requirements, decompose tasks, and plan and decide the execution flow of downstream specialized agents. Multiple specialized intelligent agents, communicatively connected to the central control intelligent agent, are used to execute specific professional and technical tasks. These specialized intelligent agents include: Natural language understanding agents are used to transform users' natural language design requirements into structured design parameters; A verification agent is used to automatically verify the rationality of the design parameters based on a built-in engineering specification database. The finite element analysis agent, based on the design parameters verified by the verification agent, is used to automatically generate analysis scripts and call the analysis engine to perform structural mechanics calculations; The diagnostic agent is used to perform compliance verification and design quality scoring on the analysis results of the finite element analysis agent. An optimized agent is used to execute a hybrid intelligent optimization algorithm to adjust design parameters; Computer-aided design intelligence agents are used to automatically generate standardized engineering drawings based on final design parameters.

2. The autonomous design system for truss structures based on multi-agent technology as described in claim 1, characterized in that: It also includes a result verification agent, which performs a final consistency and integrity check on the outputs of the entire process before the final results are delivered.

3. A multi-agent-based autonomous design method for truss structures, characterized in that: The autonomous design system for truss structures based on multi-agent technology as described in claim 1 or 2 includes the following steps: The central control agent receives and parses the user's natural language design requirements, decomposes the user's natural language design requirements into sub-tasks, and plans and decides the execution process of downstream specialized agents. Natural language understanding agents transform users' natural language design requirements into structured design parameters; Then it enters a closed-loop iterative process of "analysis-diagnosis-optimization" until the preset termination conditions are met; The closed-loop iterative process specifically includes: a. The finite element analysis agent automatically generates analysis scripts and calls the analysis engine to perform structural mechanics calculations based on the design parameters generated by the agent through natural language understanding; b. The diagnostic agent performs compliance verification on the analysis results of the finite element analysis agent and uses a design quality scoring system to quantitatively score the current design scheme; c. The overall control agent makes decisions based on the verification and scoring results of the diagnostic agent: If the termination condition is not met, the optimization agent will optimize and adjust the design parameters. Based on the optimized and adjusted design parameters, the process will return to step a to continue iterating. If the termination conditions are met, the computer-aided design agent will automatically generate standardized engineering drawings based on the final determined design parameters.

4. The autonomous design method for truss structures based on multi-agent systems as described in claim 3, characterized in that: The final step also includes the overall control agent calling the result verification agent to make a final confirmation of the completeness of the drawings and the accuracy of the parameters. After the confirmation is correct, the system presents the final design report and CAD drawing files to the user, and the task is completed.

5. The autonomous design method for truss structures based on multi-agent systems as described in claim 3, characterized in that: The optimized agent employs a hybrid intelligent optimization algorithm, which includes: The overall control agent provides macro-level, global heuristic optimization guidance based on the diagnostic report returned by the diagnostic agent. The optimization agent invokes a genetic algorithm to perform a refined parameter search in the optimization direction given by the overall control agent.

6. The autonomous design method for truss structures based on multi-agent systems as described in claim 3, characterized in that: The design quality score includes a safety score and an economic score; Safety rating is used to assess the structural safety of a design based on indicators including strength, stiffness, and construction specifications. Specifically, the strength verification follows the formula: ； in, σ For maximum stress, f This is the strength design value. Stiffness verification checks whether the structural deflection meets the code limits. For truss structures, the vertical deflection limit is 1 / 250 of the span, and the maximum deflection is less than or equal to L / 250, where L is the calculated span (for 2D trusses, take the span; for 3D trusses, take the span). When deflection exceeds the limit, the percentage exceeding the limit is calculated and the severity is classified. Recommended measures are "increasing the truss height" or "increasing the cross-sectional area"; stability is controlled by the slenderness ratio λ. ； in To calculate the length, i Where is the radius of gyration; according to the specifications, the slenderness ratio λ of a compression member is ≤150, and the slenderness ratio λ of a tension member is ≤300. The economic efficiency score is used to evaluate the material economy of the design scheme based on material utilization efficiency and unit steel consumption; the overall material utilization rate is calculated by averaging the stress ratio and deflection ratio. , where xs≤1.0 and xd≤1.0; The unit steel consumption is defined as the ratio of the total steel consumption W of the structure to its span L: Steel consumption per unit = W / L; The design quality score is calculated by weighting the safety score and the economic score to obtain a comprehensive quantitative score, which serves as the objective function for optimization iteration. The total score is calculated using the following formula: ； Wherein, Ss is the strength score, Sd is the stiffness score, Sc is the structural score, Su is the material utilization score, and Sm is the steel consumption efficiency score. When the strength or stiffness does not meet the specification requirements (i.e., Ss<0 or Sd<0), the DQS total score is directly judged as unqualified, to ensure that structural safety is always the primary premise in the optimization process.

7. The autonomous design method for truss structures based on multi-agent systems as described in claim 3, characterized in that: The method uses a directed graph-based workflow engine for process orchestration and task scheduling, where the action of each dedicated agent is defined as a node, and the transition logic between nodes is dynamically decided and conditionally routed by the overall control agent based on a central state object shared and transmitted among all nodes.

Images