An engineering design method and system based on non-structured communication data and multi-agent cooperation

Abstract

The application discloses an engineering design method and system based on multi-agent cooperation of non-structure communication data, which is suitable for industrial waste gas treatment equipment and chemical engineering design scene. The method constructs a multi-agent cooperation system containing engineer Agent, architect Agent, quality guarantee Agent, developer Agent and project management Agent, each intelligent agent is configured with a professional knowledge model and realizes communication cooperation through a cooperation center; according to a preset process, phased cooperation design is performed, multi-agent parallel review is carried out at key nodes, unified review conclusions are formed through conflict identification, grading and targeted solution strategies; based on the review results, design iteration optimization is carried out, multi-dimensional cooperation verification is performed after development, and finally reusable knowledge is extracted and stored in a structured manner. The application effectively integrates multi-field professional knowledge, realizes multi-angle design review and intelligent conflict resolution, significantly improves design efficiency and quality, reduces manual intervention, and realizes continuous accumulation and reuse of design experience.

Description

Technical Field

[0001] This invention relates to the fields of computer-aided design and artificial intelligence, specifically to an engineering design method and system that employs multiple intelligent agents with different professional roles to work collaboratively, particularly to the application of semantic parsing of unstructured communication data in industrial equipment engineering projects and driving multi-agent collaborative engineering design based on unstructured communication data. Background Technology

[0002] As engineering projects become increasingly complex, traditional design models involving single engineers or small teams face challenges such as low efficiency and difficulty in ensuring quality. In recent years, the development of artificial intelligence technology has brought new opportunities for engineering design automation; however, existing technologies, which primarily rely on human team collaboration, single AI-assisted design, and workflow automation systems, still have many shortcomings.

[0003] While human teamwork can ensure design quality, it has the following drawbacks: human communication and coordination are time-consuming, making it difficult to shorten the design cycle; multiple engineers from different disciplines are required, resulting in high labor costs; design quality heavily relies on individual experience and is difficult to standardize; project experience is difficult to accumulate and reuse, and personnel changes lead to knowledge loss.

[0004] Existing workflow automation systems are primarily based on predefined rules, which present the following problems: poor flexibility, fixed rules that make it difficult to adapt to new situations; insufficient intelligence, lacking reasoning and decision-making capabilities; simple collaboration mechanisms, with multiple modules lacking genuine collaboration and merely executing sequentially; and no learning ability, unable to learn from and improve from historical projects. Furthermore, a large amount of design input exists in the form of emails, P&IDs, bills of materials, etc., making it difficult to automatically integrate into the engineering design process.

[0005] Existing AI-assisted design tools (such as code generation assistants) typically use a single AI model to provide suggestions to engineers. The main problems with this approach include: 1. Knowledge limitations: A single AI model has a limited knowledge base and cannot simultaneously master multiple professional fields such as engineering, software, and quality management; 2. Lack of multi-perspective review: Design outcomes are evaluated from only a single perspective, easily overlooking potential problems. Summary of the Invention

[0006] The purpose of this invention is to provide an engineering design method and system based on multi-agent collaboration using unstructured communication data. This method can significantly improve design efficiency while ensuring design quality, effectively integrate knowledge from different professional fields, establish a systematic multi-angle design review mechanism, and enable automatic learning and cross-project reuse of design experience.

[0007] To achieve the above objectives, the present invention provides an engineering design method for multi-agent collaboration based on unstructured communication data, comprising the following steps: S1: Collection and management of unstructured communication data, connecting with enterprise emails and project documents, performing semantic parsing and entity extraction, and constructing a project knowledge graph as input for multiple agents; S2: Construct a multi-agent collaborative system, which consists of multiple intelligent agents with different professional roles. Each intelligent agent is configured with a knowledge model for a corresponding professional field. The knowledge model includes at least domain rule information, historical experience case information, and a reasoning processing module for generating professional decision results. Each intelligent agent communicates and collaborates through a collaboration center. S3: Execute a phased collaborative design process. Divide the engineering design process into multiple design phases according to the preset engineering design process, and let the intelligent agents of different professional roles lead the design tasks of the corresponding phases to generate phased design results. S4: Perform multi-agent collaborative review. At the preset node of the design phase, multiple intelligent agents conduct parallel reviews of the design results of the current phase based on their respective professional domain knowledge models, generate corresponding review opinions, perform consistency analysis on the collected review opinions, identify inconsistencies or conflicts, and adopt corresponding conflict handling strategies according to the severity of the conflict to form a unified review conclusion. S5: Design iteration and optimization. Based on the review conclusion, determine whether the current design result meets the preset pass conditions. If the pass conditions are not met, trigger the corresponding intelligent agent to modify and optimize the design result, and repeat the collaborative review. S6: Implement development. Once the design results have passed the review, the corresponding intelligent agent will execute the engineering implementation or development operations to generate the final engineering design results. S7: Perform multi-agent collaborative verification. After development is completed, each intelligent agent verifies the quality of the design and implementation. S8: Collaborative knowledge extraction and storage: Extract reusable design knowledge, problem solutions, or collaborative experience from the multi-agent collaborative design and review process, and store them in a structured form for subsequent engineering design tasks.

[0008] Furthermore, the intelligent agents include an engineer agent, an architect agent, a quality assurance agent, a developer agent, and a project management agent, wherein: Engineer Agent: Responsible for defining functional modules, developing engineering specifications, and determining technical constraints; Architect Agent: Responsible for designing system architecture, defining module interfaces, and planning data flow; Quality Assurance Agent: Responsible for developing testing strategies, defining acceptance criteria, and performing quality assessments; Developer Agent: Responsible for developing and implementing specific engineering solutions, performing process calculations and simulations, and verifying unit operations; Project Management Agent: Responsible for coordinating the overall progress, assigning tasks, and tracking milestones.

[0009] Furthermore, each of the intelligent agents is configured with a professional knowledge model, which includes: Domain rule base: Stores engineering rules, design specifications, and constraints for this professional domain; Experience Case Library: Stores historical design cases in this professional field and their effect evaluations; Inference engine: An algorithmic module that performs reasoning and decision-making based on domain rules and cases; Each intelligent agent exchanges information through a collaboration center, which includes: Shared knowledge base: Stores common knowledge and best practices across projects; Task queue: Manages design tasks and their dependencies at each stage; Message bus: Enables asynchronous communication between intelligent agents.

[0010] Furthermore, following the predetermined engineering design process, different intelligent agents lead each stage, including: Requirements Analysis Phase: The project management agent receives user requirements, analyzes the characteristics of the requirements, formulates an overall project plan, identifies key milestones, and assigns tasks to other intelligent agents; Engineering design phase: Based on the project plan, the engineer agent defines functional modules. Each module includes: module name and description, engineering requirements and technical specifications, input and output parameters, engineering constraints, estimated complexity and workload; Architecture design phase: The architect agent designs the system architecture based on functional modules, including: module division and organization structure, interface definition between modules, data structure design, data flow and control flow design; Test strategy phase: The quality assurance agent develops the test plan, including: unit test cases, integration test scenarios, system test plan and acceptance criteria.

[0011] Furthermore, the multi-agent collaborative review includes the following iterative steps during the engineering design phase: (a) Each intelligent agent independently reviews the functional module division scheme and generates opinions; (b) The collaboration center identifies conflicts in the review opinions; (c) For moderate conflicts, a voting mechanism based on professional weights is used to form a unified opinion; (d) Design iteration optimization is triggered based on the review conclusions until the review is passed.

[0012] Furthermore, at key junctures in the engineering design process, multi-agent collaborative reviews are conducted, with specific steps including: Parallel review: Each intelligent agent independently reviews the design results at the current stage from its own professional perspective based on its professional knowledge model, and generates review opinions; Conflict Identification: After collecting all review comments, the collaboration center uses a conflict identification algorithm to detect inconsistencies in the comments. Conflict classification: Identified conflicts are classified into minor conflicts, severe conflicts, and major conflicts for processing. Conflict resolution: Adopt appropriate resolution strategies based on the level of conflict. Furthermore, the conflict resolution strategy includes: Minor conflict resolution: An automatic negotiation mechanism is used to adjust standardized descriptions to reach consensus; Moderate conflict resolution: A final opinion is reached using a voting mechanism based on professional expertise. Major conflict resolution: Mark conflicts that require human decision-making and generate decision support reports.

[0013] Furthermore, the voting mechanism based on professional expertise weighting includes the following steps: Step 1: Analyze the professional fields involved in the conflict; Step 2: Calculate the expertise weight of each intelligent agent in this field. The formula for calculating the weight W is: W = α × E + β × H + γ × R Where E represents the agent’s level of expertise in the field (0-1). H represents the accuracy of the agent's historical decisions (0-1). R represents the degree of relevance between the conflict and the agency responsibility (0-1). α, β, and γ are weighting coefficients that satisfy α + β + γ = 1; Preferably, α=0.5, β=0.3, γ=0.2; Step 3, the opinion score of each intelligent agent = number of approvals × sum of corresponding weights; Step 4: Select the opinion with the highest score as the final decision. Furthermore, the multi-agent cooperation verification includes: Engineer Agent Verification: Whether the function meets engineering requirements, whether the calculation is accurate, and whether the parameters are reasonable; Architect Agent Verification: Whether the architecture implementation conforms to the design, whether the interfaces are clear, and whether the code quality meets the standards; Quality Assurance Agent Verification: Whether the testing is sufficient, whether the coverage meets the standard, and whether the defects have been fixed; Project management agent verification: whether the progress is on schedule, whether the documentation is complete, and whether the deliverables are complete.

[0014] Furthermore, the automatic knowledge extraction process includes: Successful design pattern extraction: Identify successful patterns in designs and extract them as structured knowledge; Problem solution extraction involves extracting reusable solutions from the problems encountered during the design process and their solutions. Collaboration experience extraction involves analyzing effective collaboration patterns and conflict resolution experiences in multi-agent collaboration processes and refining them into collaboration knowledge. Knowledge storage involves storing extracted knowledge in a structured form in a shared knowledge base.

[0015] On the other hand, the present invention also provides an engineering design system based on unstructured communication data for multi-agent collaboration, comprising multiple intelligent agents with different professional roles forming the collaborative system. These intelligent agents include an engineer agent, an architect agent, a quality assurance agent, a developer agent, and a project management agent, wherein: Engineer Agent: Responsible for defining functional modules, developing engineering specifications, and determining technical constraints; Architect Agent: Responsible for designing system architecture, defining module interfaces, and planning data flow; Quality Assurance Agent: Responsible for developing testing strategies, defining acceptance criteria, and performing quality assessments; Developer Agent: Responsible for implementing the actual development, writing code, and conducting unit tests; Project Management Agent: Responsible for coordinating the overall schedule, assigning tasks, and tracking milestones; Each of the intelligent agents is configured with a professional knowledge model, which includes: Domain rule base: Stores engineering rules, design specifications, and constraints for this professional domain; Experience Case Library: Stores historical design cases in this professional field and their effect evaluations; Inference engine: An algorithmic module that performs reasoning and decision-making based on domain rules and cases; Each intelligent agent exchanges information through a collaboration center, which includes: Shared knowledge base: Stores common knowledge and best practices across projects; Task queue: Manages design tasks and their dependencies at each stage; Message bus: Enables asynchronous communication between intelligent agents.

[0016] Compared with the prior art, the present invention has the following beneficial effects and technological advancements: 1. Through a multi-agent collaborative review mechanism, the design is comprehensively reviewed from five professional perspectives: engineering, architecture, quality, development, and management. This enables early detection of potential problems and avoidance of major design flaws. Experimental data shows that engineering projects designed using the method of this invention have a significantly lower number of design flaws compared to single AI-assisted design and traditional human team design.

[0017] 2. This invention significantly reduces manual workload and waiting time through the parallel operation and automated processing of intelligent agents. Time comparison experiments show that: to complete engineering design projects of the same complexity, the method of this invention takes less time and is more efficient; in terms of design iterations, this invention requires fewer iterations to pass review.

[0018] 3. Achieve effective integration of interdisciplinary knowledge: Through the collaboration of intelligent agents from five different disciplines, this invention can automatically integrate knowledge from multiple fields such as engineering, software, and quality without the need for manual coordination.

[0019] 4. Continuous learning and knowledge accumulation: The automatic knowledge extraction and learning mechanism of this invention enables the system to learn and improve from each project. The knowledge base also grows with the increase of projects, greatly improving the design efficiency of subsequent projects.

[0020] 5. Intelligent conflict resolution: The hierarchical conflict resolution mechanism of this invention can automatically handle most disagreements, reducing human intervention.

[0021] 6. The design process is traceable. This invention records the complete design process, including design results at each stage, review comments, conflict resolution processes, modification history, etc., which realizes complete traceability of the design process and is conducive to quality management and experience summarization. Attached Figure Description

[0022] 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.

[0023] Figure 1 This is the overall architecture diagram of the multi-agent cooperative system of the present invention.

[0024] Figure 2 This is a flowchart of the phased collaborative design process of the present invention.

[0025] Figure 3 This is a flowchart of the multi-agent collaborative review process of the present invention.

[0026] Figure 4 This is a flowchart of the conflict identification and resolution process of the present invention. Detailed Implementation

[0027] 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.

[0028] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the invention to those skilled in the art.

[0029] As shown in Figure 1, this embodiment provides an engineering design system based on unstructured communication data and multi-agent collaboration, including a collaborative system composed of multiple intelligent agents with different professional roles. The intelligent agents include an engineer agent, an architect agent, a quality assurance agent, a developer agent, and a project management agent. Each intelligent agent is configured with a professional knowledge model and exchanges information through a collaboration center.

[0030] Specifically, the professional knowledge model includes a domain rule base, an experience case base, and an inference engine; wherein, the domain rule base is used to store engineering rules, design specifications, and constraints of the professional domain; the experience case base is used to store historical design cases and their effect evaluations of the professional domain; and the inference engine is an algorithm module for reasoning and decision-making based on domain rules and cases.

[0031] The collaboration center includes a shared knowledge base, a task queue, and a message bus. The shared knowledge base stores common knowledge and best practices across projects; the task queue manages design tasks and their dependencies at each stage; and the message bus enables asynchronous communication between intelligent agents.

[0032] The responsibilities of each intelligent agent are as follows: Engineer Agent: Responsible for defining functional modules, developing engineering specifications, and determining technical constraints; Architect Agent: Responsible for designing system architecture, defining module interfaces, and planning data flow; Quality Assurance Agent: Responsible for developing testing strategies, defining acceptance criteria, and performing quality assessments; Developer Agent: Responsible for developing and implementing specific engineering solutions, performing process calculations and simulations, and verifying unit operations; Project Management Agent: Responsible for coordinating the overall schedule, assigning tasks, and tracking milestones. In this embodiment, the Project Management Agent receives user requirement input from the unstructured communication data system (including waste gas treatment requirements submitted via email, P&ID drawings provided in PDF format, and bills of materials provided in Excel spreadsheets, etc.). Through the semantic parsing module, the Agent transforms the above unstructured data into structured design parameters, stores them in a shared knowledge base, and then triggers a multi-agent collaborative design process.

[0033] Through the specialized division of labor among multiple agents and the unified scheduling of the collaboration center, the effective integration of knowledge from different professional fields is achieved, solving the problems of knowledge fragmentation and low collaboration efficiency in traditional design systems. At the same time, the professional knowledge models of each intelligent agent can be expanded and updated according to actual application scenarios, improving the system's flexibility and adaptability.

[0034] As shown in Figure 2, this embodiment provides an engineering design method based on multi-agent collaboration using unstructured communication data. It employs the engineering design system from Embodiment 1 and includes the following steps: S1: Unstructured communication data collection and management, connecting with enterprise emails and project documents, collecting and classifying attachments such as P&ID, MTR / BOM lists, etc., and establishing a communication data index by project number. Semantic parsing and entity extraction: extracting equipment tag numbers, material codes, process parameters, interlocking conditions, etc., to construct a project knowledge graph as input for multi-agent communication.

[0035] S2: Construct a multi-agent collaborative system, which consists of multiple intelligent agents with different professional roles. Each intelligent agent is configured with a knowledge model for a corresponding professional field. The knowledge model includes at least domain rule information, historical experience case information, and a reasoning processing module for generating professional decision results. Each intelligent agent communicates and collaborates through a collaboration center.

[0036] S3: Execute a phased collaborative design process. According to the preset engineering design process, the engineering design process is divided into multiple design phases. Intelligent agents with different professional roles lead the design tasks of the corresponding phases and generate phased design results.

[0037] S4: Perform multi-agent collaborative review. At the preset node of the design phase, multiple intelligent agents conduct parallel reviews of the design results of the current phase based on their respective professional domain knowledge models, generate corresponding review opinions, perform consistency analysis on the collected review opinions, identify inconsistencies or conflicts, and adopt corresponding conflict handling strategies according to the severity of the conflict to form a unified review conclusion.

[0038] S5: Design iteration and optimization. Based on the review conclusion, determine whether the current design result meets the preset pass conditions. If the pass conditions are not met, trigger the corresponding intelligent agent to modify and optimize the design result, and repeat the collaborative review.

[0039] S6: Implementation and Development. Once the design results have passed the review, the corresponding intelligent agent will execute the engineering implementation or development operations to generate the final engineering design results.

[0040] S7: Perform multi-agent collaborative verification. After development is completed, each intelligent agent verifies the quality of the design and implementation.

[0041] S8: Collaborative knowledge extraction and storage: Extract reusable design knowledge, problem solutions, or collaborative experience from the multi-agent collaborative design and review process, and store them in a structured form for subsequent engineering design tasks.

[0042] Specifically, in step S3, according to the predetermined engineering design process, different intelligent agents lead each stage, including: Requirements Analysis Phase: The project management agent receives user requirements, analyzes the characteristics of the requirements, formulates an overall project plan, identifies key milestones, and assigns tasks to other intelligent agents; Engineering design phase: Based on the project plan, the engineer agent defines functional modules. Each module includes: module name and description, engineering requirements and technical specifications, input and output parameters, engineering constraints, estimated complexity and workload; Architecture design phase: The architect agent designs the system architecture based on functional modules, including: module division and organization structure, interface definition between modules, data structure design, data flow and control flow design; Test strategy phase: The quality assurance agent develops the test plan, including: unit test cases, integration test scenarios, system test plan and acceptance criteria.

[0043] As shown in Figure 3, in step S4, at a key node in the engineering design process, a multi-agent collaborative review is performed. The specific steps include: Parallel review: Each intelligent agent independently reviews the design results at the current stage from its own professional perspective based on its professional knowledge model, and generates review opinions; Conflict Identification: After collecting all review comments, the collaboration center uses a conflict identification algorithm to detect inconsistencies in the comments. Conflict classification: Identified conflicts are classified into minor conflicts, severe conflicts, and major conflicts for processing. Conflict resolution: Adopt appropriate resolution strategies based on the level of conflict.

[0044] As shown in Figure 4, the conflict resolution strategy includes: Minor conflict resolution: An automatic negotiation mechanism is used to adjust the standardized description to reach a consensus.

[0045] Moderate conflict resolution: A voting mechanism based on professional expertise weighting is used to reach a final opinion.

[0046] Major conflict resolution: Mark conflicts that require human decision-making and generate decision support reports.

[0047] Preferably, the automatic negotiation mechanism is triggered when the conflict classifier outputs a conflict level of "minor conflict". The criteria for determining a minor conflict include: the review conclusions of each intelligent agent are substantially consistent, with only differences in expression, wording, or degree of standardization; the differences in the recommended values ​​of design parameters are within the allowable tolerance range of the engineering (e.g., the recommended values ​​of the safety factor do not differ by more than ±5%); and the conflict's impact domain is limited to within a single functional module and does not affect the interface definition between modules or the overall technical approach.

[0048] For example, in the design implementation of the exhaust gas treatment equipment, the quality assurance agent suggested "increasing the number of abnormal operating condition test scenarios to 10," while the developer agent suggested "increasing it to 8." The conflict detector normalized both suggestions to the design element of "expanding the number of test scenarios." The relative deviation was calculated to be 20% > 5%, indicating that automatic merging was not possible. However, both parties essentially agreed to expand test coverage. Therefore, the automatic negotiation mechanism determined it to be a minor conflict, taking the median value of 9 scenarios and recording the negotiation trajectory in the review log without triggering manual intervention.

[0049] A preferred voting mechanism based on expertise weighting includes the following steps: Step 1: Analyze the professional fields involved in the conflict; Step 2: Calculate the expertise weight of each intelligent agent in this field. The formula for calculating the weight W is: W = α × E + β × H + γ × R Where E represents the agent's level of expertise in the field (0-1), which is automatically extracted by the system from the agent's professional knowledge model. If the conflicting field belongs to the agent's core responsibility area (e.g., mass transfer calculation belongs to the core responsibility area of ​​the engineer agent), then E=1.0; if it belongs to a related field (e.g., mass transfer calculation belongs to the related field of the architect agent), then E=0.6; if it belongs to an unrelated field, then E=0.2. H represents the accuracy of the agent's historical decisions (0-1). The sliding window statistical method is used, where H = the number of times the most recent N decisions were ultimately adopted / N, and N = 30. The result is automatically updated after each review. R represents the degree of relevance (0-1) between the conflict and the delegated responsibility, calculated by the conflict resolution module based on the pre-configured mapping relationship between the conflict type and each delegated responsibility matrix. Core responsibility conflicts have R=1.0, related responsibilities have R=0.6, and auxiliary responsibilities have R=0.3. α, β, and γ are weighting coefficients that satisfy α + β + γ = 1; Preferably, α=0.5, β=0.3, and γ=0.2, reflecting the dominant role of professional expertise in decision-making; Step 3, the opinion score of each intelligent agent = number of approvals × sum of corresponding weights; Step 4: Select the opinion with the highest score as the final decision.

[0050] In step S6, the multi-agent cooperation verification includes: Engineer Agent Verification: Whether the function meets engineering requirements, whether the calculation is accurate, and whether the parameters are reasonable; Architect Agent Verification: Whether the architecture implementation conforms to the design, whether the interfaces are clear, and whether the code quality meets the standards; Quality Assurance Agent Verification: Whether the testing is sufficient, whether the coverage meets the standard, and whether the defects have been fixed; Project management agent verification: whether the progress is on schedule, whether the documentation is complete, and whether the deliverables are complete.

[0051] In step S7, the automatic knowledge extraction process includes: Successful design pattern extraction: Identify successful patterns in designs and extract them as structured knowledge; Problem solution extraction involves extracting reusable solutions from the problems encountered during the design process and their solutions. Collaboration experience extraction involves analyzing effective collaboration patterns and conflict resolution experiences in multi-agent collaboration processes and refining them into collaboration knowledge. Knowledge storage involves storing extracted knowledge in a structured form in a shared knowledge base.

[0052] This embodiment uses the design of a waste gas treatment device as an example to illustrate the engineering design method of the present invention in detail, specifically including the following steps: 1. Construct a multi-agent cooperative system Following the system architecture of Example 1, a collaborative system is constructed, including an engineer agent, an architect agent, a quality assurance agent, a developer agent, and a project management agent, and the professional knowledge model and collaboration center of each intelligent agent are configured. The engineer agent's domain rule base includes design specifications for waste gas treatment projects such as GB 16297-1996 "Integrated Emission Standard for Air Pollutants" and HJ 2007-2010 "Technical Policy for Pollution Prevention and Control"; the experience case library contains design parameters and operational performance data of 50 historical waste gas treatment projects; and the inference engine adopts an inference algorithm based on mass transfer theory (HTU-NTU method), fluid dynamics calculation, and chemical reaction measurement.

[0053] Preferably, taking the reasoning process based on mass transfer theory as an example, after the engineer agent's inference engine receives the structured design parameters from the project management agent, it initiates the inference calculation process based on mass transfer theory. The inference engine first calls the design specifications for waste gas scrubbing towers (such as the recommended methods for mass transfer calculation of packed towers in GB / T 151 and HJ 2007-2010) from the domain rule base, and at the same time retrieves historical project cases and their mass transfer performance data that are similar to the current operating conditions (processing capacity of 10,000 m³ / h, HCl / HF / NH3 three-component pollutants) from the experience case base.

[0054] The core reasoning process of the inference engine consists of the following steps: Step 1: Calculation of Number of Mass Transfer Units (NTUs). The inference engine calculates the number of mass transfer units for each pollutant using the logarithmic average driving force method, based on the inlet concentration of the pollutant and the target removal efficiency. For HCl (inlet concentration 100 ppm, target removal efficiency >95%), the inference engine calculates the ratio of the change in total gas phase concentration Δy to the logarithmic average driving force Δym, i.e., NTU = (y1-y2) / Δym. Where y1 is the inlet gas phase mole fraction, y2 is the outlet gas phase mole fraction, and Δym = (Δy1-Δy2) / ln(Δy1-Δy2). The inference engine performs the same NTU calculation for HF (inlet concentration 50 ppm, removal efficiency >95%) and NH3 (inlet concentration 200 ppm, removal efficiency >90%).

[0055] Step 2: Height of Transfer Unit (HTU) Calculation. The inference engine retrieves the mass transfer performance parameters of recommended packing materials from the packing database (such as the mass transfer coefficient data of Mellapak 250Y corrugated structured packing), and uses Onda correlation to calculate the gas film and liquid film mass transfer coefficients. The inference engine calculates the height of the mass transfer unit according to the formula HTU = G_m / (k_y·a), where G_m is the gas phase molar flow rate, k_y is the total mass transfer coefficient driven by the molar difference of pollutants in the gas phase, and a is the effective mass transfer area within the packing layer. The inference engine also retrieves the measured HTU values ​​of the same packing material under similar operating conditions from the experience case library in historical projects and cross-validates them with the theoretical calculation values.

[0056] Step 3: Determining the Total Height of the Packing Layer and Multi-Component Synergistic Correction. The inference engine multiplies the NTU of each pollutant by its corresponding HTU to obtain the required packing layer height H = NTU × HTU to meet the removal requirements of each pollutant. For multi-component systems, the inference engine further utilizes historical data on the synergistic removal of HCl / HF / NH3 from the empirical case library and applies a synergistic correction factor (based on the variation law of mass transfer driving force when similar pollutants coexist) to correct the calculated NTU value. The inference engine takes the maximum required packing layer height for each pollutant as the design value and adds a 10%-15% design margin.

[0057] Step 4: Pressure Drop Verification and Design Optimization. Based on the calculated packing height and tower diameter, the inference engine calculates the packing pressure drop using the Ergun equation and compares it with the pressure drop limit (<2500 Pa) in the domain rule base. If the pressure drop exceeds the limit, the inference engine automatically triggers design parameter adjustments—for example, recommending the use of packing with a higher specific surface area to reduce gas velocity, or suggesting adjustments to the liquid-to-gas ratio—and re-executes the above NTU / HTU calculations until all constraints are met.

[0058] Step 5: Output of Inference Results. The inference engine encapsulates the final calculated design parameters, such as tower diameter, packing height, pressure drop, and absorbent consumption, into structured data. Along with the calculation methods used (such as Onda correlation version, NTU calculation method source), referenced standards (such as GB 16297-1996 emission standard), and confidence scores (based on deviation analysis between theoretical calculation values ​​and historical case measured values), it outputs all of these to the knowledge model of the engineer agent for use in subsequent module design and multi-agent collaborative review.

[0059] Through the reasoning process based on mass transfer theory described above, the engineer agent can automatically complete the quantitative calculation and design optimization of the core process parameters of the waste gas scrubbing tower without human intervention, providing reliable engineering data support for subsequent architecture design, equipment selection, and multi-agent collaborative review. 2. Implement a phased collaborative design process Requirements Analysis Phase: The project management agent receives the waste gas treatment requirements input by the user, including parameters such as waste gas treatment capacity of 10,000 m³ / h, temperature of 40°C, humidity of 60%, pollutant concentration (HCl: 100 ppm, HF: 50 ppm, NH3: 200 ppm), removal efficiency requirements, and pressure drop limit of <2500 Pa. After analysis, it generates an overall project plan, determines the total duration of 3 weeks and milestones for each stage, and assigns tasks to other intelligent agents. Engineering design phase: Based on the project plan, the engineer agent defines 9 core functional modules, including InputManager, PackingDatabase, and HTUCalculator, and clarifies the engineering requirements, input and output parameters, constraints and complexity of each module; Architecture design phase: The architect (Agent) designs a layered modular architecture based on functional modules, including a data layer, computing engine layer, integration layer, and interface layer, defining the interfaces and data flow between modules; Testing strategy phase: The quality assurance agent develops a test plan, including 60 functional test cases, 12 integration test scenarios, and system test schemes, and clarifies the acceptance criteria.

[0060] 3. Perform multi-agent collaborative review Each intelligent agent reviews the design results in parallel based on its own professional knowledge model. After the collaboration center collects the review opinions, it identifies one moderate conflict (the engineer agent suggested adding a stoichiometric calculation module, which was not included in the architect agent's initial architecture). The conflict was resolved by using a voting mechanism based on professional expertise weights. The calculated professional expertise weight of the Engineer Agent in this engineering field was 0.955, which was higher than that of the Architect Agent (0.724). Ultimately, the Engineer Agent's suggestion was adopted, and a chemometrics calculation module was added.

[0061] 4. Design Iteration and Optimization The architect agent adjusts the system architecture to accommodate the new modules; the engineer agent supplements the database with high-temperature performance data; the quality assurance agent expands the abnormal operating condition test scenarios; after the modifications are completed, they are submitted for a second review, and all agents give a passing conclusion.

[0062] 5. Implement development The developer agent executes code development according to the design document, completes the code implementation of 9 modules (approximately 9400 lines of Python code), writes 103 unit test cases, and generates complete API documentation and usage instructions.

[0063] 6. Perform multi-agent collaborative verification Each intelligent agent was verified from the dimensions of functional completeness, architecture implementation, test coverage, and delivery schedule. The verification results all met the preset standards and were ultimately verified.

[0064] 7. Collaborative Knowledge Extraction and Storage Reusable knowledge, such as multi-component pollutant treatment strategies and high-frequency database query performance optimization schemes, is extracted from the project design and review process and stored in a structured form in a shared knowledge base for use in subsequent projects.

[0065] In this embodiment, the design and development of the waste gas treatment equipment was successfully completed using the engineering design method of the present invention. The design cycle was shortened by 50% compared with the traditional manual method, the number of design defects was reduced by 28%, and the calculation accuracy deviated from the actual operating data by less than 3%, which fully demonstrates the high efficiency and reliability of the present invention.

[0066] Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of 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. An engineering design method for multi-agent collaboration based on unstructured communication data, characterized in that, Includes the following steps: S1: Collection and management of unstructured communication data, connecting with enterprise emails and project documents, performing semantic parsing and entity extraction, and constructing a project knowledge graph as input for multiple agents; S2: Construct a multi-agent collaborative system, which consists of multiple intelligent agents with different professional roles. Each intelligent agent is configured with a knowledge model for a corresponding professional field. The knowledge model includes at least domain rule information, historical experience case information, and a reasoning processing module for generating professional decision results. Each intelligent agent communicates and collaborates through a collaboration center. S3: Execute a phased collaborative design process. Divide the engineering design process into multiple design phases according to the preset engineering design process, and let the intelligent agents of different professional roles lead the design tasks of the corresponding phases to generate phased design results. S4: Perform multi-agent collaborative review. At the preset node of the design stage, multiple intelligent agents conduct parallel reviews of the design results of the current stage based on their respective professional domain knowledge models, generate corresponding review opinions, perform consistency analysis on the collected review opinions, identify inconsistencies or conflicts, and adopt corresponding conflict handling strategies according to the severity of the conflict to form a unified review conclusion. S5: Design iteration and optimization. Based on the review conclusion, determine whether the current design result meets the preset pass conditions. If the pass conditions are not met, trigger the corresponding intelligent agent to modify and optimize the design result, and repeat the collaborative review. S6: Implement development. Once the design results have passed the review, the corresponding intelligent agent will execute the engineering implementation or development operations to generate the final engineering design results. S7: Perform multi-agent collaborative verification. After development is completed, each intelligent agent verifies the quality of the design and implementation. S8: Collaborative knowledge extraction and storage: Extract reusable design knowledge, problem solutions, or collaborative experience from the multi-agent collaborative design and review process, and store them in a structured form for subsequent engineering design tasks.

2. The engineering design method for multi-agent collaboration based on unstructured communication data as described in claim 1, characterized in that, The intelligent agents include engineer agents, architect agents, quality assurance agents, developer agents, and project management agents, among which: Engineer Agent: Responsible for defining functional modules, developing engineering specifications, and determining technical constraints; Architect Agent: Responsible for designing system architecture, defining module interfaces, and planning data flow; Quality Assurance Agent: Responsible for developing testing strategies, defining acceptance criteria, and performing quality assessments; Developer Agent: Responsible for developing and implementing specific engineering solutions, performing process calculations and simulations, and verifying unit operations; Project Management Agent: Responsible for coordinating the overall progress, assigning tasks, and tracking milestones.

3. The engineering design method for multi-agent collaboration based on unstructured communication data as described in claim 2, characterized in that, Each of the intelligent agents is configured with a professional knowledge model, which includes: Domain rule base: Stores engineering rules, design specifications, and constraints for this professional domain; Experience Case Library: Stores historical design cases in this professional field and their effect evaluations; Inference engine: An algorithmic module that performs reasoning and decision-making based on domain rules and cases; Each intelligent agent exchanges information through a collaboration center, which includes: Shared knowledge base: Stores common knowledge and best practices across projects; Task queue: Manages design tasks and their dependencies at each stage; Message bus: Enables asynchronous communication between intelligent agents.

4. The engineering design method for multi-agent collaboration based on unstructured communication data as described in claim 3, characterized in that, According to the predetermined engineering design process, different intelligent agents lead each stage, including: Requirements Analysis Phase: The project management agent receives user requirements, analyzes the characteristics of the requirements, formulates an overall project plan, identifies key milestones, and assigns tasks to other intelligent agents; Engineering design phase: Based on the project plan, the engineer agent defines functional modules. Each module includes: module name and description, engineering requirements and technical specifications, input and output parameters, engineering constraints, estimated complexity and workload; Architecture design phase: The architect agent designs the system architecture based on functional modules, including: module division and organization structure, interface definition between modules, data structure design, data flow and control flow design; Test strategy phase: The quality assurance agent develops the test plan, including: unit test cases, integration test scenarios, system test plan and acceptance criteria.

5. The engineering design method for multi-agent collaboration based on unstructured communication data as described in claim 4, characterized in that, The multi-agent collaborative review process includes the following iterative steps during the engineering design phase: Each intelligent agent independently reviews the functional module division plan and generates its own opinions; The collaboration center identifies conflicts in review comments; For moderate conflicts, a voting mechanism based on professional expertise should be used to reach a consensus; The design is iterated and optimized based on the review results until the review is passed.

6. The engineering design method for multi-agent collaboration based on unstructured communication data as described in claim 4, characterized in that, At key junctures in the engineering design process, multi-agent collaborative reviews are conducted, with specific steps including: Parallel review: Each intelligent agent independently reviews the design results at the current stage from its own professional perspective based on its professional knowledge model, and generates review opinions; Conflict Identification: After collecting all review comments, the collaboration center uses a conflict identification algorithm to detect inconsistencies in the comments. Conflict classification: Identified conflicts are classified into minor conflicts, severe conflicts, and major conflicts for processing. Conflict resolution: Adopt appropriate resolution strategies based on the level of conflict.

7. The engineering design method for multi-agent collaboration based on unstructured communication data as described in claim 6, characterized in that, The conflict resolution strategies include: Minor conflict resolution: An automatic negotiation mechanism is used to adjust standardized descriptions to reach consensus; Moderate conflict resolution: A final opinion is reached using a voting mechanism based on professional expertise. Major conflict resolution: Mark conflicts that require human decision-making and generate decision support reports.

8. The engineering design method for multi-agent collaboration based on unstructured communication data as described in claim 7, characterized in that, The voting mechanism based on professional expertise includes the following steps: Step 1: Analyze the professional fields involved in the conflict; Step 2: Calculate the expertise weight of each intelligent agent in this field. The formula for calculating the weight W is: W = α × E + β × H + γ × R Where E represents the agent’s level of expertise in the field (0-1). H represents the accuracy of the agent's historical decisions (0-1). R represents the degree of relevance between the conflict and the agency responsibility (0-1). α, β, and γ are weighting coefficients that satisfy α + β + γ = 1; Preferably, α=0.5, β=0.3, γ=0.2; Step 3, the opinion score of each intelligent agent = number of approvals × sum of corresponding weights; Step 4: Select the opinion with the highest score as the final decision.

9. The engineering design method for multi-agent collaboration based on unstructured communication data as described in claim 1, characterized in that, The multi-agent collaboration verification includes: Engineer Agent Verification: Whether the function meets engineering requirements, whether the calculation is accurate, and whether the parameters are reasonable; Architect Agent Verification: Whether the architecture implementation conforms to the design, whether the interfaces are clear, and whether the code quality meets the standards; Quality Assurance Agent Verification: Whether the testing is sufficient, whether the coverage meets the standard, and whether the defects have been fixed; Project management agent verification: whether the progress is on schedule, whether the documentation is complete, and whether the deliverables are complete.

10. The engineering design method for multi-agent collaboration based on unstructured communication data as described in claim 1, characterized in that, The automatic knowledge extraction process includes: Successful design pattern extraction: Identify successful patterns in designs and extract them as structured knowledge; Problem solution extraction involves extracting reusable solutions from the problems encountered during the design process and their solutions. Collaboration experience extraction involves analyzing effective collaboration patterns and conflict resolution experiences in multi-agent collaboration processes and refining them into collaboration knowledge. Knowledge storage involves storing extracted knowledge in a structured form in a shared knowledge base.

11. An engineering design system based on multi-agent collaboration using unstructured communication data, characterized in that, The collaborative system comprises multiple intelligent agents with different professional roles, including an engineer agent, an architect agent, a quality assurance agent, a developer agent, and a project management agent, wherein: Engineer Agent: Responsible for defining functional modules, developing engineering specifications, and determining technical constraints; Architect Agent: Responsible for designing system architecture, defining module interfaces, and planning data flow; Quality Assurance Agent: Responsible for developing testing strategies, defining acceptance criteria, and performing quality assessments; Developer Agent: Responsible for implementing the actual development, writing code, and conducting unit tests; Project Management Agent: Responsible for coordinating the overall schedule, assigning tasks, and tracking milestones; Each of the intelligent agents is configured with a professional knowledge model, which includes: Domain rule base: Stores engineering rules, design specifications, and constraints for this professional domain; Experience Case Library: Stores historical design cases in this professional field and their effect evaluations; Inference engine: An algorithmic module that performs reasoning and decision-making based on domain rules and cases; Each intelligent agent exchanges information through a collaboration center, which includes: Shared knowledge base: Stores common knowledge and best practices across projects; Task queue: Manages design tasks and their dependencies at each stage; Message bus: Enables asynchronous communication between intelligent agents.

Images