Metal structure material service behavior evaluation method and system based on large model multi-modal analysis technology

Abstract

The invention discloses a metal structure material service behavior evaluation method and system based on a large-model multi-modal analysis technology, belongs to the technical field of metal structure materials, and solves the problem of obvious limitation in coping with a complex and changeable service environment. Comprising the steps of selecting a proper data source, determining a data crawling target, cleaning and labeling crawled data, performing training by using a deep learning model, and constructing a knowledge base; designing an experiment, collecting experiment data, combining the experiment data with other data through an intelligent agent, and carrying out multi-modal data fusion analysis; comparing a multi-modal data fusion analysis result with experimental data, calibrating a prediction model, and optimizing an analysis process; an intelligent decision support system is constructed, an experiment scheme is automatically generated, the experiment process is optimized, and suggestions are provided. The dynamic change of the material in the service process can be responded in real time, the prediction model is automatically adjusted, and the adaptability and real-time performance of the system in a complex environment are ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of metal structural materials, and specifically relates to a method and system for evaluating the service behavior of metal structural materials based on large model multimodal analysis technology. Background Art

[0002] Metal structural materials, especially light metal materials such as aluminum alloys, are key materials in modern industry and are widely used in aerospace, automobile, energy, construction and other fields. In actual service, metal structural materials often face complex and changeable environmental conditions, such as high temperature, high pressure, corrosion, fatigue, etc. These conditions will seriously affect the performance and service life of the materials. Therefore, accurate prediction and evaluation of the behavior of metal structural materials in service is of great significance to ensure equipment safety and improve material utilization.

[0003] At present, the evaluation of the service behavior of metal structural materials mainly relies on traditional methods such as finite element analysis (FEA), empirical formulas, and experimental test data, which evaluate the long-term service performance of materials by simulating the stress-strain state, fatigue behavior, corrosion degree, etc. of the materials in a specific service environment. The core of the solution is to perform numerical simulation on the mechanical behavior of materials, and verify and calibrate the model with experimental data, so as to obtain service life prediction and failure analysis of materials.

[0004] The applicant has found that the prior art has at least the following technical problems:

[0005] (1) Model dependence and lack of universality: Existing finite element models and empirical formulas are highly dependent on specific experimental data and service conditions. Therefore, for unknown or inadequately tested new materials, the prediction accuracy is low and it is difficult to generalize and apply them to different materials or complex and changing service environments.

[0006] (2) Insufficient data fusion capabilities: Existing solutions have limited capabilities in multimodal data fusion. They usually only use a single type of data (such as mechanical data) for analysis, while ignoring the influence of other physical fields (such as thermal fields, acoustic fields, etc.), resulting in insufficient comprehensive understanding of material behavior.

[0007] (3) Poor dynamic response capability: Finite element analysis is usually based on static or quasi-static assumptions, and it is difficult to respond in real time to dynamic changes in materials during service, such as sudden crack expansion and fluctuations in the corrosive environment.

[0008] (4) High computing resource requirements: Finite element analysis, especially in the case of complex geometric shapes and multi-physical field coupling, requires huge computing resources and takes a long time to simulate, making it difficult to meet real-time requirements.

[0009] (5) Lack of adaptive capabilities: Existing systems have difficulty self-learning and adapting to new service conditions or material changes. They lack intelligent decision-making support capabilities and cannot automatically update models to reflect changes in material properties in actual use.

[0010] In recent years, intelligent agent technology has gradually penetrated into various fields of industrial production and scientific research. With the deepening of scientific computing and the integration of large and small models, intelligent agents have shown superb ability in autonomous learning and adaptation to the environment, and can also make autonomous decisions based on the current situation and state changes. In the research of metal structural materials, there are a lot of physical and chemical changes. To deeply analyze the laws of change, it is necessary to comprehensively grasp the theoretical research, physical laws, chemical changes, data analysis and other fields. The emergence of intelligent agents provides a powerful technical means for realizing the fusion application and analysis in multi-modal, multi-scenario and multi-disciplinary situations. It can be used to comprehensively integrate data at different stages, different states and different situations, simulate the evolution process of materials under different service conditions, update the prediction model in real time, and form a highly adaptable metal service behavior analysis system through the combination with large models and multi-modal technologies. Summary of the invention

[0011] The purpose of the present invention is to provide a method and system for evaluating the service behavior of metal structural materials based on large model multimodal analysis technology, so as to solve the technical problem that traditional methods have obvious limitations when dealing with complex and changeable service environments.

[0012] To achieve the above object, the present invention provides the following technical solutions:

[0013] The method for evaluating the service behavior of metal structural materials based on large model multimodal analysis technology provided by the present invention comprises the following steps:

[0014] S1. Acquire at least one text data related to the service behavior of metal structural materials, use a large model to perform deep learning and semantic analysis on the collected data, and build a knowledge base covering the evaluation of the service behavior of metal structural materials;

[0015] S2, constructing an expert system for text data according to the knowledge base;

[0016] S3, conduct experiments and collect experimental data, build an agent group, and integrate the experimental data with the expert system. The agent group includes at least an understanding and analysis agent, a program planning agent, and an experimental process agent;

[0017] S4, copy the physical property data of experimental equipment and experimental samples, feed back the physical property data in real time to the digital twin system, and dynamically update the model;

[0018] S5. Update the agent group and build an intelligent decision support system.

[0019] Furthermore, the S1 comprises the following sub-steps:

[0020] S11. Data source selection: Determine the target of data crawling, including academic databases, journal websites, patent databases, technical reports, and conference proceedings. These sources will provide basic theories, experimental data, and case studies related to the service behavior of metal structural materials;

[0021] S12. Crawler development and deployment: Develop data crawler programs, set directional crawling rules, and automatically obtain relevant literature and data from selected data sources. The crawler will run regularly to ensure that the data in the knowledge base can be updated in a timely manner;

[0022] S13, Data cleaning and labeling: Clean the crawled data, remove duplicate, irrelevant or low-quality data, and then label the data, including material type, service conditions, experimental methods, and results, to facilitate subsequent classification and analysis;

[0023] S14. Large model training and knowledge base construction: Use deep learning models to train the cleaned data, extract key features and semantic information, and form a knowledge base covering the service behavior evaluation of metal structural materials.

[0024] Furthermore, the understanding and analysis agent includes a natural language understanding agent, a material domain knowledge agent and an experimental requirement analysis agent; the natural language understanding agent uses natural language processing technology to extract key information from documents, scientific research papers and experimental requests, and identifies and parses key information in experimental requests, including scientific terms, experimental conditions, and target requirements; the material domain knowledge agent provides in-depth analysis of material properties, environmental factors and stress conditions based on a large-scale material science database and knowledge base; the experimental requirement analysis agent combines the natural language understanding agent and the material domain knowledge agent to generate specific experimental goals and operational step suggestions, forming a multi-dimensional understanding of experimental problems.

[0025] Furthermore, the scheme planning agent includes an experimental process planning agent and an experimental scheme generating agent. The experimental process planning agent designs the overall process and steps of the experiment based on the analysis results of the understanding and analysis agent, using an automatic planning algorithm and a deep learning model. The experimental scheme generating agent uses a generative adversarial network and Bayesian optimization technology, combined with existing experimental data and multimodal analysis results, to generate the optimal experimental scheme.

[0026] Furthermore, the experimental process agent includes an experimental process recording agent, an experimental process analysis agent, an experimental data fusion agent, an experimental data verification agent and a multimodal data analysis agent; the experimental process recording agent records each step of the experiment in real time through the Internet of Things technology and a high-precision sensor network; the experimental process analysis agent uses machine learning and deep learning technologies to analyze the real-time experimental data and identify key changes and potential problems; the experimental data fusion agent uses multimodal data fusion technology to fuse data from different sources for unified analysis; the experimental data verification agent verifies the accuracy and consistency of the experimental data through statistical analysis and verification algorithms; the multimodal data analysis agent combines multiple modal data sources to conduct an all-round, multi-level in-depth analysis of the experimental process.

[0027] Furthermore, in S4, the implementation of the digital twin system includes the following steps:

[0028] S41. Equipment and sample twinning: The digital twin system first fully replicates the experimental equipment, including the physical properties and operation procedures of the equipment. Through high-precision 3D scanning technology, a digital twin model of the experimental sample is established to ensure that the physical properties are consistent with the actual sample;

[0029] S42, process data driven: the real-time data generated during the experiment will be dynamically fed back to the digital twin system to update the status of the experimental samples and the operating parameters of the equipment, ensuring that the process and results of the experiment are highly consistent with the actual situation;

[0030] S43. Result twin and analysis: After the experiment, the digital twin system will conduct a comprehensive analysis of the experimental results and generate a digital twin model that is consistent with the actual results.

[0031] Furthermore, the experiment tuning agent includes an experiment dynamic feedback agent, an experiment decision optimization agent, an experiment result analysis agent and an experiment optimization scheme agent; the experiment dynamic feedback agent uses a deep reinforcement learning algorithm to dynamically adjust the experiment parameters according to real-time data and feedback; the experiment decision optimization agent integrates multimodal analysis results and experiment feedback data, and provides experiment strategy recommendations through a multi-objective optimization algorithm; the experiment result analysis agent uses data mining technology and deep neural networks to conduct in-depth analysis of the experiment results, explore potential experimental trends and data patterns, and put forward improvement recommendations; the experiment optimization scheme agent combines the experiment analysis results and feedback, and continuously improves the experiment process and optimization strategy through meta-learning technology.

[0032] The system of the metal structure material service behavior evaluation method based on the large model multimodal analysis technology provided by the present invention comprises:

[0033] Data crawler and knowledge base construction module: used to automatically collect all papers, theoretical materials, and experimental data related to the service behavior of metal structure materials, and build a knowledge base for the evaluation of the service behavior of metal structures through a large model;

[0034] Industry Expert System Module: used to analyze material behavior under complex service conditions and provide intelligent suggestions for experimental design;

[0035] Experiment and data fusion module: It is used to conduct comprehensive experiments and data collection based on the service behaviors of aluminum alloys under atmospheric corrosion, corrosion fatigue, and creep conditions. The experimental data will be fused and analyzed through intelligent agent technology and expert knowledge base, experimental operation records, result data, image analysis data, as well as mechanism models, data models, and intelligent analysis models.

[0036] Digital twin module: used to realize the full process analysis and application of metal service behavior, including comprehensive replication of experimental equipment, physical property twin of experimental samples, data-driven experimental process, and comprehensive twin of experimental results and analysis results;

[0037] Intelligent decision support module: used to provide suggestions on material selection, design optimization and maintenance strategies.

[0038] Based on the above technical solution, the embodiments of the present invention can at least produce the following technical effects:

[0039] The present invention constructs an intelligent, real-time and dynamic prediction and evaluation system for the service behavior of metal structural materials by utilizing large-scale data analysis, intelligent agent technology and digital twin technology. Through multimodal data fusion and deep learning of large models, the mechanical properties, corrosion resistance, fatigue life, etc. of materials can be accurately predicted under various service conditions, which comprehensively improves the accuracy of the evaluation of the service behavior of metal structural materials, especially under complex service conditions, and can provide more accurate life prediction and failure analysis; through the combination of intelligent agent technology and digital twin system, the automation level in the experimental process is greatly improved, and intelligent decision support can be achieved from experimental design, data collection to result analysis, reducing human intervention and operation errors to achieve automation and intelligence of the experimental process; through digital twin technology, comprehensive twinning of experimental equipment, experimental samples, experimental process and results is achieved, and the virtual reproduction of experimental equipment and samples greatly reduces the time and resource consumption required for actual experiments, optimizes the experimental process, and reduces the experimental cost; it can respond to the dynamic changes of materials in the service process in real time, automatically adjust the prediction model, and ensure the adaptability and real-time performance of the system in complex environments.

[0040] The system solves the problems of poor model universality, insufficient data fusion capability, weak dynamic response capability, high computing resource requirements, and lack of adaptive capability in existing technologies, and significantly improves the accuracy, real-time, and intelligent level of material service behavior evaluation. The practical application of the system will greatly improve the reliability prediction of metal structural materials in complex service environments, and provide strong scientific support for engineering design, material selection, and maintenance strategy formulation. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in this field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0042] The objective of the present invention is achieved through the following technical solutions:

[0043] The present invention relates to a metal structural material service behavior evaluation system based on large model multimodal analysis technology. The system uses a large model to perform fusion analysis on multimodal data (including mechanics, thermal, chemical, image, etc.) to extract the behavior patterns of metal structural materials under various service conditions. The large model builds a professional knowledge base covering the service behavior evaluation of metal structural materials through deep learning and semantic analysis technology.

[0044] The system mainly includes the following key modules:

[0045] (1) Data crawler and knowledge base construction module: Through data crawler technology, the full amount of papers, theoretical materials, and experimental data related to the service behavior of metal structure materials are automatically collected, and a knowledge base for evaluating the service behavior of metal structures is constructed through a large model.

[0046] (2) Industry expert system module: Based on the large model and knowledge base, an industry expert system is built in the field of metal structural materials. The system can analyze the material behavior under complex service conditions and provide intelligent suggestions for experimental design.

[0047] (3) Experiment and data fusion module: Combined with the service behavior of aluminum alloys under various conditions such as atmospheric corrosion, corrosion fatigue, creep, etc., comprehensive experiments and data collection are carried out. The experimental data will be fused and analyzed through intelligent agent technology and expert knowledge base, experimental operation records, result data, image analysis data, as well as mechanism models, data models and intelligent analysis models.

[0048] (4) Digital twin system module: Through digital twin technology, the full process analysis and application of metal service behavior can be realized, including comprehensive replication of experimental equipment, physical property twinning of experimental samples, data-driven experimental process, and comprehensive twinning of experimental results and analysis results.

[0049] (5) Intelligent decision support module: Based on the comprehensive data and analysis results of the aforementioned modules, an intelligent decision support system is constructed to automatically generate experimental plans, optimize the experimental process, and provide suggestions on material selection, design optimization, and maintenance strategies.

[0050] The key technical means involved in the above modules are as follows:

[0051] (1) Data crawler technology: Data crawler technology is used to automatically obtain papers, theoretical materials and experimental data related to the service behavior of metal structural materials from various sources such as academic databases, journal websites, patent databases, technical reports, etc. The crawler technology will preliminarily classify and annotate these data and store them in the database as the basis for building a knowledge base.

[0052] (2) Construction of large model and knowledge base: Based on the crawled data, the large model is used to conduct deep learning and semantic analysis on the data to build a professional database covering the service behavior evaluation knowledge of metal structural materials. This knowledge base not only contains basic data such as the physical and chemical properties, mechanical properties, and service conditions of the materials, but also integrates theoretical models, experimental results, and expert experience to form a comprehensive knowledge system.

[0053] (3) Multimodal data fusion: Through agent-based technology, multimodal data (including mechanical data, chemical data, thermal data, image data, etc.) from different experiments are fused and analyzed. Agent-based technology can extract relevant features from multidimensional data, identify the behavior patterns of materials under different service conditions, and form a unified prediction model.

[0054] (4) Digital twin technology: The digital twin system uses high-precision simulation technology to replicate the physical properties of experimental equipment and experimental samples, realizing a comprehensive digital drive of the experimental process. The data generated during the experiment will be fed back to the twin system in real time, and the model will be dynamically updated to reflect the real-time status of the material and future evolution trends.

[0055] (5) Intelligent agents and intelligent decision support: Intelligent agents are not only responsible for data integration and dynamic adjustment of models, but can also autonomously learn data from the experimental process and continuously optimize the prediction model. Combined with expert knowledge base and digital twin technology, intelligent agents can automatically design experimental plans, optimize experimental processes, and provide intelligent support in material selection, design optimization, and maintenance decisions.

[0056] The above system is mainly implemented in the following way, which specifically includes the following steps:

[0057] S1. Use data crawler technology to automatically collect all papers, theoretical materials and experimental data related to the service behavior of metal structural materials, use large models to conduct deep learning and semantic analysis on the collected data, and build a knowledge base covering the evaluation of the service behavior of metal structural materials;

[0058] S11. Data source selection: First, determine the target of data crawling, including academic databases (such as Web of Science, Scopus, etc.), journal websites, patent databases, technical reports, conference proceedings, etc. These sources will provide basic theories, experimental data, case studies, etc. related to the service behavior of metal structural materials.

[0059] S12. Crawler development and deployment: Develop data crawler programs, set directional crawling rules, and automatically obtain relevant literature and data from selected data sources. The crawler will run regularly to ensure that the data in the knowledge base can be updated in a timely manner.

[0060] S13. Data cleaning and labeling: Clean the crawled data to remove duplicate, irrelevant or low-quality data. Then, label the data, including material type, service conditions, experimental methods, results, etc., to facilitate subsequent classification and analysis.

[0061] S14. Large model training and knowledge base construction: Use deep learning models to train the cleaned data, extract key features and semantic information, and form a knowledge base covering the service behavior evaluation of metal structural materials. The knowledge base will be integrated with the industry expert system to provide support for subsequent experimental design and data analysis.

[0062] S2. Based on the constructed knowledge base, an industry expert system is created in the field of metal structural materials service. The system is used to analyze the material behavior under complex service conditions and provide intelligent suggestions for experimental design.

[0063] S3. Through understanding and analysis agents, accurate interpretation and analysis of complex experimental requirements can be achieved. Through program planning agents, the optimal experimental process can be provided. Service behavior experiments of aluminum alloys under atmospheric corrosion, corrosion fatigue, and creep conditions can be designed and executed. Through the experimental process agent, a comprehensive and multi-level in-depth analysis of the experimental process can be conducted. The experimental data can be integrated with the expert knowledge base, experimental operation records, result data, image analysis data, as well as mechanism models, data models, and intelligent analysis models for analysis.

[0064] S4. Use digital twin technology to fully replicate the physical properties of experimental equipment and experimental samples, realize comprehensive digital drive of the experimental process, dynamically feed back real-time data during the experiment to the digital twin system, dynamically update the model, and reflect the real-time status of the material and future evolution trend;

[0065] The implementation of the digital twin system includes the following steps:

[0066] S41. Equipment and sample twinning: The digital twin system first fully replicates the experimental equipment, including the physical properties of the equipment, operating procedures, etc. In addition, through high-precision 3D scanning technology, a digital twin model of the experimental sample is established to ensure that the physical properties are consistent with the actual sample.

[0067] S42. Process data driven: The real-time data generated during the experiment will be dynamically fed back to the digital twin system to update the status of the experimental samples and the operating parameters of the equipment, ensuring that the process and results of the experiment are highly consistent with the actual situation.

[0068] S43. Result twinning and analysis: After the experiment, the digital twin system will conduct a comprehensive analysis of the experimental results, including the prediction and verification of mechanical properties, corrosion behavior, fatigue life, etc., and generate a digital twin model consistent with the actual results.

[0069] S5. Based on the comprehensive data and analysis results of the above steps, the intelligent agent is tuned through experiments to build an intelligent decision support system; the system automatically generates experimental plans, optimizes the experimental process, and provides suggestions on material selection, design optimization, and maintenance strategies;

[0070] S5 includes the following specific steps:

[0071] S51. Automatic generation of experimental plans: Based on the knowledge base and multimodal data fusion analysis results, the intelligent agent will automatically generate experimental plans, including experimental steps, parameter settings, expected results, etc., to reduce human intervention and errors.

[0072] S52. Experimental process optimization: The intelligent agent dynamically adjusts the experimental parameters and operating procedures based on the real-time feedback of experimental data, optimizes the experimental process, and ensures the accuracy of the data and the efficiency of the experiment.

[0073] S53. Result application and decision support: Based on the experimental results and the expert experience in the knowledge base, the intelligent agent provides intelligent decision support in material selection, design optimization, maintenance strategy, etc., providing a scientific basis for engineering applications.

[0074] Specifically, the intelligent agents include natural language understanding agents, material domain knowledge agents, and experimental requirements analysis agents involved in problem understanding and demand analysis, experimental process planning agents and experimental scheme generation agents involved in scheme planning, experimental process recording agents, experimental process analysis agents, experimental data fusion agents, experimental data verification agents, and multimodal data analysis agents involved in the experimental process, experimental dynamic feedback agents, experimental decision optimization agents, experimental results analysis agents, and experimental optimization scheme agents involved in experimental tuning, as well as data transmission agents, data classification agents, and twin application agents involved in twin applications; the intelligent agents in this scheme, on the one hand, are based on traditional AI machine learning or deep learning models for calculation and analysis, and on the other hand, they are combined with the multimodal analysis capabilities of large models, and have stronger capabilities in problem understanding, scheme planning, historical memory, tool use, and other aspects. These intelligent agents are described in detail below:

[0075] In the process of building the intelligent agent, the first problem that needs to be solved is how to accurately understand the experimental objectives and requirements. This part is jointly implemented by the natural language understanding intelligent agent, the material domain knowledge intelligent agent and the experimental requirement analysis intelligent agent. The natural language understanding intelligent agent uses advanced natural language processing (NLP) technology, including language models such as BERT and GPT, to extract key information from documents, scientific research papers and experimental requests. The intelligent agent can identify and parse scientific terms, experimental conditions, target requirements, etc. to ensure a comprehensive understanding of the problem. The material domain knowledge intelligent agent plays a further role. It provides in-depth analysis of material properties, environmental factors, stress conditions, etc. based on large-scale material science databases and knowledge bases, such as material property databases, phase diagram databases, corrosion behavior databases, etc. The experimental requirement analysis intelligent agent is responsible for combining natural language understanding and domain knowledge, making accurate technical descriptions of experimental requirements, generating specific experimental objectives and operation step suggestions, and forming a multi-dimensional understanding of experimental problems. This process realizes the accurate interpretation and analysis of complex experimental requirements through the linkage of intelligent agents, ensuring the scientificity and feasibility of experimental design.

[0076] After a comprehensive understanding and analysis of the experimental problem, the agent enters the solution planning stage, which is led by the experimental process planning agent and the experimental solution generation agent. The experimental process planning agent designs the overall process and steps of the experiment based on the problem analysis results. It uses automatic planning algorithms (such as task planning algorithms) and deep learning models to consider factors such as experimental conditions, resource constraints, and equipment configuration to generate an operational experimental process. Next, the experimental solution generation agent uses generative adversarial networks (GANs) and Bayesian optimization techniques to combine existing experimental data and multimodal analysis results to generate the optimal experimental solution. This agent can dynamically adjust experimental parameters and steps to maximize the effectiveness and efficiency of the experiment. In addition, through reinforcement learning algorithms (such as proximal policy optimization PPO), the experimental solution generation agent can perform multiple iterations and optimizations in a simulated environment, continuously improve the experimental design, and ensure that the experimental solution is scientifically reasonable and highly feasible. The entire solution planning process aims to provide an optimal path for actual experiments and improve the success rate of experiments and data reliability.

[0077] The experimental process is the core stage of the agent's work, which is jointly implemented by the experimental process recording agent, the experimental process analysis agent, the experimental data fusion agent, the experimental data verification agent and the multimodal data analysis agent. The experimental process recording agent is responsible for recording every step, operating conditions and real-time data of the experiment in real time, and uses the Internet of Things (IoT) technology and high-precision sensor network to ensure the comprehensiveness and accuracy of the data. The experimental process analysis agent uses machine learning and deep learning technologies (such as convolutional neural networks CNN and long short-term memory networks LSTM) to analyze the real-time data of the experiment and identify key changes and potential problems in the experiment. The experimental data fusion agent uses multimodal data fusion technology (such as multi-view learning and adaptive weight algorithm) to fuse and unify data from different sources (such as temperature, pressure, microscopic images, etc.) to improve the overall understanding of experimental data. The experimental data verification agent verifies the accuracy and consistency of experimental data through statistical analysis and verification algorithms (such as chi-square test and Monte Carlo simulation). Multimodal data analysis agents play an important role in this stage. By combining data sources of different modalities (such as text, images, videos, and sensor data), they can achieve a comprehensive, multi-level in-depth analysis of the experimental process and improve the ability to understand and control the experimental process.

[0078] The process of experiment tuning involves experimental dynamic feedback agents, experimental decision optimization agents, experimental result analysis agents, and experimental optimization scheme agents. The experimental dynamic feedback agent uses deep reinforcement learning algorithms (such as DQN and DDPG) to dynamically adjust experimental parameters such as temperature, stress, time, etc. according to real-time data and feedback during the experiment to optimize experimental conditions. The experimental decision optimization agent integrates multimodal analysis results and experimental feedback data, and provides optimized experimental strategy recommendations through multi-objective optimization algorithms such as genetic algorithms and particle swarm optimization. The experimental result analysis agent uses data mining technology and deep neural networks to conduct in-depth analysis of experimental results, explore potential experimental trends and data patterns, and put forward improvement suggestions. The experimental optimization scheme agent uses meta-learning technology, combined with the aforementioned experimental analysis results and feedback, to continuously improve the experimental process and optimization strategy, generate new experimental scheme recommendations, and ensure continuous optimization and efficient execution of the experimental process.

[0079] The twin application part is the key link for the intelligent agent to achieve a high degree of integration between virtual and reality. It is mainly composed of data transmission intelligent agent, data classification intelligent agent and twin application intelligent agent. The data transmission intelligent agent is responsible for the two-way real-time transmission of data between the physical experiment and the digital twin model to ensure the synchronization of the virtual environment and the physical experiment and the consistency of the data. It uses efficient data transmission protocols (such as MQTT, CoAP) and edge computing technology to reduce latency and improve transmission efficiency. The data classification intelligent agent uses machine learning classification algorithms (such as support vector machine SVM and decision tree) to automatically classify and label the transmitted data to provide accurate data support for the twin model. The twin application intelligent agent combines multiple data and models (such as finite element models, physical models and data-driven models) to create a high-precision digital twin, realizing all-round virtual simulation and predictive analysis of physical experiments. Through digital twin technology, the intelligent agent can conduct a large number of virtual experiments and simulation analysis, reduce the actual experimental cost, and improve the experimental efficiency and reliability of the results. The combination of twin applications enables the intelligent agent to establish a close connection between virtual and reality, realize the comprehensive optimization of the experiment and the digital reproduction of the process.

[0080] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements fall within the scope of the present invention to be protected. The scope of the present invention to be protected is defined by the attached claims and their equivalents.

Claims

1. A method for evaluating the service behavior of metal structural materials based on large model multimodal analysis technology, characterized in that: include: S1. Acquire at least one text data related to the service behavior of metal structural materials, use a large model to perform deep learning and semantic analysis on the collected data, and build a knowledge base covering the evaluation of the service behavior of metal structural materials; S2, constructing an expert system for text data according to the knowledge base; S3, conduct experiments and collect experimental data, build an agent group, and integrate the experimental data with the expert system. The agent group includes at least an understanding and analysis agent, a program planning agent, and an experimental process agent; S4, copy the physical property data of experimental equipment and experimental samples, feed back the physical property data in real time to the digital twin system, and dynamically update the model; S5. Update the agent group and build an intelligent decision support system.

2. The method for evaluating the service behavior of metal structural materials based on large model multimodal analysis technology according to claim 1 is characterized in that: The S1 includes: S11. Data source selection: Determine the target of data crawling, including academic databases, journal websites, patent databases, technical reports, and conference proceedings. These sources will provide basic theories, experimental data, and case studies related to the service behavior of metal structural materials; S12. Crawler development and deployment: Develop data crawler programs, set directional crawling rules, and automatically obtain relevant literature and data from selected data sources. The crawler will run regularly to ensure that the data in the knowledge base can be updated in a timely manner; S13, Data cleaning and labeling: Clean the crawled data, remove duplicate, irrelevant or low-quality data, and then label the data, including material type, service conditions, experimental methods, and results, to facilitate subsequent classification and analysis; S14. Large model training and knowledge base construction: Use deep learning models to train the cleaned data, extract key features and semantic information, and form a knowledge base covering the service behavior evaluation of metal structural materials.

3. The method for evaluating the service behavior of metal structural materials based on large model multimodal analysis technology according to claim 1 is characterized in that: The understanding and analysis agent includes a natural language understanding agent, a material domain knowledge agent, and an experimental requirement analysis agent; The natural language understanding agent uses natural language processing technology to extract key information from documents, scientific research papers and experimental requests, and identifies and parses key information in experimental requests, including scientific terms, experimental conditions, and target requirements; the material domain knowledge agent provides in-depth analysis of material properties, environmental factors and stress conditions based on a large-scale material science database and knowledge base; the experimental requirement analysis agent combines the natural language understanding agent and the material domain knowledge agent to generate specific experimental goals and operational step suggestions, forming a multi-dimensional understanding of experimental problems.

4. The method for evaluating the service behavior of metal structural materials based on large model multimodal analysis technology according to claim 1 is characterized in that: The scheme planning agent includes an experimental process planning agent and an experimental scheme generating agent. The experimental process planning agent designs the overall process and steps of the experiment based on the analysis results of the understanding and analysis agent, using an automatic planning algorithm and a deep learning model. The experimental scheme generating agent uses a generative adversarial network and Bayesian optimization technology, combined with existing experimental data and multimodal analysis results, to generate the optimal experimental scheme.

5. The method for evaluating the service behavior of metal structural materials based on large model multimodal analysis technology according to claim 1 is characterized in that: The experimental process agent includes an experimental process recording agent, an experimental process analysis agent, an experimental data fusion agent, an experimental data verification agent and a multimodal data analysis agent; the experimental process recording agent records each step of the experiment in real time through the Internet of Things technology and a high-precision sensor network; the experimental process analysis agent uses machine learning and deep learning technologies to analyze the real-time experimental data and identify key changes and potential problems; the experimental data fusion agent uses multimodal data fusion technology to fuse data from different sources for unified analysis; the experimental data verification agent verifies the accuracy and consistency of the experimental data through statistical analysis and verification algorithms; the multimodal data analysis agent combines multiple modal data sources to conduct an all-round, multi-level in-depth analysis of the experimental process.

6. The method for evaluating the service behavior of metal structural materials based on large model multimodal analysis technology according to claim 1 is characterized in that: In S4, the implementation of the digital twin system includes the following steps: S41. Equipment and sample twinning: The digital twin system first fully replicates the experimental equipment, including the physical properties and operation procedures of the equipment. Through high-precision 3D scanning technology, a digital twin model of the experimental sample is established to ensure that the physical properties are consistent with the actual sample; S42, process data driven: the real-time data generated during the experiment will be dynamically fed back to the digital twin system to update the status of the experimental samples and the operating parameters of the equipment, ensuring that the process and results of the experiment are highly consistent with the actual situation; S43. Result twin and analysis: After the experiment, the digital twin system will conduct a comprehensive analysis of the experimental results and generate a digital twin model that is consistent with the actual results.

7. The method for evaluating the service behavior of metal structural materials based on large model multimodal analysis technology according to claim 1 is characterized in that: The experimental tuning agent includes an experimental dynamic feedback agent, an experimental decision optimization agent, an experimental result analysis agent, and an experimental optimization solution agent; the experimental dynamic feedback agent uses a deep reinforcement learning algorithm to dynamically adjust experimental parameters according to real-time data and feedback; the experimental decision optimization agent integrates multimodal analysis results and experimental feedback data, and provides experimental strategy recommendations through a multi-objective optimization algorithm; The experimental result analysis agent uses data mining technology and deep neural network to conduct in-depth analysis of the experimental results, explore potential experimental trends and data patterns, and put forward improvement suggestions; The experimental optimization scheme agent combines the experimental analysis results and feedback, and continuously improves the experimental process and optimization strategy through meta-learning technology.

8. A system for implementing the method for evaluating the service behavior of metal structural materials based on large model multimodal analysis technology as described in any one of claims 1 to 7, characterized in that: include: Data crawler and knowledge base construction module: used to automatically collect all papers, theoretical materials, and experimental data related to the service behavior of metal structure materials, and build a knowledge base for the evaluation of the service behavior of metal structures through a large model; Industry Expert System Module: used to analyze material behavior under complex service conditions and provide intelligent suggestions for experimental design; Experiment and data fusion module: It is used to conduct comprehensive experiments and data collection based on the service behaviors of aluminum alloys under atmospheric corrosion, corrosion fatigue, and creep conditions. The experimental data will be fused and analyzed through intelligent agent technology and expert knowledge base, experimental operation records, result data, image analysis data, as well as mechanism models, data models, and intelligent analysis models. Digital twin module: used to realize the full process analysis and application of metal service behavior, including comprehensive replication of experimental equipment, physical property twin of experimental samples, data-driven experimental process, and comprehensive twin of experimental results and analysis results; Intelligent decision support module: used to provide suggestions on material selection, design optimization and maintenance strategies.