General design inspection architecture method for comprehensively scheduling small models based on AI large models

Abstract

The invention relates to the technical field of design examination architecture, in particular to a general design examination architecture method for comprehensively scheduling small models based on AI large models. Comprising the following steps: constructing an intelligent scheduling system which takes a general AI large model as a core and comprises a visual analysis model, a text understanding and segmentation conversion model and a rule generation model, enabling a user to submit a design review task through a dialogue interface or a file uploading mode, receiving a task request through a large language model, and carrying out preliminary analysis; through multi-module cooperation and intelligent scheduling, the automation degree of design drawing review is greatly improved, the overall review efficiency is improved, visual analysis, text processing and rule generation technologies are fused, compared with a rule-based drawing review system, wider review content can be covered, and the design drawing review efficiency is improved. The versatility of the architecture enables the architecture to rapidly adapt to new rules and standards, and enables the iteration updating of the drawing checking system to be faster.

Description

Technical Field

[0001] The present invention relates to the technical field of design review architecture, and specifically to a general design review architecture method based on AI large model comprehensive scheduling of small models. Background Art

[0002] Drawing design plays an important role in industrial manufacturing, especially in the field of high-end equipment manufacturing such as locomotives. The review of design drawings is a complex and delicate process. When faced with large quantities of complex drawings, the efficiency and accuracy of traditional manual review methods cannot meet the high requirements of modern engineering. Under the general trend of digital transformation, intelligent drawing review systems have emerged, which significantly improve the efficiency and accuracy of review through automated procedures. However, most of the existing drawing review systems are based on rule-based programming, which makes it difficult to fully respond to the following challenges: the diversity and complexity of standards, the difficulty of quantifying specific scenarios, and the unstructured nature of review experience.

[0003] However, in general, rule-based methods cannot automatically review projects that are beyond the scope of the rule definition. This requires manual organization of standards and rules and manual code writing, which is a cumbersome and time-consuming process. Even programs implemented based on rule programming have a small coverage range and are difficult to meet the needs of diverse scenarios. At the same time, the intelligent drawing review system in the construction industry usually relies on industry-specific data for training. Its tasks are relatively simple, the drawings are of low complexity, and the accuracy requirements are not high. These models cannot be directly reused in the field of industrial design, limiting their applicability across industries.

[0004] Based on this, the present invention provides a general design review architecture method based on the comprehensive scheduling of small models by a large AI model to solve the technical problems raised above. Summary of the invention

[0005] The purpose of the present invention is to provide a general design review architecture method based on AI large model comprehensive scheduling small model to solve the problems raised by the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A general design review architecture method based on AI large model comprehensive scheduling small model is provided, including the following steps:

[0008] S1. Build an intelligent scheduling system with a general AI big model as the core. Users submit design review tasks through a dialogue interface or file upload. The big language model receives task requests and performs preliminary analysis to understand user intentions and review requirements. Through natural language processing technology, it enhances the semantic understanding of user input, supports fuzzy queries and intent recognition, and improves the flexibility and convenience of task submission.

[0009] S2. The large model breaks down complex tasks into multiple subtasks based on the analysis results, and distributes them to the corresponding single-task visual analysis model, text processing model, and rule generation model according to the requirements and attributes of the subtasks. Through the task priority and urgency evaluation mechanism, the task distribution strategy is optimized to ensure that high-priority tasks are processed first;

[0010] S3. Through the visual analysis model, the design drawings are deeply analyzed to identify and verify the symbols, dimensioning, and table graphic elements in the drawings. Combined with deep learning technology, the recognition ability of complex graphics and special symbols is improved to support the review of cross-industry drawings;

[0011] S4. The rule generation model analyzes industry specifications and enterprise standards, generates a standard rule base in XML format and supports dynamic updates. The user-defined rule function allows users to input custom rules in natural language, and the model automatically analyzes and generates corresponding rule entries, thereby improving the flexibility and adaptability of the rules.

[0012] S5. Receive and execute large model distribution tasks through the rule-based review system, perform mathematical and logical checks to ensure that the design drawings meet the established standards, and continuously optimize the rule-based review system through machine learning technology to improve the accuracy and efficiency of the review;

[0013] S6. The single-task model integrates the visual analysis model, text processing model, and rule generation model through a general AI big model to integrate data and results from different modules, and improves the efficiency and scalability of multi-module collaboration through distributed computing and cloud computing technologies;

[0014] S7. Construct a drawing review report generation module, generate a report based on the review results through the drawing review report generation module, provide structured and hierarchical report output, support multiple formats of export, and use data visualization technology to intuitively display the review results in the form of charts and animations to improve the readability and comprehensibility of the report;

[0015] S8. Integrate the review report with the design software, and directly mark the problem location and modification suggestions in the feedback closed loop. Through the intelligent recommendation system built in step S1, possible solutions or improvement suggestions are recommended to users based on the review results to further improve the design quality;

[0016] S9. Optimize the system based on key data and user feedback during task execution, continuously iterate and upgrade, and speed up system updates through continuous integration and continuous deployment (CI / CD) technology to ensure that the system always remains in the best state. At the same time, establish a user feedback mechanism to encourage users to participate in the system improvement and upgrade process and jointly promote technological progress.

[0017] Preferably, the implementation steps of step S1 are:

[0018] S1.1. Users submit design review tasks through a graphical interface or API interface. Supported file formats include DWG, DXF, and PDF. The system records the task submission time, user information, and basic drawing attributes.

[0019] S1.2. The large language model monitors the task submission interface, automatically receives the task request submitted by the user, and performs a preliminary format check on the task request to ensure file integrity and readability;

[0020] S1.3. Use natural language processing technology to parse the task descriptions and review requirements submitted by users and identify user intentions, including the scope of review and key areas of focus;

[0021] S1.4. Support fuzzy queries input by users, recognize user intent even if the description is incomplete, and generate a preliminary review task summary based on the recognition results;

[0022] S1.5. Show the user a summary of the identified review requirements and request the user to confirm. The user can adjust or supplement the review requirements, and the system will make corrections based on the user's feedback;

[0023] S1.6. Based on the review requirements confirmed by the user, the system initializes the review task and prepares to enter the next step.

[0024] Preferably, the implementation steps of step S2 are:

[0025] S2.1. Analyze the complexity of the review task for large models, including the number of drawings, layer complexity, and number of review points. Based on the complexity assessment results, decide whether the task needs to be disassembled;

[0026] S2.2. Break down complex tasks into multiple subtasks, each corresponding to a specific review point or drawing area, and assign a unique identifier and priority to each subtask;

[0027] S2.3. Define the required model type, parameter settings, and expected outputs according to the subtask content, ensuring that the subtask description is clear and easy for subsequent model understanding and execution;

[0028] S2.4. Introduce task priority and urgency evaluation mechanism, formulate task distribution strategy, and intelligently select the most appropriate model for task distribution based on model load and performance;

[0029] S2.5. Distribute subtasks to corresponding single-task models, including visual analysis models and text processing models, monitor the task distribution process, and ensure that the tasks are successfully delivered and started;

[0030] S2.6. Collect feedback from the model on task distribution, including whether to accept the task and the estimated completion time, adjust the task distribution strategy based on the feedback, and optimize resource allocation.

[0031] Preferably, the implementation steps of step S3 are:

[0032] S3.1. The visual analysis model loads the drawing file submitted by the user and performs preprocessing operations, including denoising, scaling, and color correction to improve recognition accuracy;

[0033] S3.2. Use deep learning technology to identify graphic elements in drawings, including lines, symbols, and dimension annotations, conduct preliminary verification of the recognition results, and eliminate erroneous or redundant elements;

[0034] S3.3. Verify the identified symbols and annotations according to industry specifications and enterprise standards to check the correctness of the symbols, accuracy and consistency of the annotations;

[0035] S3.4. Classify the graphic elements that have been identified and verified, including structural elements and electrical elements, to provide classification information for subsequent review and improve review efficiency;

[0036] S3.5. For complex graphics and special symbols, use the advanced feature extraction capabilities of deep learning models to identify them, and combine contextual information to improve the recognition accuracy of complex graphics;

[0037] S3.6. Record the verification results of the graphic elements into the database, including pass, fail and problem details, and generate a verification report for use in subsequent steps.

[0038] Preferably, the implementation steps of step S4 are:

[0039] S4.1. The rule generation model parses industry specifications and enterprise standard documents, extracts key information, and structures the documents to facilitate subsequent rule generation;

[0040] S4.2. Based on the parsing results, a standard rule base in XML format is automatically generated, which includes rule entries, descriptions, and applicable condition information;

[0041] S4.3. Verify the generated rules to ensure their accuracy and applicability, and optimize and adjust the rules based on user feedback and actual needs;

[0042] S4.4. Support dynamic update of the rule base. When the specifications or standards change, update the rule base in time to ensure that the review system always uses the latest rules for review;

[0043] S4.5. Introduce the user-defined rule function, allowing users to input custom rules through natural language. The system automatically parses the user input, generates corresponding rule entries, and adds them to the rule library;

[0044] S4.6. Perform version management on the rule base, record the content and time of each update, provide version rollback function, and ensure the stability and traceability of the rule base.

[0045] Preferably, the implementation steps of step S5 are:

[0046] S5.1. The rule-based review system receives tasks distributed by the large model, matches corresponding rules according to task requirements, and ensures that each review point has corresponding rules for verification;

[0047] S5.2. Perform mathematical and logical checks, including dimensional calculations, material selection, and process requirements, to ensure that the design drawings meet the mathematical and logical requirements of the established standards;

[0048] S5.3. Record the review results in the database, including pass, fail and problem details, and generate a review report for subsequent steps;

[0049] S5.4. Locate the failed review points, mark the problem locations and modification suggestions, and provide intuitive feedback to facilitate user understanding and modification;

[0050] S5.5. Introduce machine learning technology to continuously optimize the rule-based review system and improve the accuracy and efficiency of review based on historical review data and user feedback;

[0051] S5.6. Feedback the review results to the big model for subsequent processing and integration, and provide preview and export functions for the review report to facilitate user viewing and use.

[0052] Preferably, the implementation steps of step S6 are:

[0053] S6.1. Collect data from each single-task model, including visual review results, text processing results, and rule checking results, to ensure data integrity and consistency;

[0054] S6.2. Integrate the collected data to form a unified review result data set, clean and format the data to facilitate subsequent analysis and use;

[0055] S6.3. Verify the integrated data to ensure its accuracy and reliability, check the relevance and consistency between the data, and eliminate data conflicts;

[0056] S6.4. Convert and store data according to the needs of subsequent steps, supporting multiple data formats and storage methods, including databases and file systems;

[0057] S6.5. Introduce distributed computing and cloud computing technologies to improve the efficiency and scalability of multi-module collaboration and utilize cloud computing resources to achieve rapid data processing and storage;

[0058] S6.6. Provide a data sharing interface to allow other systems or modules to access the review result data, ensure the security and stability of the data interface, and prevent data leakage or damage.

[0059] Preferably, the implementation steps of step S7 are:

[0060] S7.1. Select appropriate report templates based on user needs and review results, and support customized report templates to meet the needs of different users;

[0061] S7.2. Fill in the review results, problem descriptions, and modification suggestions into the report template to ensure the accuracy and completeness of the report content and facilitate user understanding;

[0062] S7.3. Adjust the report format according to user needs, including font, font size, and layout, and support export in multiple formats, including PDF, Word, and Excel, to facilitate user viewing and sharing;

[0063] S7.4. Introduce data visualization technology to intuitively display the review results in the form of charts and animations to improve the readability and comprehension of the report and facilitate users to quickly locate problems;

[0064] S7.5. Review the generated reports to ensure that they meet user requirements and industry standards, publish the approved reports to users or related systems, and provide download or viewing links;

[0065] S7.6. Perform version management on the report, record the time and content of each generation and modification, provide version rollback function, and ensure the stability and traceability of the report.

[0066] Preferably, the implementation steps of step S8 are:

[0067] S8.1. Mark the problems found in the review results on the design drawings, provide modification suggestions, and feedback the marked results and modification suggestions to the user or design software to form a closed-loop management;

[0068] S8.2. Integration with design software supports direct viewing and modification of marked issues in the design software, improving design rectification efficiency and reducing user operation steps.

[0069] Preferably, the implementation steps of step S9 are:

[0070] S9.1. Establish a regular review and evaluation mechanism to regularly evaluate the effectiveness of question annotation, feedback, and intelligent recommendation systems. Based on the evaluation results, continuously optimize and improve the annotation process, feedback mechanism, and intelligent recommendation system.

[0071] S9.2. Set clear evaluation indicators, including the improvement rate of problem rectification efficiency, the reduction rate of design errors, and user satisfaction. Through data analysis, quantitatively evaluate the implementation effect of various improvement measures.

[0072] Compared with the prior art, the present invention has the following beneficial effects:

[0073] Through multi-module collaboration and intelligent scheduling, the present invention has greatly improved the automation level of design drawing review, significantly reduced manual intervention, and improved overall review efficiency. It integrates visual analysis, text processing, and rule generation technologies, and can cover a wider range of review content than rule-based review systems. The universality of the architecture enables it to quickly adapt to new rules and standards, making the iteration and update of the review system faster. The system supports flexible configuration and customized rules, and can meet the review needs of different industries and scenarios, including industrial manufacturing, construction engineering, scientific research and design. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 This is a general design review architecture diagram of the present invention based on the AI ​​large model comprehensive scheduling small model;

[0075] Figure 2 The present invention is a flow chart of a general design review architecture method for comprehensively scheduling small models based on an AI large model. DETAILED DESCRIPTION

[0076] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of 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.

[0077] Example 1

[0078] See also Figure 1 The present invention proposes a general design review architecture method based on AI large model comprehensive scheduling small model, comprising the following steps:

[0079] S1. Build an intelligent scheduling system with a general AI big model as the core, including a visual analysis model, a text understanding and segmentation conversion model, and a rule generation model, so that users can submit design review tasks through a dialogue interface or file upload, and receive task requests through a big language model for preliminary analysis;

[0080] S2. The large model breaks down the complex task into multiple subtasks based on the analysis results, and distributes them to the corresponding single-task visual analysis model, text processing model, and rule generation model according to the requirements and attributes of the subtasks;

[0081] S3. Perform in-depth analysis of design drawings through visual analysis models to identify and verify symbols, dimensioning, and table graphic elements in the drawings;

[0082] S4. Analyze industry specifications and enterprise standards through the rule generation model, generate a standard rule base in XML format and support dynamic updates;

[0083] S5. Receive and execute tasks distributed by large models through a rule-based review system, and perform checks at the mathematical and logical level;

[0084] S6. The single-task model integrates the visual analysis model, text processing model, and rule generation model through a general AI big model to integrate data and results from different modules;

[0085] S7. Construct a drawing review report generation module, generate a report based on the review results through the drawing review report generation module, and provide structured and hierarchical report output;

[0086] S8. Integrate the review report with the design software, and directly mark the problem location and modification suggestions in the feedback closed loop;

[0087] S9. Optimize the system based on key data and user feedback during task execution.

[0088] In this embodiment, it should be noted that the implementation steps of step S1 are:

[0089] S1.1. Users submit design review tasks through a graphical interface or API interface. Supported file formats include DWG, DXF, and PDF. Task submission time, user information, and basic drawing attributes are recorded.

[0090] S1.2. The large language model monitors the task submission interface, automatically receives the task request submitted by the user, and performs preliminary format verification on the task request;

[0091] S1.3. Use natural language processing technology to parse the task descriptions and review requirements submitted by users and identify user intentions;

[0092] S1.4. Support fuzzy queries input by users, recognize user intent even if the description is incomplete, and generate a preliminary review task summary based on the recognition results;

[0093] S1.5. Display the identified review requirements summary to the user and request the user to confirm;

[0094] S1.6. Initialize the review task according to the review requirements confirmed by the user and prepare to enter the next step of the process.

[0095] In this embodiment, it should be noted that the implementation steps of step S2 are:

[0096] S2.1. The big model analyzes the complexity of the review task and decides whether the task needs to be disassembled based on the complexity assessment results;

[0097] S2.2. Break down complex tasks into multiple subtasks, each corresponding to a specific review point or drawing area, and assign a unique identifier and priority to each subtask;

[0098] S2.3. Define the required model type, parameter settings and expected output according to the subtask content;

[0099] S2.4. Introduce task priority and urgency evaluation mechanism, formulate task distribution strategy, and select the most appropriate model for task distribution based on model load and performance;

[0100] S2.5. Distribute the subtasks to the corresponding single-task models and monitor the task distribution process;

[0101] S2.6. Collect feedback from the model on task distribution, adjust the task distribution strategy based on the feedback, and optimize resource allocation.

[0102] In this embodiment, it should be noted that the implementation steps of step S3 are:

[0103] S3.1. The visual analysis model loads the drawing file submitted by the user and performs preprocessing operations;

[0104] S3.2. Use deep learning technology to identify graphic elements in drawings, conduct preliminary verification of the recognition results, and eliminate erroneous or redundant elements;

[0105] S3.3. Verify the identified symbols and annotations according to industry specifications and enterprise standards to check the correctness of the symbols, accuracy and consistency of the annotations;

[0106] S3.4. Classify the identified and verified graphic elements, including structural elements and electrical elements, to provide classification information for subsequent review;

[0107] S3.5. For complex graphics and special symbols, use the advanced feature extraction capabilities of deep learning models to identify them and combine them with contextual information;

[0108] S3.6. Record the verification results of the graphic elements into the database, including pass, fail and problem details, and generate a verification report.

[0109] In this embodiment, it should be noted that the implementation steps of step S4 are:

[0110] S4.1. The rule generation model parses industry specifications and enterprise standard documents, extracts key information, and performs structured processing on the documents;

[0111] S4.2. Based on the parsing results, a standard rule base in XML format is automatically generated, which includes rule entries, descriptions, and applicable condition information;

[0112] S4.3. Verify the generated rules to ensure their accuracy and applicability, and optimize and adjust the rules based on user feedback and actual needs;

[0113] S4.4. Support dynamic update of rule base. When the specification or standard changes, update the rule base in time.

[0114] S4.5. Introduce the user-defined rule function, allowing users to input custom rules through natural language. The system automatically parses the user input, generates corresponding rule entries, and adds them to the rule library;

[0115] S4.6. Perform version management on the rule base, record the content and time of each update, and provide version rollback function.

[0116] In this embodiment, it should be noted that the implementation steps of step S5 are:

[0117] S5.1. The rule-based review system receives tasks distributed by the large model and matches corresponding rules according to task requirements;

[0118] S5.2. Perform mathematical and logical checks, including dimensional calculations, material selection, and process requirements;

[0119] S5.3. Record the review results in the database, including pass, fail and problem details, and generate a review report;

[0120] S5.4. Locate the failed review points, mark the problem locations and modification suggestions, and provide intuitive feedback;

[0121] S5.5. Introduce machine learning technology to continuously optimize the rule-based review system based on historical review data and user feedback;

[0122] S5.6. Feedback the review results to the big model and provide preview and export functions for the review report.

[0123] In this embodiment, it should be noted that the implementation steps of step S6 are:

[0124] S6.1. Collect data from each single-task model, including visual review results, text processing results, and rule checking results;

[0125] S6.2. Integrate the collected data to form a unified review result data set, and clean and format the data;

[0126] S6.3. Verify the integrated data to check the relevance and consistency between the data and eliminate data conflicts;

[0127] S6.4. Convert and store data according to the requirements of subsequent steps;

[0128] S6.5. Introduce distributed computing and cloud computing technologies to improve the efficiency and scalability of multi-module collaboration and use cloud computing resources to achieve rapid data processing and storage;

[0129] S6.6. Provide a data sharing interface to allow other systems or modules to access the review result data and ensure the security and stability of the data interface.

[0130] In this embodiment, it should be noted that the implementation steps of step S7 are:

[0131] S7.1. Select appropriate report templates based on user needs and review results, and support customized report templates;

[0132] S7.2. Fill in the report template with the review results, problem description, and modification suggestions;

[0133] S7.3. Adjust the report format according to user needs, including font, font size, and layout, and support export in multiple formats, including PDF, Word, and Excel;

[0134] S7.4. Introduce data visualization technology to intuitively display the review results in the form of charts and animations to improve the readability and comprehension of the report;

[0135] S7.5. Review the generated reports, publish the approved reports to users or related systems, and provide download or viewing links;

[0136] S7.6. Perform version management on the report, record the time and content of each generation and modification, and provide version rollback function.

[0137] In this embodiment, it should be noted that the implementation steps of step S8 are:

[0138] S8.1. Mark the problems found in the review results on the design drawings, provide modification suggestions, and feedback the marked results and modification suggestions to the user or design software to form a closed-loop management;

[0139] S8.2.Integrate with design software to support direct viewing and modification of annotated issues in the design software.

[0140] In this embodiment, it should be noted that the implementation steps of step S9 are:

[0141] S9.1. Establish a regular review and evaluation mechanism to regularly evaluate the effectiveness of question annotation, feedback, and intelligent recommendation systems. Based on the evaluation results, continuously optimize and improve the annotation process, feedback mechanism, and intelligent recommendation system.

[0142] S9.2. Set clear evaluation indicators, including the improvement rate of problem rectification efficiency, the reduction rate of design errors, and user satisfaction. Through data analysis, quantitatively evaluate the implementation effect of various improvement measures.

[0143] Example 2

[0144] See also Figure 1 and Figure 2 In the practical application of various drawing review in rail transit industry manufacturing, the above-mentioned general design review architecture method based on AI large model comprehensive scheduling small model specifically includes the following processes:

[0145] 1. Architecture Design

[0146] See also Figure 1 , build an intelligent scheduling system with a general AI big model as the core, and support the full process automated review of design drawings;

[0147] The architecture can dynamically call the review system, visual analysis model, text understanding and segmentation conversion model, rule generation model and review report generation module. The modules can collaborate efficiently to form a flexible and scalable intelligent review solution.

[0148] This architecture is applicable to various drawing review requirements of rail transit industry manufacturing;

[0149] 2. Visual Analysis Model

[0150] The visual analysis model is based on deep learning and computer vision technology and has powerful multimodal recognition capabilities;

[0151] It can accurately identify and verify various graphic elements in design drawings (such as symbols, dimensioning, tables, flow charts, and view marks, etc.), and support the adaptation of industry standards;

[0152] 3. Rule Generation Model

[0153] The rule generation model is based on a large language model, which can automatically parse industry specifications, enterprise standards and design experience, extract executable rule entries from them, and generate a standard rule base in XML format;

[0154] The rules are associated with relevant knowledge to form a dynamic update and traceability link of the rules, ensuring the comprehensiveness and real-time nature of the review standards;

[0155] Users can also input custom rules through natural language, and the model will automatically parse and generate corresponding rule entries, greatly improving the efficiency and flexibility of rule generation;

[0156] 4. Drawing review report generation module

[0157] The drawing review report generation module provides structured and hierarchical report output, including an overall review overview, a problem classification list, a detailed problem description, and targeted modification suggestions.

[0158] The report can be exported in multiple formats such as PDF, Word, and Excel to suit different application scenarios. It also provides intuitive visual display to help users quickly understand the review results.

[0159] The module is also tightly integrated with the design software, directly marking problem locations and modification suggestions through a closed feedback loop to improve the efficiency of design rectification;

[0160] 5. Multi-module collaboration and dynamic scheduling

[0161] Multi-module collaboration and dynamic scheduling aims to achieve efficient management and resource allocation of each sub-module through the intelligent scheduling capabilities of the general AI large model, ensuring that the system can flexibly respond and operate efficiently in complex and diverse review scenarios;

[0162] Intelligent task analysis and distribution: The general AI large model has powerful task understanding and analysis capabilities. It can decompose complex tasks submitted by users into multiple subtasks and distribute them to corresponding small models (such as visual analysis models, text processing models, and rule generation models) for processing based on task attributes and requirements.

[0163] The big model is not only responsible for the decomposition of tasks, but also sorts out the dependencies between subtasks to ensure the orderly connection of the overall process;

[0164] Dynamic task supervision and feedback: During the task execution process, the general AI model monitors the operating status of each module in real time and dynamically adjusts resource allocation and collaboration between modules;

[0165] For example, when a module encounters a bottleneck or delay in processing a task, the large model can reallocate resources or dynamically modify the process to ensure the timely completion of the overall task;

[0166] At the same time, the large model can record key data during task execution and provide support for subsequent system optimization;

[0167] Flexible collaboration between modules: Small models in the system (such as visual analysis models, text understanding models, etc.) achieve efficient collaboration through a dynamic scheduling mechanism;

[0168] The large model flexibly combines and calls each sub-module according to the requirements of different review tasks to cope with design review scenarios with different requirements, different standards, and even cross-languages;

[0169] For example, when processing design drawings, the visual analysis model is responsible for verifying graphic elements, the rule generation model is responsible for updating review rules, and the big model coordinates the task flow between the two to ensure data consistency and processing efficiency;

[0170] Resource optimization and expansion capabilities

[0171] The dynamic scheduling mechanism can intelligently adjust the resource allocation method according to the task load and module capabilities to avoid resource waste or module overload;

[0172] In the scenario of multi-task concurrent processing, the system can efficiently utilize available resources to ensure that each task can be completed under the best conditions;

[0173] In addition, the mechanism is highly scalable and can quickly adapt to new review requirements or technical standards by introducing new small models or upgrading existing modules;

[0174] Ability to adapt to complex and diverse scenarios

[0175] The core of dynamic scheduling lies in flexibility and adaptability. Whether it is for the review of traditional design drawings or complex scenarios involving complex engineering standards, multi-language document processing, knowledge review, etc., this function can achieve efficient and reliable task execution through intelligent combination of modules and optimized scheduling resources;

[0176] VI. Please refer to Figure 2 , the specific task process is:

[0177] S1. Build an intelligent scheduling system with a general AI big model as the core, including a visual analysis model, a text understanding and segmentation conversion model, and a rule generation model, so that users can submit design review tasks through a dialogue interface or file upload, and receive task requests through a big language model for preliminary analysis;

[0178] S2. The large model breaks down the complex task into multiple subtasks based on the analysis results, and distributes them to the corresponding single-task visual analysis model, text processing model, and rule generation model according to the requirements and attributes of the subtasks;

[0179] S3. Perform in-depth analysis of design drawings through visual analysis models to identify and verify symbols, dimensioning, and table graphic elements in the drawings;

[0180] S4. Analyze industry specifications and enterprise standards through the rule generation model, generate a standard rule base in XML format and support dynamic updates;

[0181] S5. Receive and execute tasks distributed by large models through a rule-based review system, and perform checks at the mathematical and logical level;

[0182] S6. The single-task model integrates the visual analysis model, text processing model, and rule generation model through a general AI big model to integrate data and results from different modules;

[0183] S7. Construct a drawing review report generation module, generate a report based on the review results through the drawing review report generation module, and provide structured and hierarchical report output;

[0184] S8. Integrate the review report with the design software, and directly mark the problem location and modification suggestions in the feedback closed loop;

[0185] S9. Optimize the system based on key data and user feedback during task execution

[0186] If a user uploads a new standard file, the large language model will call the rule generation model to automatically generate an XML format rule file that is suitable for the review system;

[0187] The drawing review system will then update the corresponding functions to ensure that it can adapt to the new standard requirements in a timely manner and realize the dynamic upgrade of system functions;

[0188] In summary, the technical effects of the present invention are:

[0189] Further improvement in efficiency: Through multi-module collaboration and intelligent scheduling, the automation level of design drawing review has been greatly improved, which significantly reduces manual intervention and improves overall review efficiency;

[0190] Expand the scope of inspection: By integrating visual analysis, text processing and rule generation technologies, it can cover a wider range of inspection contents than rule-based review systems;

[0191] Enhanced versatility: The versatility of the architecture enables it to quickly adapt to new rules and standards, making the iteration and update of the review system faster;

[0192] Applicable to multiple scenarios: The system supports flexible configuration and customized rules, and can meet the review needs of different industries and scenarios, including industrial manufacturing, construction engineering, scientific research and design.

[0193] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0194] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A general design review architecture method based on AI big model comprehensive scheduling small model, characterized by: The following steps are involved: S1. Build an intelligent scheduling system with a general AI big model as the core, including a visual analysis model, a text understanding and segmentation conversion model, and a rule generation model, so that users can submit design review tasks through a dialogue interface or file upload, and receive task requests through a big language model for preliminary analysis; S2. The large model breaks down the complex task into multiple subtasks based on the analysis results, and distributes them to the corresponding single-task visual analysis model, text processing model, and rule generation model according to the requirements and attributes of the subtasks; S3. Perform in-depth analysis of design drawings through visual analysis models to identify and verify symbols, dimensioning, and table graphic elements in the drawings; S4. Analyze industry specifications and enterprise standards through the rule generation model, generate a standard rule base in XML format and support dynamic updates; S5. Receive and execute tasks distributed by large models through a rule-based review system, and perform checks at the mathematical and logical level; S6. The single-task model integrates the visual analysis model, text processing model, and rule generation model through a general AI big model to integrate data and results from different modules; S7. Construct a drawing review report generation module, generate a report based on the review results through the drawing review report generation module, and provide structured and hierarchical report output; S8. Integrate the review report with the design software, and directly mark the problem location and modification suggestions in the feedback closed loop; S9. Optimize the system based on key data and user feedback during task execution.

2. The general design review architecture method based on AI large model comprehensive scheduling small model according to claim 1 is characterized in that: The implementation steps of step S1 are: S1.

1. Users submit design review tasks through a graphical interface or API interface. Supported file formats include DWG, DXF, and PDF. Task submission time, user information, and basic drawing attributes are recorded. S1.

2. The large language model monitors the task submission interface, automatically receives the task request submitted by the user, and performs preliminary format verification on the task request; S1.

3. Use natural language processing technology to parse the task descriptions and review requirements submitted by users and identify user intentions; S1.

4. Support fuzzy queries input by users, recognize user intent even if the description is incomplete, and generate a preliminary review task summary based on the recognition results; S1.

5. Display the identified review requirements summary to the user and request the user to confirm; S1.

6. Initialize the review task according to the review requirements confirmed by the user and prepare to enter the next step of the process.

3. The general design review architecture method based on AI large model comprehensive scheduling small model according to claim 2 is characterized in that: The implementation steps of step S2 are: S2.

1. The big model analyzes the complexity of the review task and decides whether the task needs to be disassembled based on the complexity assessment results; S2.

2. Break down complex tasks into multiple subtasks, each corresponding to a specific review point or drawing area, and assign a unique identifier and priority to each subtask; S2.

3. Define the required model type, parameter settings and expected output according to the subtask content; S2.

4. Introduce task priority and urgency evaluation mechanism, formulate task distribution strategy, and select the most appropriate model for task distribution based on model load and performance; S2.

5. Distribute the subtasks to the corresponding single-task models and monitor the task distribution process; S2.

6. Collect feedback from the model on task distribution, adjust the task distribution strategy based on the feedback, and optimize resource allocation.

4. The general design review architecture method based on AI large model comprehensive scheduling small model according to claim 3 is characterized in that: The implementation steps of step S3 are: S3.

1. The visual analysis model loads the drawing file submitted by the user and performs preprocessing operations; S3.

2. Use deep learning technology to identify graphic elements in drawings, conduct preliminary verification of the recognition results, and eliminate erroneous or redundant elements; S3.

3. Verify the identified symbols and annotations according to industry specifications and enterprise standards to check the correctness of the symbols, accuracy and consistency of the annotations; S3.

4. Classify the identified and verified graphic elements, including structural elements and electrical elements, to provide classification information for subsequent review; S3.

5. For complex graphics and special symbols, use the advanced feature extraction capabilities of deep learning models to identify them and combine them with contextual information; S3.

6. Record the verification results of the graphic elements into the database, including pass, fail and problem details, and generate a verification report.

5. The general design review architecture method based on AI large model comprehensive scheduling small model according to claim 4 is characterized in that: The implementation steps of step S4 are: S4.

1. The rule generation model parses industry specifications and enterprise standard documents, extracts key information, and performs structured processing on the documents; S4.

2. Based on the parsing results, a standard rule base in XML format is automatically generated, which includes rule entries, descriptions, and applicable condition information; S4.

3. Verify the generated rules to ensure their accuracy and applicability, and optimize and adjust the rules based on user feedback and actual needs; S4.

4. Support dynamic update of rule base. When the specification or standard changes, update the rule base in time. S4.

5. Introduce the user-defined rule function, allowing users to input custom rules through natural language. The system automatically parses the user input, generates corresponding rule entries, and adds them to the rule library; S4.

6. Perform version management on the rule base, record the content and time of each update, and provide version rollback function.

6. The general design review architecture method based on AI large model comprehensive scheduling small model according to claim 5 is characterized in that: The implementation steps of step S5 are: S5.

1. The rule-based review system receives tasks distributed by the large model and matches corresponding rules according to task requirements; S5.

2. Perform mathematical and logical checks, including dimensional calculations, material selection, and process requirements; S5.

3. Record the review results in the database, including pass, fail and problem details, and generate a review report; S5.

4. Locate the failed review points, mark the problem locations and modification suggestions, and provide intuitive feedback; S5.

5. Introduce machine learning technology to continuously optimize the rule-based review system based on historical review data and user feedback; S5.

6. Feedback the review results to the big model and provide preview and export functions for the review report.

7. The general design review architecture method based on AI large model comprehensive scheduling small model according to claim 6 is characterized in that: The implementation steps of step S6 are: S6.

1. Collect data from each single-task model, including visual review results, text processing results, and rule checking results; S6.

2. Integrate the collected data to form a unified review result data set, and clean and format the data; S6.

3. Verify the integrated data to check the relevance and consistency between the data and eliminate data conflicts; S6.

4. Convert and store data according to the requirements of subsequent steps; S6.

5. Introduce distributed computing and cloud computing technologies to improve the efficiency and scalability of multi-module collaboration and use cloud computing resources to achieve rapid data processing and storage; S6.

6. Provide a data sharing interface to allow other systems or modules to access the review result data and ensure the security and stability of the data interface.

8. The general design review architecture method based on AI large model comprehensive scheduling small model according to claim 7 is characterized in that: The implementation steps of step S7 are: S7.

1. Select appropriate report templates based on user needs and review results, and support customized report templates; S7.

2. Fill in the report template with the review results, problem description, and modification suggestions; S7.

3. Adjust the report format according to user needs, including font, font size, and layout, and support export in multiple formats, including PDF, Word, and Excel; S7.

4. Introduce data visualization technology to intuitively display the review results in the form of charts and animations to improve the readability and comprehension of the report; S7.

5. Review the generated reports, publish the approved reports to users or related systems, and provide download or viewing links; S7.

6. Perform version management on the report, record the time and content of each generation and modification, and provide version rollback function.

9. The general design review architecture method based on AI large model comprehensive scheduling small model according to claim 8 is characterized in that: The implementation steps of step S8 are: S8.

1. Mark the problems found in the review results on the design drawings, provide modification suggestions, and feedback the marked results and modification suggestions to the user or design software to form a closed-loop management; S8.2.Integrate with design software to support direct viewing and modification of annotated issues in the design software.

10. The general design review architecture method based on AI large model comprehensive scheduling small model according to claim 9 is characterized in that: The implementation steps of step S9 are: S9.

1. Establish a regular review and evaluation mechanism to regularly evaluate the effectiveness of question annotation, feedback, and intelligent recommendation systems. Based on the evaluation results, continuously optimize and improve the annotation process, feedback mechanism, and intelligent recommendation system. S9.

2. Set clear evaluation indicators, including the improvement rate of problem rectification efficiency, the reduction rate of design errors, and user satisfaction. Through data analysis, quantitatively evaluate the implementation effect of various improvement measures.

Images