Ship test plan generation system and method based on knowledge graph and AI large model

Abstract

The application relates to the technical field of ship manufacturing, and particularly discloses a ship test scheme generation method and system based on an AI large model and a knowledge graph, which comprises the following steps: acquiring national standards, industry standards and ship classification society standards related to ship manufacturing, and constructing a structured knowledge graph; guiding a general large language model deployed locally based on the knowledge graph, generating a test scheme text and a record table conforming to the standards according to user input requirements; and automatically determining the results and generating a test report in a corresponding format according to the recorded data after the test is completed. The application realizes intelligent generation of test schemes and reports, and significantly improves the efficiency and standardization of ship manufacturing quality control.

Description

Technical Field

[0001] This invention relates to the field of quality inspection and intelligent application in shipbuilding, and in particular to a ship test scheme generation system and method based on knowledge graphs and AI large models. Background Technology

[0002] During the shipbuilding process, corresponding tests and inspections need to be carried out according to the characteristics of different stages and equipment to ensure compliance with various standards and guarantee delivery quality. Traditional methods rely on manual preparation of test plans and reports, which is cumbersome and prone to errors. In recent years, although large language models have performed well in text generation, they have problems such as lagging knowledge updates, non-standard output formats, and poor content controllability in professional fields, making it difficult to directly meet the professional needs of the ship testing field.

[0003] This invention aims to provide an intelligent generation system and method for ship test plans that integrates knowledge graphs and AI large-scale models. It addresses the problems of low efficiency and poor accuracy in generating ship test plans and reports in existing technologies, caused by a lack of professional knowledge in large-scale language models, uncontrollable generated content, and non-standard formats. The core of this invention lies in combining domain knowledge in the form of structured knowledge graphs with the natural language generation capabilities of large-scale language models to achieve intelligent and standardized processes from test plan generation and test process support to report generation. Summary of the Invention

[0004] The purpose of this invention is to provide a ship test scheme generation system based on knowledge graphs and AI large models to solve the problems existing in the prior art.

[0005] The above-mentioned technical objective of the present invention is achieved through the following technical solution: The ship test scheme generation system based on knowledge graph and AI big model includes a cloud platform, a database, a control terminal and a mobile terminal. The cloud platform is used to realize wireless data interaction between the mobile terminal and the control terminal. The database is used to store historical data generated by the control terminal. The control terminal is used to generate test schemes based on historical data and requests sent by the mobile terminal.

[0006] In a further embodiment, the database includes a standard file storage area, a knowledge graph storage area, and a historical data storage area. The standard file storage area is used to store standard files, the knowledge graph storage area is used to construct testing process specification documents based on national standards, industry standards, and classification society standards, and the historical data storage area is used to store historical data.

[0007] In a further embodiment, the control terminal includes an intelligent agent and a computing power center. The intelligent agent is used to acquire manually entered instruction information, and to retrieve relevant files from the knowledge graph storage area based on the manually entered instruction information to generate a detection process specification document. The intelligent agent is also used to update the verification process document based on the final detection result conforming to the detection process. The computing power center is used to allocate computing power to the intelligent agent.

[0008] This invention also discloses a production method for a ship test scheme generation system based on knowledge graphs and AI large models, comprising the following steps: Step S1, Preparation Stage: A knowledge graph is established using standard documents, and search terms are selected based on the knowledge graph. The search terms are then used to perform a rapid enhanced search on the knowledge graph. Step S2: Generate a detection process specification document. By confirming the manually entered target to be detected, the intelligent agent retrieves the corresponding document from the knowledge graph storage area and sends it to the mobile terminal for manual verification. Once the manual verification is correct, the document is sent out. Step S3: Result Feedback. The target to be detected is detected according to the detection process specification document. By recording the detection process, the agent checks whether there are any loopholes in the detection process specification document. If there are, the loopholes are filled and the results are fed back to the knowledge graph storage area, and the corresponding knowledge graph is iteratively updated. If there are no loopholes, the structure is recorded and saved.

[0009] In a further embodiment, the standard file is obtained from publicly available information channels using a periodic update method, and the storage process of the standard file is as follows: Step S11: Confirm the standard core vocabulary and periodically check whether the publicly available information channels have been updated using the core vocabulary. Step S12: Determine the update cycle and perform periodic updates based on the historical update data of the region corresponding to the standard document; Step S13: Standard document entry verification. The standard documents are manually verified before being entered into the database.

[0010] In a further embodiment, a search term filtering process is established, which is as follows: Step S14: Retrieve the detection area of ​​the item to be detected, and determine the search range of the standard document based on the detection area; Step S15: Search for the commercial nature of the item to be tested, and determine the testing standards based on the commercial nature; Step S16: Retrieve the category of the item to be tested. Based on the process category of the item to be tested, retrieve the corresponding classification file in the standard file. Step S17: Obtain the final standard document for the corresponding process step according to the process steps.

[0011] In a further embodiment, the method for generating the knowledge graph is as follows: Step S18: Obtain the standard file and recognize the standard text information using OCR; Step S19: Using the main body of the detection project as the node, and the standard attributes and standard specific text as the node attributes, establish a knowledge graph database through the nodes and attributes.

[0012] In a further embodiment, the detection process is recorded by a combination of manual intervention and audio-visual files.

[0013] In summary, the present invention has the following beneficial effects: 1. Through the deep integration of knowledge graphs and AI big models, the entire process of intelligent generation of ship test plans and reports has been realized, which greatly reduces the workload of manual compilation, significantly improves the efficiency and response speed of document compilation, and meets the needs of shipyards and other users for rapid delivery of test documents. 2. By constructing a four-level domain knowledge graph, national standards, industry standards, and classification society specifications are stored in a structured manner and used as the core basis for generating content for large models. This effectively solves the problems of outdated professional domain knowledge, uncontrollable content, and non-standard format in general large models, ensuring the accuracy and compliance of test plans and reports. 3. By deploying large models locally and supporting multimodal automatic detection, the security of enterprise data is ensured. At the same time, the automatic judgment and report generation of test results are realized, which further improves the automation level of quality control, reduces the human error rate, and has good scalability and adaptability, which can be extended to other testing links in shipbuilding. Attached Figure Description

[0014] Figure 1 This is a system block diagram of the intelligent agent operation process of the present invention; Figure 2 This is a flowchart of the knowledge graph manufacturing process of the present invention; Figure 3 This is a four-layer structured system block diagram of the standard system knowledge graph used to embody the present invention. Detailed Implementation

[0015] The present invention will be further described in detail below with reference to the accompanying drawings.

[0016] It should be noted that in the description of this invention, any descriptions of orientation, such as up, down, front, back, left, right, etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings. They are only for the purpose of facilitating the description of this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed or operated in a specific orientation, and should not be construed as a limitation of this invention.

[0017] Example 1: like Figures 1-3 As shown, a ship test scheme generation system based on knowledge graphs and AI large models includes a cloud platform, a database, a control terminal, and a mobile terminal. The cloud platform enables wireless data interaction between the mobile terminal and the control terminal. The database stores historical data generated by the control terminal, and the control terminal generates test schemes based on historical data and requests sent by the mobile terminal. The database includes a standard file storage area, a knowledge graph storage area, and a historical data storage area. The standard file storage area stores standard files, and the knowledge graph storage area is used to construct test scheme specification documents based on national standards, industry standards, and classification society standards. The historical data storage area stores historical data. The control terminal has an intelligent agent and a computing power center. The intelligent agent acquires manually entered instructions and retrieves relevant files from the knowledge graph storage area to generate a test process specification document. The intelligent agent also updates the verification process document based on the final test results conforming to the test process. The computing power center allocates computing power to the intelligent agent.

[0018] like Figures 1-3 As shown, the generation method includes the following steps: Step S1, Preparation Stage: A knowledge graph is established using standard documents, and search terms are selected based on the knowledge graph. The search terms are then used to perform a rapid enhanced search on the knowledge graph. Step S2: Generate a detection process specification document. By confirming the manually entered target to be detected, the intelligent agent retrieves the corresponding document from the knowledge graph storage area and sends it to the mobile terminal for manual verification. Once the manual verification is correct, the document is sent out. Step S3, Result Feedback: The target to be detected is detected according to the detection process specification document. By recording the detection process, the agent checks whether there are any loopholes in the detection process specification document. If there are, the loopholes are filled and the results are fed back to the knowledge graph storage area, and the corresponding knowledge graph is iteratively updated. If there are no loopholes, the structure is recorded and saved.

[0019] The standard documents are obtained from publicly available information channels using a periodic update method. The storage process for the standard documents is as follows: Step S11: Confirm the standard core vocabulary and periodically check whether the publicly available information channels have been updated using the core vocabulary. Step S12: Determine the update cycle and perform periodic updates based on the historical update data of the region corresponding to the standard document; Step S13: Standard document entry verification. The standard documents are manually verified before being entered into the database.

[0020] Establish a search term filtering process, which is as follows: Step S14: Retrieve the detection area of ​​the item to be detected, and determine the search range of the standard document based on the detection area; Step S15: Search for the commercial nature of the item to be tested, and determine the testing standards based on the commercial nature; Step S16: Retrieve the category of the item to be tested. Based on the process category of the item to be tested, retrieve the corresponding classification file in the standard file. Step S17: Obtain the final standard document for the corresponding process step according to the process steps.

[0021] The method for generating the knowledge graph is as follows: Step S18: Obtain the standard file and recognize the standard text information using OCR; Step S19: Using the main body of the detection project as the node, and the standard attributes and standard specific text as the node attributes, establish a knowledge graph database through the nodes and attributes.

[0022] The detection process is recorded by a combination of manual intervention and audio-visual documentation.

[0023] Example 2: This invention can also filter the acquired information by embedding a large language model at the system layer. The large language model adopts a general open source model, is deployed on a local high-computing-power server, and provides dialogue and reasoning services.

[0024] The core of this invention lies in combining domain knowledge in the form of a structured knowledge graph with the natural language generation capabilities of a large language model to achieve intelligent and standardized processes from experimental scheme generation and experimental process support to report generation.

[0025] The system includes a knowledge graph construction module, a large language model service module, a retrieval and generation module, an experiment execution support module, and a report generation module.

[0026] The knowledge graph construction module is responsible for extracting entity, relation, and detection process information from multi-source standard documents to build a domain knowledge graph with a four-level structure; the large language model service module provides a locally deployed general-purpose large language model to ensure data security and response efficiency; the retrieval and generation module performs multi-level retrieval in the knowledge graph based on user input and uses the retrieval results to construct prompt words to guide the large model to generate standardized content; the test execution support module provides both manual and automatic detection methods, with automatic detection achieved using multimodal models or computer vision technology; and the report generation module automatically determines compliance and outputs a formatted report based on test results and standard requirements.

[0027] This invention also provides a method for intelligent generation of ship tests based on the above system, comprising the following steps: System construction phase: Deploy the large language model on the local server, and extract knowledge from standard documents based on OCR and entity recognition technologies to build a structured, dynamically updatable knowledge graph; Experimental scheme generation stage: Obtain the standard content corresponding to the user's instructions through multi-level vector retrieval, and use it as context input to the large model to generate experimental schemes and record tables that meet the format requirements; During the test execution phase: test results can be entered manually or automatically filled in through image recognition; Test report generation stage: The test results are assessed for compliance. For projects that meet the requirements, a final report is automatically generated. For projects that do not meet the requirements, a review or retest is prompted.

[0028] Through the above system and method, the present invention can significantly improve the accuracy, efficiency and standardization of ship test document generation, reduce manual intervention, reduce error rate, and has good scalability and adaptability.

[0029] Regarding the deployment of large language models, this solution selects an open-source general-purpose large language model, deploys it on a local server equipped with a GPU, and uses inference frameworks such as vLLM or Ollama to achieve efficient inference under high concurrency. The model must have complete instruction following and text generation capabilities, and provide services through dialogue interfaces (such as OpenAI-compatible APIs).

[0030] Regarding the construction of knowledge graphs, taking welding as an example, it is necessary to first collect all standard documents related to weld inspection, including but not limited to national standards, industry standards, classification society specifications such as CCS / LR / DNV, and internal enterprise standards. For paper documents, a high-precision scanner is first used to convert them into electronic images, and then PaddleOCR or Tesseract engines are used for text recognition to output structured text. The national standards mentioned here are not limited to Chinese national standards, but also include relevant standards formulated by various sovereign states. These standards are stored in partitions according to country.

[0031] The knowledge graph construction process includes the following steps: Step A1: Entity recognition. Using the deployed large language model, entities and relationships are extracted by designing prompt words. The model output results are parsed and manually verified to ensure accuracy. The results can also be reviewed through the configuration annotation platform interface.

[0032] Step A2: Graph Construction. The identified entities and relationships are imported into the Neo4j graph database. The graph schema is designed as a four-level structure, such as... Figure 3 The structure shown: The first level is the standard system, such as the classification rules for steel seagoing vessels based on the country. The second level is the applicable objects, which refer to the ship's positional structure, such as the hull structure and piping system. The third level includes inspection items, such as visual inspection of butt welds and non-destructive testing. Level 4: Testing procedures and acceptance criteria, such as testing environment requirements and acceptance standards; Vectorization involves using a text embedding model to convert all text nodes in the knowledge graph into vectors, which are then stored in a vector database for subsequent retrieval.

[0033] Regarding the test plan generation stage, that is, the method for generating the test process specification document, the method is as follows: The user inputs instructions through the Web interface or API, such as generating a visual inspection test plan for the thick plate butt welds of the hull structure. In this sentence, it is necessary to extract the three keywords of hull structure, thick plate butt welds, and visual inspection in sequence, and retrieve the corresponding data information from the knowledge graph based on these three keywords. The retrieval process includes the system vectorizing the user query and performing an approximate nearest neighbor (ANN) search in the vector database. The retrieval strategy is to refine the process step by step, first matching the standard system, then the applicable objects, then the testing items, and finally locating the specific testing steps and acceptance standard nodes. Regarding the construction and generation of prompt words, the relevant text node content retrieved will be inserted into the following preset prompt word templates. The text content will be extracted according to the testing steps, instrument requirements, and safety precautions. The preset prompt word templates include: You are a ship test program expert; Please strictly follow the following standard requirements; Generate test programs and record forms for users.

[0034] Standard requirement: Insert the retrieved standard text; User requirements: Original user instructions; Please generate one or more of the following: a test plan overview, a list of required equipment, detailed test procedures, and a blank test record form. The form should include columns for header, test items, standard requirements, measured values, and result judgment. A prompt should also be provided; please use Markdown format for the prompt output, ensuring completeness and proper formatting.

[0035] The large model generates content based on this prompt word. The system has a verification mechanism: if the key content (such as steps and standard values) in the model output is seriously inconsistent with the knowledge graph, it will trigger regeneration or an alarm.

[0036] Regarding the execution phase of the test process, the generated test forms are first printed or downloaded to mobile devices for on-site use.

[0037] Then comes the manual inspection: the inspector performs the test according to the test procedure and manually fills in the measurement results in the form.

[0038] Finally, automated inspection is performed (taking weld appearance inspection as an example): the inspector takes photos of the weld and uploads them to the system. The system uses a trained computer vision model or a multimodal large model to analyze the photos.

[0039] In this process, the prompts are designed as follows: Please analyze the weld image to determine if there are defects such as cracks, undercut, or lack of fusion. If so, please output the defect type, location, and severity. Finally, please give a conclusion of whether the weld is acceptable or unacceptable.

[0040] After the model output is analyzed, the results are automatically filled into the result judgment column of the test table.

[0041] Regarding the test report generation stage, after all test results are summarized, they are input into the large model to generate the final report. The results here include those entered manually and those automatically recognized.

[0042] Furthermore, compliance assessment is required: the large model compares the measured values ​​of each test item with the acceptance criteria stored in the knowledge graph to determine whether it is qualified.

[0043] Finally, report generation: For all qualified projects, the large model generates the final test report according to the preset report template (Word or LaTeX format required).

[0044] Example prompt: "Please generate a formal test report based on the following test data and standard requirements. The report must include: test overview, test basis, test equipment, test results summary table (including test items, standard requirements, measured values, and conclusions), overall conclusions, and an approval column. Test data: [Insert structured test data] Standard requirements: [Insert relevant standard text] Note: The report format must strictly follow the company's document specifications (a template is provided in the attachment)." If any non-conformities are found, the system will generate a non-conformity report and trigger a notification process to remind relevant personnel to take action.

[0045] Through the above specific implementation methods, this invention achieves full automation and intelligence in the weld inspection process, from scheme generation to report delivery, ensuring the accuracy of technical content and the standardization of format, and greatly improving the efficiency and reliability of quality inspection in the shipbuilding process. In the description of this invention, unless otherwise explicitly defined, terms such as "set up," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0046] This specific embodiment is merely an explanation of the present invention and is not intended to limit the invention. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A ship test scheme generation system based on knowledge graphs and AI large models, characterized by: It includes a cloud platform, a database, a control terminal, and a mobile terminal. The cloud platform is used to enable wireless data interaction between the mobile terminal and the control terminal. The database is used to store historical data generated by the control terminal. The control terminal is used to generate experimental plans based on the historical data and requests sent by the mobile terminal.

2. The ship test scheme generation system based on knowledge graph and AI large model according to claim 1, characterized in that: The database includes a standard file storage area, a knowledge graph storage area, and a historical data storage area. The standard file storage area is used to store standard files. The knowledge graph storage area is used to construct testing scheme specification documents based on national standards, industry standards, and classification society standards. The historical data storage area is used to store historical data.

3. The ship test scheme generation system based on knowledge graph and AI large model according to claim 1, characterized in that: The control terminal includes an intelligent agent and a computing power center. The intelligent agent is used to obtain manually entered instruction information, and retrieves relevant files from the knowledge graph storage area to generate a detection process specification document based on the manually entered instruction information. The intelligent agent is also used to update the verification process document based on the final detection result conforming to the detection process. The computing power center is used to allocate computing power to the intelligent agent.

4. A method for generating a ship test scheme generation system based on knowledge graphs and AI large models according to any one of claims 1-3, characterized in that, Includes the following steps: Step S1, Preparation Stage: A knowledge graph is established using standard documents, and search terms are selected based on the knowledge graph. The search terms are then used to perform a rapid enhanced search on the knowledge graph. Step S2: Generate a detection process specification document. By confirming the manually entered target to be detected, the intelligent agent retrieves the corresponding document from the knowledge graph storage area and sends it to the mobile terminal for manual verification. Once the manual verification is correct, the document is sent out. Step S3, Result Feedback: The target to be detected is detected through the detection process specification document. By recording the detection process, the intelligent agent checks whether there are any loopholes in the detection process specification document. If there are, the loopholes are filled and the results are fed back to the knowledge graph storage area, and the corresponding knowledge graph is iteratively updated. If it does not exist, record the structure and save it.

5. The generation method according to claim 4, characterized in that: The standard documents are obtained from publicly available information channels using a periodic update method. The storage process for the standard documents is as follows: Step S11: Confirm the standard core vocabulary and periodically check whether the publicly available information channels have been updated using the core vocabulary. Step S12: Determine the update cycle and perform periodic updates based on the historical update data of the region corresponding to the standard document; Step S13: Standard document entry verification. The standard documents are manually verified before being entered into the database.

6. The generation method according to claim 4, characterized in that: Establish a search term filtering process, which is as follows: Step S14: Retrieve the detection area of ​​the item to be detected, and determine the search range of the standard document based on the detection area; Step S15: Search for the commercial nature of the item to be tested, and determine the testing standards based on the commercial nature; Step S16: Retrieve the category of the item to be tested. Based on the process category of the item to be tested, retrieve the corresponding classification file in the standard file. Step S17: Obtain the final standard document for the corresponding process step according to the process steps.

7. The generation method according to claim 4, characterized in that: The method for generating the knowledge graph is as follows: Step S18: Obtain the standard file and recognize the standard text information using OCR; Step S19: Using the main body of the detection project as the node, and the standard attributes and standard specific text as the node attributes, establish a knowledge graph database through the nodes and attributes.

8. The generation method according to claim 4, characterized in that: The detection process is recorded by a combination of manual intervention and audio-visual documentation.

Images