Product external inspection AI copilot system based on large model, construction method and storage medium

Abstract

The invention belongs to the technical field of appearance detection, and particularly relates to a product external inspection AI co-driving system based on a large model, a construction method and a storage medium. The invention discloses an external inspection AI co-driver system construction method based on a large model, and the method comprises the steps: obtaining the historical experience data of an enterprise, carrying out the fragmentation modularization processing of the obtained historical experience data, and forming an external inspection post knowledge base; data in the external inspection post knowledge base are subjected to vector model coding vectorization processing and stored in a vector database; establishing a vector index model and a vector recall model; establishing a language interaction relationship between the vector database and the large model through a vector index model and a vector recall model; the user side obtains the content of the vector database through large model questions to complete real-time questions and answers; and performing information supplement on real-time dialogue data of the user side, performing key information extraction on the dialogue data, and supplementing the dialogue data into the vector database. Standardized operation guidance is provided for an on-site appearance detection post, so that the appearance detection defect identification efficiency and disposal efficiency are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of appearance inspection, and specifically relates to an AI co-pilot system for product external inspection based on a large model. Background Art

[0002] During the production and maintenance of steel products, appearance quality inspection is required. The current appearance inspection process for steel products requires manual inspection of each finished product in the warehouse, relying on manual experience to discover and judge appearance defects. Common defect types include cracks, pits, scratches, etc. Each major defect category has several subcategories.

[0003] Traditional product appearance inspection methods rely on manual labor and have the following pain points:

[0004] (1) The defect recognition methods and judgment results of appearance inspection vary from person to person, and there is a lack of unified standard guidance;

[0005] (2) Experienced operators are required to accurately determine the defect type, cause, and treatment measures;

[0006] (3) For situations that are difficult to identify and judge, it is necessary to organize experienced technical personnel to make comprehensive judgments, which is inefficient;

[0007] (4) Appearance inspection work experience remains in the minds of technicians or computers and has not yet formed an enterprise knowledge base;

[0008] (5) Relying on the traditional master-apprentice model, the training cycle for new employees is long.

[0009] Traditional product appearance inspections make it difficult for the historical experience of many outstanding individuals to be fully shared and passed on. Existing defect detection methods, which rely on cameras and deep learning modules for automated inspection, suffer from inaccurate recognition and judgment, leading to inaccurate inspections.

[0010] Based on the above problems, the applicant has constructed an AI co-pilot system for external inspection of steel industry products that can transform historical experience of appearance inspection into knowledge and form unified standards for appearance inspection. It is used for standardized operation guidance of on-site appearance inspection positions to improve the efficiency of appearance inspection defect identification and handling. Summary of the Invention

[0011] The purpose of the present invention is to overcome the defects in the prior art that the appearance inspection work experience lacks standards and the inspection efficiency and accuracy vary from person to person, and to provide an AI co-pilot system for product external inspection based on a large model, a construction method and a storage medium for standardized work guidance in on-site appearance inspection positions.

[0012] The technical solution adopted by the present invention to solve its technical problem is:

[0013] The first aspect is a method for building an external inspection AI co-pilot system based on a large model, which includes the following:

[0014] Obtain the company's historical experience data and perform segmented and modular processing on the acquired historical experience data to form a knowledge base for external inspection positions;

[0015] The data in the external inspection position knowledge base is vectorized through vector model encoding and stored in the vector database;

[0016] Establishing a vector index model and a vector recall model, i.e., labeling and editing the data in the vector database, and using the data stored in the vector database to label and recall questions and answers;

[0017] The vector database and the big model establish a language interaction relationship through the vector index model and the vector recall model; the user terminal asks questions through the big model to obtain the content of the vector database and complete real-time question and answer;

[0018] The real-time conversation data of the user end is supplemented with information and key information is extracted from the conversation data and supplemented into the vector database.

[0019] Specifically, the acquired historical experience data includes one or more of picture data, text data, voice data, video data, and report documents;

[0020] The acquired historical experience data is processed in a fragmented and modular manner, including the following contents:

[0021] Separate historical experience data according to data type;

[0022] Organize and clean the separated image data and / or text data and / or voice data and / or video data and / or report documents, and process them into a unified standard format;

[0023] The data in the standard format is encoded and segmented according to the data length, and then divided into several small blocks according to semantics.

[0024] Specifically, the data in the external inspection post knowledge base is encoded and vectorized by a vector model, including the following contents:

[0025] Using natural language processing technology to parse the text information divided into several small blocks according to semantics, and establishing a vector unit for natural language understanding and reasoning through a model encoder;

[0026] Computer vision algorithms are used to perform defect recognition and defect labeling on the image information that is semantically separated into several small blocks, and the generated label information is converted into a vector form.

[0027] Specifically, the defect marking includes any one of defect frame marking and defect arrow marking.

[0028] Specifically, the process of supplementing the user's real-time conversation data and extracting key information from the conversation data and adding it to the vector database includes the following:

[0029] Dynamically adjust the size of the user-side context window;

[0030] Using information extraction technology, key information in each round of conversation is extracted and vectorized;

[0031] The key information after vectorization processing is supplemented and stored in the vector database.

[0032] Specifically, the answer data includes one or more of defect type data, cause data, solution data, and example pictures.

[0033] The second aspect is a large-scale model-based AI co-pilot system for product inspection, including:

[0034] Enterprise historical experience data file; the historical experience data file includes at least one of image data, text data, voice data, video data and report documents;

[0035] A document parsing and splitting module performs document parsing on the historical experience data file and splits it into image data and / or text data and / or voice data and / or video data and / or report documents;

[0036] Data vectorization module, used to perform vectorization processing on split and classified data;

[0037] A vector database, used to store the data processed by the data vectorization module;

[0038] The AI big model serves as the client input port and matches the data information of the query problem with the vector database through the vector index module and the vector recall module;

[0039] The information supplement module includes a dynamic adjustment unit and a key information extraction unit; the dynamic adjustment unit is used to dynamically adjust the size of the user terminal context window, and the key information extraction unit is used to extract key information in the conversation.

[0040] Specifically, it also includes a quantitative annotation module and a retrieval enhancement module. The vectorized annotation module is used to provide quantitative annotations to the administrator background and / or users, and the retrieval enhancement module is used to filter and bind the annotated correct answers according to the administrator background standards and / or the quantitative annotations of users' likes and dislikes, so as to enhance the index matching degree.

[0041] Specifically, it also includes a data detection and analysis module for analyzing usage data and calculating key indicators for display; the key indicators include at least one of the number of users, number of active users, large language model call consumption, usage type trend, total number of messages, average number of appropriate interactions, and output speed.

[0042] The third aspect is a storage medium storing processor-executable instructions, which, when executed by the processor, are used to execute a method for constructing a large-model-based product external inspection AI co-pilot system as described above.

[0043] The beneficial effects of the large-model-based product inspection AI co-pilot system, construction method, and storage medium of the present invention are:

[0044] The present invention performs fragmented and modular processing on the text data and image data in the historical experience data, that is, independently processes the text data and image data by adapting their respective characteristics to form an external inspection post knowledge base, improves the accuracy of the data in the external inspection post knowledge base, vectorizes and stores the data in the external inspection post knowledge base, uses semantic vectorization technology to improve retrieval efficiency and accuracy, and makes content-based retrieval more accurate and effective. It also realizes language interaction between the vector database and the large model by establishing a vector index module and a vector recall module, and provides standardized operation guidance for on-site appearance inspection posts to improve the efficiency of appearance inspection defect recognition and disposal.

[0045] The present invention supplements user-side conversation data to avoid the problem of losing access to previous conversation content due to conversations exceeding context limits. By dynamically adjusting the context window, it can flexibly adapt to conversation histories of varying lengths based on actual conversation needs, ensuring that information is not lost and that its integrity and continuity are maintained. Furthermore, the present invention extracts key information from user-side conversation data and supplements it into vector database 4, ensuring that the data in vector database 4 is updated in real time, improving the accuracy and relevance of responses. In subsequent conversations, this vectorized key information can also be queried to maintain conversation continuity and ensure conversation quality, thereby improving the user experience and enhancing the system's flexibility and responsiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] Figure 1 This is a flowchart of the method for constructing the product external inspection AI co-pilot system in this embodiment 1.

[0048] Figure 2This is a diagram of the construction process of the product external inspection AI co-pilot system in this embodiment 1.

[0049] Figure 3 This is a simplified diagram of the formation of the vector database in the first embodiment.

[0050] Figure 4 This is a diagram of the large model application framework in the first embodiment.

[0051] Figure 5 This is a flow chart of the sharding and modular processing of historical experience data in the first embodiment.

[0052] Figure 6 This is a flowchart of the information supplement part in this embodiment 1.

[0053] Figure 7 This is a screenshot of the software interface for natural language dialogue in the first embodiment.

[0054] Figure 8 This is a screenshot of the software interface for marking and replying to a defect frame with a crack defect in the first embodiment.

[0055] Figure 9 This is a structural diagram of the product external inspection AI co-pilot system in the second embodiment.

[0056] Figure 10 This is a diagram showing the analysis data of the data detection and analysis module on the software interface in the second embodiment.

[0057] In the figure: 1. Enterprise historical experience data file, 2. Document parsing and splitting module, 3. Data vectorization module, 4. Vector database, 5. AI large model, 6. Vector indexing module, 7. Vector recall module, 8. Information supplement module, 81. Dynamic adjustment unit, 82. Key information extraction unit, 9. Quantitative labeling module, 10. Retrieval enhancement module, 11. Data detection and analysis module. DETAILED DESCRIPTION

[0058] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0059] Example 1

[0060] This embodiment takes the appearance inspection of products in the steel industry as an example. Figures 1-8 The specific embodiment of the method for constructing a large-scale model-based product inspection AI co-pilot system of the present invention includes the following contents:

[0061] S10: Acquire the company's historical experience data and perform segmented and modular processing on the acquired historical experience data to form a knowledge base for external inspection positions in the steel industry;

[0062] S20: The data in the external inspection post knowledge base is vectorized by vector model encoding and stored in the vector database 4;

[0063] S30: Establishing a vector index model and a vector recall model, i.e., annotating and editing the data in the vector database 4, and annotating and recalling questions and answers using the data stored in the vector database 4;

[0064] S40: Establishing a language interaction relationship between the vector database and the large model through the vector index model and the vector recall model; the user terminal asks questions through the large model to obtain the content of the vector database and complete real-time question and answer;

[0065] S50: Supplement the real-time conversation data of the user end and extract key information from the conversation data and add it to the vector database 4.

[0066] This embodiment performs fragmented and modular processing on the text data and image data in the historical experience data, that is, independently processes the text data and image data by adapting their respective characteristics to form an external inspection position knowledge base, improves the accuracy of the data in the external inspection position knowledge base, vectorizes and stores the data in the external inspection position knowledge base, uses semantic vectorization technology to improve retrieval efficiency and accuracy, and makes content-based retrieval more accurate and effective. It also realizes language interaction between the vector database 4 and the large model by establishing a vector index module 6 and a vector recall module 7, and provides standardized operation guidance for on-site appearance inspection positions to improve the efficiency of appearance inspection defect recognition and disposal.

[0067] The historical experience data acquired in step S10 includes one or more of image data, text data, voice data, video data, and report documents. In this embodiment, the enterprise's historical experience data includes image data and text data related to appearance inspections in the steel industry. It should be understood that the historical experience data in this embodiment only lists text data and image data, but is not limited to text data and image data. Appearance defects in the steel industry include cracks, pits, scratches, etc. Cracks can be refined according to the causes of their formation. The causes of pits and scratches can be refined and divided into subcategories based on their size and dimensions to ensure the integrity and accuracy of the data. This embodiment takes the appearance of products in the steel industry as an example, but of course it can also be applied to other industries, and no absolute limitation is made here.

[0068] For further explanation, see Figure 5 In this embodiment, the historical experience data obtained is processed in a fragmented and modular manner, including the following contents:

[0069] S101: Separate historical experience data according to data type;

[0070] S102: arranging and cleaning the separated image data and / or text data and / or voice data and / or video data and / or report documents, and processing them into a unified standard format;

[0071] S103: Encode the data in the standard format and segment it according to the data length, and divide it into several small blocks according to semantics.

[0072] This step, S10, extracts the company's accumulated historical experience to form high-quality corpus for large-scale model input, including text data and image data. This data is then segmented and modularized to construct a knowledge base for external inspection positions. This knowledge base incorporates the work experience of multiple experienced technicians to form an enterprise-level external inspection knowledge base, which can be flexibly maintained and expanded as experience accumulates. This knowledge base is primarily used for on-site operational guidance and new employee training, greatly facilitating knowledge sharing and learning within the company and promoting knowledge sharing and experience transfer.

[0073] In step S20, the data in the external inspection position knowledge base is vectorized through vector model encoding, and the text information divided into several small blocks according to semantics is parsed using natural language processing technology. A vector unit for natural language understanding and reasoning is established through a model encoder, and a computer vision algorithm is used to perform defect identification and defect labeling on the image information divided into several small blocks according to semantics, and the generated label information is converted into a vector form.

[0074] In this embodiment, the text data and image data in the external inspection post knowledge base are separated and processed, and vectorized using vector model encoding and modules suitable for themselves. The text data is parsed using natural language processing technology. Specifically, a vector unit for natural language understanding and reasoning is established through a model encoder. Computer vision algorithms are used to identify and annotate defects in image data, and the generated label information is converted into vector form. The vectorized text data and image data are stored in the vector database 4. Semantic vectorization technology is used to improve retrieval efficiency and accuracy, making content-based retrieval more accurate and effective. Through this meticulous multimodal information processing strategy, not only the response speed of the retrieval operation is enhanced, but also the relevance and accuracy of the retrieval results are improved, thereby optimizing the overall information management process.

[0075] The defect marking includes any one of defect frame marking and defect arrow marking. In this embodiment, the defect marking adopts defect frame marking, such as Figure 8 As shown, the defect frame marking of the crack defect is shown.

[0076] The data stored in the vector database 4 is used to mark and recall questions and answers to increase the accuracy of the answers. There are manual background annotations and recall data content for editing and correction to ensure the accuracy of the answer data.

[0077] In step S40 of this embodiment, a vector index model and a vector recall model are set between the vector database 4 and the large model, and a language interaction relationship is established between the vector database 4 and the large model. The constructed vector index model is used to match and index the question and answer data, and the vector recall model is used to recall the answer data, edit and correct it, and obtain work instruction information in real time to update the stored data in the vector database 4.

[0078] This embodiment uses a large model as the user end, and uses the large model's strong semantic understanding and association analysis capabilities to achieve efficient interactive question-answering in the form of natural language or language dialogue. In step S50, the real-time conversation data of the user end is supplemented with information and key information is extracted from the conversation data and added to the vector database 4. Figure 6 The specific contents are as follows:

[0079] S501: Dynamically adjust the size of the user terminal context window;

[0080] S502: Using information extraction technology, extract key information from each round of conversation and perform vector processing;

[0081] S503: The key information after vectorization processing is added and stored in the vector database 4.

[0082] When processing multi-turn conversations, if the conversation exceeds the model's context limits, previous conversation content cannot be effectively utilized, impacting the user experience. To address this issue, two optimization solutions can be adopted. First, by dynamically adjusting the size of the model's context window, the model can flexibly adapt to conversation histories of varying lengths based on actual conversation needs, ensuring that relevant information is not lost. Second, information extraction techniques can be used to extract and vectorize the key information from each conversation turn. This allows the system to maintain conversation coherence in subsequent conversations, even if the original context limits are exceeded, by querying this vectorized key information. This improves the accuracy and relevance of responses, thereby enhancing the user experience. Combining these two approaches can enhance the system's flexibility and responsiveness while maintaining conversation quality.

[0083] This embodiment applies the semantic understanding and association analysis capabilities of the large model, and dynamically adjusts the size of the user-side context window, that is, optimizes multi-round dialogue context management, to achieve efficient interactive question-and-answer based on natural language or voice intercom, and can obtain relevant standardized work instructions for product surface inspection in real time, such as Figure 7 and Figure 8 As shown, it includes representative pictures of relevant defects, causes and treatment measures, as well as analysis reports of relevant historical similar defects, etc., to promote more efficient external inspection work.

[0084] Specifically, this embodiment summarizes and refines historical data scattered across human brains or computers. Through large-scale semantic understanding and vectorization, a preprocessing phase establishes a multimodal document parsing system, forming an external inspection knowledge base encompassing text and images to support standardized work instructions. By building a unified knowledge base management system, maintenance and management of relevant text and images is achieved, including debugging question-and-answer results and adding annotations. This enables knowledge management, such as information updates and authorized objects, as well as experience transfer.

[0085] Example 2

[0086] like Figure 9 and Figure 10 As shown, the external inspection AI co-pilot system based on the large model and the same inventive concept based on the construction method of the above-mentioned external inspection AI co-pilot system includes:

[0087] Enterprise historical experience data file 1; the historical experience data file includes at least one of image data, text data, voice data, video data, and report documents;

[0088] Document parsing and splitting module 2, which parses the historical experience data file and splits it into image data and / or text data and / or voice data and / or video data and / or report documents;

[0089] Data vectorization module 3, used for vectorizing the split and classified data;

[0090] Vector database 4, used to store data processed by data vectorization module 3;

[0091] The AI big model 5 serves as the client input port and matches the data information of the query problem with the vector database 4 through the vector index module 6 and the vector recall module 7;

[0092] The information supplement module 8 includes a dynamic adjustment unit 81 and a key information extraction unit 82; the dynamic adjustment unit 81 is used to dynamically adjust the size of the user terminal context window, and the key information extraction unit 82 is used to extract key information in the conversation.

[0093] This embodiment takes the historical experience data file 1 including text data, image data and report documents as an example. In this embodiment, the AI large model 5 adopts the qwen2.572B large model (Transformer architecture) as the core engine of the entire system. Users can directly obtain relevant text and image information in the knowledge base through natural language dialogue or voice intercom, realize job standardization guidance, and serve as an AI co-pilot to assist efficient operation. The data is converted into a vector model and stored in the vector database 4. The vector index module 6 and the vector recall module 7 are used to index and match the data information of the query question to realize language interaction and real-time question and answer. The actual question and answer effect is as follows: Figure 7 and Figure 8 As shown, this system can obtain relevant standardized operating instructions for product surface inspection in real time, including representative pictures of relevant defects, causes and treatment measures, as well as relevant historical analysis reports of similar defects and other textual information.

[0094] The information supplementation module 8 creates a system application for different domains by combining different categorized data documents for specialized personnel to query. Document content information is supplemented and uploaded, then encoded and stored in the database for data updates and processing to facilitate retrieval results. The information supplementation module 8 allows for both user questions and answers to be searched and edited, and the modified answers are then saved to the vector database 4. This not only creates data storage space for different domains, facilitating user queries, but also allows for real-time updates of the data within the vector database 4, ensuring accurate answers for user queries and large-scale model summaries.

[0095] This system also includes a quantitative annotation module 9 and a retrieval enhancement module. The quantitative annotation module 9 is used to provide quantitative annotations to the administrator background and / or users. The retrieval enhancement module is used to filter and bind the quantitative annotations of the administrator background and users' likes and dislikes to the correct answers, thereby enhancing the index matching degree.

[0096] The quantitative annotation module 9 enables the system to have an annotation reply function, and provides customized high-quality question-answer response capabilities for applications through manual editing. It can directly provide predefined answers without going through the generation step of a large language model (LLM), which not only improves the accuracy and reliability of the answers, but also effectively avoids the content hallucination problem that may occur in the retrieval-based generation (RAG) process. Specifically, the annotation reply function correctly binds the question and the answer and stores them in the vector library, increasing the accuracy of the retrieval, and increasing the accuracy of the search through the administrator's background standards or the quantitative annotation of the user's likes and dislikes, providing customized high-quality question-answering capabilities for the application. The processing steps to avoid the content hallucination problem are as follows:

[0097] A10: Standardize the formats of text and image data and add special marks and symbols;

[0098] A20: Intelligently slice text data into segments based on their length and adjust the length to optimize storage.

[0099] A30: Binding of vector-stored retrieval filtering and annotated correct answers to eliminate hallucination questions.

[0100] By setting up the quantitative annotation module 9 to annotate and reply to the image data, the stability and credibility of the system are enhanced, ensuring that users can obtain more accurate and consistent information services, such as Figure 8 The image data is shown with annotations.

[0101] The setting of the retrieval enhancement module forms a closed loop of question-answer retrieval, answer generation, and feedback, which specifically includes the following processing steps: data organization and cleaning and structuring into a standard unified format, intelligent slicing to separate long documents into semantic blocks according to length to improve retrieval granularity, and data vector storage, combined with word segmentation index storage in the vector database 4.

[0102] In order to better present the usage and key data of the system, this system also includes a data detection and analysis module, which is used to analyze the usage data and calculate key indicators for display; key indicators include at least one of the number of users, number of active users, large language model call consumption, usage type trends, total number of messages, average number of appropriate interactions, and output speed.

[0103] The data detection and analysis module displays key indicators such as user volume, number of active users, and large language model (LLM) call consumption. These data provide a solid foundation for continuous optimization of applications, user activity, and economic efficiency. By monitoring user volume and number of active users, we can gain a deeper understanding of application usage trends and user behavior patterns, thereby formulating more precise operational strategies to improve user engagement. At the same time, analyzing AI large model 5 call consumption helps evaluate cost-effectiveness, optimize resource allocation, and improve the economic efficiency of services. Combining these insights promotes continuous optimization of applications, ensuring that they meet user needs while achieving more efficient operational management. Figure 10 This is a graph showing data in actual use.

[0104] The present invention divides the text data and image data in the enterprise historical experience data file 1 into fragments and modularizes them through the document parsing and splitting module 2 to form an external inspection position knowledge base, and vectorizes and stores the data in the external inspection position knowledge base through the data vectorization module 3, and stores them in the vector database 4. It uses semantic vectorization technology to improve retrieval efficiency and accuracy, making content-based retrieval more accurate and effective, and realizes language interaction between the vector database 4 and the AI large model 5 by establishing a vector index module 6 and a vector recall module 7, and provides standardized operation guidance for on-site appearance inspection positions.

[0105] At the same time, the quantitative annotation module 9 annotates and responds to the image data, enhancing the stability and credibility of the system. The retrieval enhancement module forms a closed loop between question-and-answer retrieval, answer generation, and feedback, improving retrieval accuracy and improving the efficiency of defect recognition and handling in appearance inspection. The data detection and analysis module monitors key indicators such as user volume, number of active users, and large language model (LLM) call consumption to ensure that the system achieves more efficient operation and management while meeting user needs. It should be understood that this system can not only be directly applied in external inspections, but can also provide training and development for new employees, with pictures and texts, natural language interaction, and is more intuitive and efficient.

[0106] Example 3

[0107] The construction method described in Example 1 and the system described in Example 2 are compiled into corresponding computer codes and written into a storage medium. When the storage medium is connected to a controller, the computer program codes therein can be read out and executed, thereby automatically executing steps S10-S50 and achieving the same technical effects as in Example 1 or Example 2.

[0108] It should be appreciated that embodiments of the present invention may be implemented or practiced by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable memory. The methods may be implemented in a computer program using standard programming techniques - including a non-transitory computer-readable storage medium configured with a computer program, wherein the storage medium so configured causes the computer to operate in a specific and predefined manner - according to the methods and drawings described in the specific embodiments. Each program may be implemented in a high-level procedural or target terminal-oriented programming language to communicate with the computer system. However, if desired, the program may be implemented in assembly or machine language. In any case, the language may be a compiled or interpreted language. In addition, the program may be run on a programmed application-specific integrated circuit for this purpose.

[0109] It should be understood that the specific embodiments described above are only used to explain the present invention and are not intended to limit the present invention. Obvious changes or modifications derived from the spirit of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for constructing an external inspection AI co-pilot system based on a large model, characterized by: Includes the following: Obtain the company's historical experience data and perform segmented and modular processing on the acquired historical experience data to form a knowledge base for external inspection positions; The data in the external inspection position knowledge base is vectorized through vector model encoding and stored in the vector database; Establishing a vector index model and a vector recall model, i.e., labeling and editing the data in the vector database, and using the data stored in the vector database to label and recall questions and answers; The vector database and the big model establish a language interaction relationship through the vector index model and the vector recall model; the user terminal asks questions through the big model to obtain the content of the vector database and complete real-time question and answer; The real-time conversation data of the user end is supplemented with information and key information is extracted from the conversation data and supplemented into the vector database.

2. The method for constructing a large-scale model-based product inspection AI co-pilot system according to claim 1, characterized in that: The acquired historical experience data includes one or more of picture data, text data, voice data, video data and report documents; The acquired historical experience data is processed in a fragmented and modular manner, including the following contents: Separate historical experience data according to data type; Organize and clean the separated image data and / or text data and / or voice data and / or video data and / or report documents, and process them into a unified standard format; The data in the standard format is encoded and segmented according to the data length, and then divided into several small blocks according to semantics.

3. The method for constructing a large-scale model-based product inspection AI co-pilot system according to claim 2, characterized in that: The vectorization processing of the data in the external inspection post knowledge base through vector model encoding includes the following contents: Using natural language processing technology to parse the text information divided into several small blocks according to semantics, and establishing a vector unit for natural language understanding and reasoning through a model encoder; Computer vision algorithms are used to perform defect recognition and defect labeling on the image information that is semantically separated into several small blocks, and the generated label information is converted into a vector form.

4. The method for constructing a large-scale model-based product inspection AI co-pilot system according to claim 3, characterized in that: The defect marking includes any one of defect frame marking and defect arrow marking.

5. The method for constructing a large-scale model-based product inspection AI co-pilot system according to claim 1, characterized in that: The real-time conversation data of the user terminal is supplemented with information and key information of the conversation data is extracted and added to the vector database. Includes the following: Dynamically adjust the size of the user-side context window; Using information extraction technology, key information in each round of conversation is extracted and vectorized; The key information after vectorization processing is supplemented and stored in the vector database.

6. The method for constructing a large-scale model-based product inspection AI co-pilot system according to claim 5, characterized in that: The answer data includes one or more of defect type data, cause data, solution data, and example pictures.

7. A large-scale model-based product inspection AI co-pilot system, characterized by: include: Enterprise historical experience data file; the historical experience data file includes at least one of image data, text data, voice data, video data and report documents; A document parsing and splitting module performs document parsing on the historical experience data file and splits it into image data and / or text data and / or voice data and / or video data and / or report documents; Data vectorization module, used to perform vectorization processing on split and classified data; A vector database, used to store the data processed by the data vectorization module; The AI big model serves as the client input port and matches the data information of the query problem with the vector database through the vector index module and the vector recall module; The information supplement module includes a dynamic adjustment unit and a key information extraction unit; the dynamic adjustment unit is used to dynamically adjust the size of the user terminal context window, and the key information extraction unit is used to extract key information in the conversation.

8. The AI co-pilot system for product inspection based on a large model according to claim 7, characterized in that: It also includes a quantitative annotation module and a retrieval enhancement module. The vectorized annotation module is used to provide quantitative annotation to the administrator background and / or users, and the retrieval enhancement module is used to filter and bind the annotated correct answers according to the administrator background standards and / or the quantitative annotations of users' likes and dislikes, so as to enhance the index matching degree.

9. The AI co-pilot system for product inspection based on a large model according to claim 7, characterized in that: It also includes a data detection and analysis module for analyzing usage data and calculating key indicators for display; the key indicators include at least one of the number of users, number of active users, large language model call consumption, usage type trend, total number of messages, average number of appropriate interactions, and output speed.

10. A storage medium storing instructions executable by a processor, characterized in that: The processor-executable instructions, when executed by the processor, are used to execute a method for constructing a large-model-based product external inspection AI co-pilot system as described in any one of claims 1-7.

Images