Intelligent knowledge question-answering method and system for aviation part production process

Abstract

The invention belongs to the technical field of aviation intelligent manufacturing, and discloses an aviation part production process-oriented knowledge intelligent question and answer method and system. Aiming at complex structure knowledge including form data in a business platform, long text data in a standard file, short text data of a quality fault record and the like in the aviation part production process, technologies including multivariate knowledge data preprocessing, two-stage vector coding, two-stage vector indexing, enhanced guidance and the like are adopted; and professional questions proposed by the user can be quickly and accurately responded. According to the method, the defects of long time consumption, high labor cost, difficulty in quickly dealing with technical challenges and the like in retrieving complex aviation manufacturing professional knowledge can be solved, so that the scientific research and production efficiency of aviation manufacturing is improved.

Description

Technical Field

[0001] The present invention application belongs to the field of aviation intelligent manufacturing technology, and specifically relates to a knowledge intelligent question-answering method and system for the production process of aviation parts. Background Art

[0002] The rapid development of the aviation manufacturing industry has led to an increase in the complexity and breadth of aviation business knowledge. The acquisition and application of knowledge is crucial to ensuring the quality and safety of aviation products and improving the efficiency of scientific research and production. The existing knowledge acquisition method of the aviation parts production process mainly relies on manual review of documents, manuals or consultation with experts, which is time-consuming and labor-intensive, and it is difficult to respond to complex production and R&D problems in real time.

[0003] Large language model technology has demonstrated powerful processing capabilities in the field of natural language processing, bringing new possibilities for the realization of intelligent question-answering systems. However, existing question-answering systems are mostly used in general fields, and there are still many technical challenges for professional knowledge question-answering systems in the aviation manufacturing field. For example, how to deeply combine large language models with complex and huge aviation manufacturing professional knowledge to achieve accurate and efficient intelligent question-answering and provide professional knowledge services is still a key problem that needs to be solved.

[0004] Therefore, it is desired to have a technical solution to overcome or at least alleviate at least one of the above-mentioned defects of the prior art. Summary of the invention

[0005] The present invention application proposes a knowledge intelligent question-answering method and system for the aviation parts production process to solve the shortcomings of the prior art in that aviation business knowledge retrieval is time-consuming, labor-intensive, and difficult to respond to complex problems in real time.

[0006] In order to achieve the above invention objectives, the technical solutions adopted in this application are as follows:

[0007] A knowledge intelligent question answering method for aviation parts production process, comprising the following steps:

[0008] The multivariate knowledge data of the aviation parts production process is processed, and a pre-trained text embedding model is used for two-level vector encoding to construct a multivariate knowledge structure database of the aviation parts production process.

[0009] Vector encoding is performed on the user's question, and vector retrieval is performed in the multivariate knowledge structure database through a two-level index to obtain the similarity between the relevant knowledge items and the corresponding text;

[0010] Based on the text similarity, adding corresponding attention prompt words in the relevant knowledge items;

[0011] The user question, relevant knowledge items, attention prompt words and scenario application prompt words are input into the large language model to obtain a data-enhanced structured output.

[0012] As a further solution of the present invention: the multi-dimensional knowledge data of the aviation parts production process includes form data in the business platform, long text data in the standard file and short text data of quality failure records.

[0013] As a further solution of the present invention: the method for constructing the multivariate knowledge structure database specifically comprises the following steps:

[0014] The content processing includes dynamically and real-time extracting the form data in the platform, extracting knowledge sections from long text data, and merging short text data;

[0015] Using the pre-trained text embedding model to perform primary keyword encoding and secondary paragraph encoding on the multivariate knowledge data, and obtaining a vector set through normalization processing;

[0016] The processing formula for vector normalization is:

[0017]

[0018] Where v is the original vector. ||v|| is the L2 norm of vector v. For a given vector v = (v1, v2, ..., v n ), and its L2 norm calculation formula is: Therefore, the normalized vector v norm The calculation formula for each component of is:

[0019]

[0020] A multivariate knowledge structure database of the aviation parts production process is generated based on the multivariate knowledge data and the vector set.

[0021] As a further solution of the present invention: the two-level index includes a primary keyword index and a secondary paragraph index, and the vector retrieval method is an approximate nearest neighbor search, which is used to obtain the similarity between relevant knowledge items and corresponding texts, thereby achieving efficient and accurate capture of aviation parts production process knowledge from coarse to fine and hierarchical and step-by-step.

[0022] The processing method of the approximate nearest neighbor search is:

[0023] sim(q,v j )=q·v j ;

[0024] Where q is the query vector, sim(q, v j) is the query vector q and each vector v in the set j The similarity is calculated using the inner product. Select the k vectors that are most similar to the query vector q from the n vectors in the set, that is, select the set of k vectors with the greatest similarity.

[0025] As a further solution of the present invention: the attention prompt words are a group of guide words or phrases predefined according to the text similarity, which guide the large language model to give priority to important content with higher relevance.

[0026] As a further solution of the present invention: the scenario application prompt words are output structure guide words set according to the usage requirements of different business scenarios, ensuring that the system can provide appropriate output formats according to the scenario application, thereby improving the flexibility and practicality of the system.

[0027] As a further solution of the present invention: the method further comprises:

[0028] The output natural language special characters are displayed or processed in a specific format to form structured natural language text to ensure the accuracy and neatness of the content;

[0029] Through the interface service framework, the generated results are encapsulated as API services and provided to the outside world for access, ensuring that other systems can call and integrate through standardized interfaces.

[0030] A knowledge intelligent question-answering system for the production process of aviation parts, comprising:

[0031] Multivariate knowledge data preprocessing module: used for knowledge preprocessing of different types of data, including dynamic and real-time extraction of form data in the platform, knowledge segmentation extraction of long text data, and merging of short text data;

[0032] Two-level vector encoding module: used to perform vector encoding on the natural language text obtained by preprocessing multivariate knowledge data, including first-level keyword encoding and second-level paragraph encoding;

[0033] Two-level vector indexing module: used to vectorize user questions and perform two-level indexing in the multi-knowledge database, including a primary keyword index and a secondary paragraph index, to obtain knowledge items related to the questions and to retrieve large-scale knowledge efficiently and accurately;

[0034] Enhanced guidance module: used to input user questions, relevant knowledge items, scenario application prompt words and attention prompt words into the large language model and generate scenario application knowledge;

[0035] Structured output module: displays or processes the output natural language special characters in a specific format to form structured natural language text, which is convenient for subsequent data processing, storage and retrieval;

[0036] The interface publishing module is used to publish the knowledge question and answer service through a standardized API interface, so that other systems or users can call the system service to obtain aviation business knowledge question and answer results.

[0037] Compared with the prior art, the beneficial effects of this application are:

[0038] 1. In view of the complex structured knowledge such as form data in the business platform, long text data in standard files, and short text data in quality fault records contained in the aviation parts production process, we use multivariate knowledge data preprocessing, two-level vector encoding, two-level vector indexing, enhanced guidance and other technologies to quickly and accurately respond to professional questions raised by users and significantly improve the knowledge acquisition efficiency of the aviation parts production process.

[0039] 2. This method reduces the reliance on manual document review and expert consultation through intelligent means, thereby reducing labor costs and error rates.

[0040] 3. The system can continuously learn and optimize, adapt to various complex business scenarios in the field of aviation manufacturing, help improve aviation research and production efficiency, and meet the requirements of high-efficiency, high-quality, and low-cost processing of business knowledge.

[0041] The present application is further described below in conjunction with the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0043] In the attached picture:

[0044] Figure 1 A flow chart of a knowledge intelligent question-answering method for an aviation parts production process according to an embodiment of the present application;

[0045] Figure 2 A schematic diagram of a preprocessing method for constructing a multivariate knowledge structure database of an aviation parts production process according to an embodiment of the present application;

[0046] Figure 3 This is a schematic diagram of a knowledge intelligent question-answering system for aviation parts production process according to one embodiment of the present application.

[0047] The reference numerals in the figure are: 100 - multivariate knowledge data preprocessing module; 200 - two-level vector encoding module; 300 - two-level vector indexing module; 400 - enhanced guidance module, 500 - structured output module, 600 - interface publishing module. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0049] The following is combined with Figures 1 to 3 This application is described in further detail.

[0050] The first aspect of the present application provides a knowledge intelligent question answering method for the production process of aviation parts, such as Figure 1 As shown, the following steps are included:

[0051] Step S1, performing content processing on the multivariate knowledge data of the aviation parts production process, and using a pre-trained text embedding model to perform two-level vector encoding to construct a multivariate knowledge structure database of the aviation parts production process;

[0052] Step S2: vector encoding is performed for the user question, and vector retrieval is performed in the multivariate knowledge structure database through a two-level index to obtain the similarity between the relevant knowledge item and the corresponding text;

[0053] Step S3: adding corresponding attention prompt words to the relevant knowledge items based on the text similarity;

[0054] Step S4: Input the user question, relevant knowledge items, attention prompt words and scenario application prompt words into the large language model to obtain a data-enhanced structured output. The large language model can adopt the Transformer large language model architecture.

[0055] Step S5: Display or process the output natural language special characters (such as quotation marks, brackets, slashes, separators, etc.) in a specific format to form a structured natural language text to ensure the accuracy and neatness of the content.

[0056] Step S6: Through the interface service framework, the generated results are encapsulated as API services and provided to the outside world for access, ensuring that other systems can call and integrate through standardized interfaces.

[0057] In the knowledge intelligent question-answering method for aviation parts production process of the present application, in step S1, the multi-dimensional knowledge data of the aviation parts production process includes form data in the business platform, long text data in the standard file and short text data of quality fault records.

[0058] In at least one embodiment of the present application, step S1 further includes:

[0059] Step S11, the content processing includes dynamically extracting the form data in real time on the platform, extracting knowledge sections of the long text data, and merging the short text data; in this embodiment, Figure 2 The figure shows a schematic diagram of the method for preprocessing multivariate knowledge data in the aviation production process. The form data in the business platform is dynamically extracted in real time, including the header and table content, the long text data in the standard file is processed and summarized, and the short text data of each professional quality failure record is merged.

[0060] Step S12: Use the pre-trained text embedding model to perform primary keyword encoding and secondary paragraph encoding on the multivariate knowledge data. Keywords can be professional vocabulary, technical vocabulary, operation verbs, business process vocabulary, standards and specifications, performance indicator vocabulary, etc.; paragraphs can be content summaries and abstracts, background information, causes and results, detailed descriptions, case analysis and application scenarios, theoretical explanations and concept definitions, references and citations, decisions and suggestions, etc. Obtain a vector set through normalization processing;

[0061] The processing method of the vector normalization is:

[0062]

[0063] Where v is the original vector. ||v|| is the L2 norm of vector v. For a given vector v = (v1, v2, ..., v n ), whose L2 norm is calculated as Therefore, the normalized vector v norm Each component of is calculated as:

[0064]

[0065] Step S13: Generate a multivariate knowledge structure database of the aviation parts production process based on the multivariate knowledge data and the vector set.

[0066] In the knowledge intelligent question-answering method for aviation parts production process of the present application, in step S2, the two-level index includes a primary keyword index and a secondary paragraph index, and the vector retrieval method is an approximate nearest neighbor search, which is used to obtain the similarity between relevant knowledge items and corresponding texts, so as to realize efficient and accurate capture of aviation parts production process knowledge from coarse to fine and hierarchical and step-by-step.

[0067] The processing method of the approximate nearest neighbor search is:

[0068] sim(q,v j )=q·v j ;

[0069] Where q is the query vector, sim(q, v j ) is the query vector q and each vector v in the set j The similarity is calculated using the inner product. Select the k vectors that are most similar to the query vector q from the n vectors in the set, that is, select the set of k vectors with the greatest similarity.

[0070] In the knowledge intelligent question-answering method for aviation parts production process of the present application, in step S3, the attention prompt words are a group of guide words or phrases predefined according to the size of text similarity, which guide the large language model to give priority to important content with higher relevance.

[0071] In the knowledge intelligent question-answering method for aviation parts production process of the present application, in step S4, the scenario application prompt words are output structure guide words set according to the usage requirements of different business scenarios, ensuring that the system can provide appropriate output formats according to the scenario application, thereby improving the flexibility and practicality of the system.

[0072] The knowledge intelligent question-answering method for the aviation parts production process of this application adopts multivariate knowledge data preprocessing, two-level vector encoding, two-level vector indexing, enhanced guidance and other technologies to quickly and accurately respond to professional questions raised by users, and significantly improve the knowledge acquisition efficiency of the aviation parts production process, targeting complex structured knowledge such as form data in the business platform, long text data in standard files, and short text data of quality fault records contained in the aviation parts production process. This method reduces the reliance on manual document review and expert consultation through intelligent means, reduces labor costs and error rates. At the same time, the system can continuously learn and optimize, adapt to a variety of complex business scenarios in the field of aviation manufacturing, help improve aviation research and production efficiency, and meet the requirements of high-efficiency, high-quality, and low-cost processing of business knowledge.

[0073] Based on the above-mentioned knowledge intelligent question answering method for the aviation parts production process, the second aspect of the present application provides a knowledge intelligent question answering system for the aviation parts production process, such as Figure 3 As shown, including:

[0074] Multivariate knowledge data preprocessing module 100: used for knowledge preprocessing of different types of data, including dynamic and real-time extraction of the form data in the platform, knowledge segmentation extraction of long text data, and merging processing of short text data;

[0075] Two-level vector coding module 200: used to perform vector coding on the natural language text obtained by preprocessing the multivariate knowledge data, including first-level keyword coding and second-level paragraph coding;

[0076] Two-level vector indexing module 300: used to vectorize user questions and perform two-level indexing in a multi-knowledge database, including a primary keyword index and a secondary paragraph index, to obtain knowledge items related to the questions and to retrieve large-scale knowledge efficiently and accurately;

[0077] Enhanced guidance module 400: used to input user questions, relevant knowledge items, scenario application prompt words and attention prompt words into the large language model and generate scenario application knowledge;

[0078] Structured output module 500: displays or processes the output natural language special characters in a specific format to form structured natural language text, which is convenient for subsequent data processing, storage and retrieval;

[0079] The interface publishing module 600 is used to publish the knowledge question and answer service externally through a standardized API interface, so that other systems or users can call the system service to obtain aviation business knowledge question and answer results.

[0080] So far, the purpose of the present invention has been achieved.

[0081] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A knowledge intelligent question answering method for the production process of aviation parts, characterized in that: The steps include: The multivariate knowledge data of the aviation parts production process is processed, and a pre-trained text embedding model is used for two-level vector encoding to construct a multivariate knowledge structure database of the aviation parts production process. Vector encoding is performed on the user's question, and vector retrieval is performed in the multivariate knowledge structure database through a two-level index to obtain the similarity between the relevant knowledge items and the corresponding text; Based on the text similarity, adding corresponding attention prompt words in the relevant knowledge items; The user question, relevant knowledge items, attention prompt words and scenario application prompt words are input into the large language model to obtain a data-enhanced structured output.

2. The knowledge intelligent question-answering method for the aviation parts production process according to claim 1 is characterized in that: The multi-dimensional knowledge data of the aviation parts production process includes form data in the business platform, long text data in the standard file, and short text data of quality failure records.

3. The knowledge intelligent question-answering method for the aviation parts production process according to claim 2 is characterized in that: The method for constructing the multivariate knowledge structure database specifically comprises the following steps: The content processing includes dynamically and real-time extracting the form data in the platform, extracting knowledge sections from long text data, and merging short text data; Using the pre-trained text embedding model to perform primary keyword encoding and secondary paragraph encoding on the multivariate knowledge data to obtain a vector set; A multivariate knowledge structure database of the aviation parts production process is generated based on the multivariate knowledge data and the vector set.

4. The knowledge intelligent question-answering method for aviation parts production process according to claim 1 is characterized in that: The two-level index includes a primary keyword index and a secondary paragraph index. The vector retrieval method is an approximate nearest neighbor search, which is used to obtain the similarity between relevant knowledge items and corresponding texts, and realize efficient and accurate capture of aviation parts production process knowledge from coarse to fine and hierarchical and step-by-step.

5. The knowledge intelligent question-answering method for aviation parts production process according to claim 1 is characterized in that: The attention prompt words are a group of guide words or phrases predefined according to the text similarity, which guide the large language model to give priority to important content with high relevance.

6. The knowledge intelligent question-answering method for aviation parts production process according to claim 1 is characterized in that: The scenario application prompt words are output structure guide words set according to the usage requirements of different business scenarios, ensuring that the system can provide appropriate output formats according to the scenario application, thereby improving the flexibility and practicality of the system.

7. The knowledge intelligent question answering method for aviation parts production process according to any one of claims 1 to 6 is characterized in that: The method further comprises: The output natural language special characters are displayed or processed in a specific format to form structured natural language text to ensure the accuracy and neatness of the content; Through the interface service framework, the generated results are encapsulated as API services and provided to the outside world for access, ensuring that other systems can call and integrate through standardized interfaces.

8. A knowledge intelligent question-answering system for the production process of aviation parts, characterized in that: Implementing the question-answering method as claimed in claim 7, comprising: Multivariate knowledge data preprocessing module: used for knowledge preprocessing of different types of data, including dynamic and real-time extraction of form data in the platform, knowledge segmentation extraction of long text data, and merging of short text data; Two-level vector encoding module: used to perform vector encoding on the natural language text obtained by preprocessing multivariate knowledge data, including first-level keyword encoding and second-level paragraph encoding; Two-level vector indexing module: used to vectorize user questions and perform two-level indexing in the multi-knowledge database, including a primary keyword index and a secondary paragraph index, to obtain knowledge items related to the questions and to retrieve large-scale knowledge efficiently and accurately; Enhanced guidance module: used to input user questions, relevant knowledge items, scenario application prompt words and attention prompt words into the large language model and generate scenario application knowledge; Structured output module: displays or processes the output natural language special characters in a specific format to form structured natural language text, which is convenient for subsequent data processing, storage and retrieval; The interface publishing module is used to publish the knowledge question and answer service through a standardized API interface, so that other systems or users can call the system service to obtain aviation business knowledge question and answer results.

Images