FastGPT-based intelligent question-answering system for construction project quality inspection and detection standards

Abstract

The invention relates to the technical field of data retrieval, in particular to a FastGPT-based intelligent question-answering system for construction project quality inspection and detection standards. According to the technical scheme, the method comprises the steps of external knowledge acquisition, document processing, vectorization, vector database, retrieval and context generation and intelligent question and answer, data can be collected by a system, a semantic segmentation technology, a QA pair generation technology and a high-dimensional vectorization technology are utilized, in combination with M3E and OpenAIEmbedding models, Top-K similarity retrieval and context dynamic recombination are achieved, and structured and hierarchical professional answers are generated. According to the method, the intelligent management and real-time application capability of the construction project quality inspection and detection standard can be remarkably improved, the manual retrieval and interpretation cost is reduced, and the digital and intelligent requirements of an engineering project are met.

Description

Technical Field

[0001] This invention relates to the field of information retrieval technology, and in particular to an intelligent question-and-answer system for construction engineering quality inspection and testing standards based on FastGPT. Background Technology

[0002] Construction project quality inspection and testing standards cover multiple dimensions, including structural safety, material performance, construction techniques, environmental impact, and post-construction maintenance. These standards are numerous, frequently updated, and originate from national and industry standards, local regulations, and internal corporate management systems. Existing construction projects often face the following problems when implementing quality inspection and testing: Information is scattered and lacks unified management: Standard provisions are distributed across multiple platforms or documents with inconsistent formats. Manual retrieval is time-consuming and labor-intensive, and omissions or version errors are prone to occur, resulting in untimely and inaccurate application of standards. High cost of manual interpretation: Construction engineering standards usually contain a large number of professional terms, numerical indicators and cross-referenced clauses. Engineering technicians find it difficult to fully and accurately understand the relevant clauses in a short period of time, which increases compliance risks. Insufficient intelligence: Most existing information systems only support keyword retrieval and cannot perform semantic understanding and intelligent question answering. The search results have low relevance to user questions and require manual secondary screening and sorting, which is inefficient. Lack of multi-source knowledge fusion capability: Existing systems have difficulty integrating national standards, industry norms, internal enterprise testing data and historical cases at the same time, resulting in fragmented knowledge bases and an inability to form a unified standard interpretation and application support platform; To address this issue, we propose an intelligent question-and-answer system based on FastGPT for construction engineering quality inspection and testing standards. Summary of the Invention

[0003] The purpose of this invention is to address the problems existing in the background technology by proposing an intelligent question-and-answer system for construction engineering quality inspection and testing standards based on FastGPT.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an intelligent question-and-answer system for construction engineering quality inspection and testing standards based on FastGPT, comprising an external knowledge acquisition module, a document processing module, a vectorization module, a vector database, a user interaction module, a retrieval and context generation module, and an intelligent question-and-answer module. The external knowledge acquisition module is used to automatically and manually access technical documents related to standards, specifications, and testing reports in the field of construction engineering, and supports the import of the following data formats, including PDF, Word, Excel, and web page data, to ensure the comprehensiveness and authority of the knowledge source. The document processing module is used to preprocess the external knowledge. The preprocessing includes removing redundant information, text cleaning, semantic segmentation, and dividing the document content into multiple logically complete small text units (Chunks) to facilitate subsequent vectorization processing. The vectorization module is used to perform high-dimensional semantic vectorization processing on the text unit, extract semantic features using a deep language model, and generate corresponding vector representations. The vector database is used to store the vectorized text units, supports fast retrieval based on vector similarity, and dynamically returns the most relevant context information for query requests. The user interaction module is used to receive natural language query questions input by users and convert the query questions into semantic vectors to ensure matching with the vector database. The retrieval and context generation module is used to perform similarity retrieval in the vector database based on the query vector, call the text units related to the query, and dynamically reorganize their context information to form logically consistent context content. The intelligent question-answering module is used to combine the user query with the context information to form a unified prompt, and input the large language model (LLM) fine-tuned with data from the construction engineering field to automatically output a professional answer that conforms to the construction engineering quality inspection and testing standards.

[0005] Furthermore, the document processing module further includes: The QA splitting unit automatically generates question-answer pairs based on the document's semantic structure, enhancing the targeting of subsequent searches; Semantic segmentation units are used to segment text based on the document's logical structure and semantic features, ensuring the semantic integrity of the segmented text units. The manual correction unit is used to manually correct and supplement the results of automatic processing to ensure the accuracy and completeness of professional knowledge.

[0006] Furthermore, the vectorization module supports multi-model collaboration, including at least one of the M3E vector model and the OpenAIembedding model. The system can automatically select and weightedly fuse the outputs of different vector models based on the characteristics of the retrieval task to improve the accuracy and recall of the retrieval.

[0007] Furthermore, the intelligent question-answering module includes: The model fine-tuning unit is used to perform transfer learning on a general language model based on labeled data in the construction engineering field, enabling it to recognize and correctly understand professional terms and standard clauses in the construction engineering field. The multi-task learning unit, based on the fine-tuned model, jointly performs multi-dimensional task training such as clause classification, compliance judgment, and standard clause correlation analysis to improve the model's comprehensive reasoning and question-answering capabilities in complex scenarios. The continuous learning unit is used to automatically collect and update knowledge data based on newly released industry standards, specifications or regulations, while performing incremental training of model parameters to ensure that the knowledge base and language model of the question answering system are always up-to-date.

[0008] Furthermore, the vector database is built on PostgreSQL, supports a hybrid mode of vector similarity retrieval and keyword retrieval, and can combine historical query records for contextual joint retrieval to improve the contextual continuity and professional completeness of the answers.

[0009] Furthermore, the retrieval and context generation module uses FlowController to dynamically orchestrate retrieval strategies. The orchestration includes automatic adjustment of the retrieval scope, relevance ranking of candidate contexts, removal of invalid contexts, and final context concatenation, in order to improve question-answering accuracy and professional reliability while ensuring reasoning efficiency.

[0010] Furthermore, the intelligent question-answering module can support multilingual formatting, hierarchical summary extraction, and structured result display when outputting answers, in order to meet the intelligent question-answering needs in cross-language and multi-professional scenarios.

[0011] Furthermore, the question-answering method of the system includes the following steps: Step 1: Obtain relevant external knowledge data such as construction engineering standards, specifications, and test reports, and store them in a formatted manner; Step 2: Semantically segment the external knowledge to form multiple logically independent text units; Step 3: Vectorize the text units and store them in a vector database that supports similarity retrieval; Step 4: Receive the natural language query input by the user and vectorize it into a query vector; Step 5: Retrieve the text unit that best matches the query vector from the vector database and reorganize it into context information; Step 6: Concatenate the user query with the context information to generate a prompt, and input it into the finely tuned large language model; Step 7: Output structured answers related to construction project quality inspection and testing standards.

[0012] Furthermore, the system further includes a security and access control module, which implements multi-level security strategies for the access to external knowledge, the transmission of query data, and the output of answers, including data encryption, access permission classification, and operation log recording, to ensure the data security and compliance of the system in actual engineering projects.

[0013] Furthermore, the system further includes a multi-source knowledge fusion module, which is used to fuse multi-source data from government standard databases, third-party testing agency documents, and enterprise internal documents to solve the heterogeneity and redundancy problems of cross-source standards and improve the universality and completeness of the knowledge base in multiple scenarios.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: Multi-source knowledge fusion and unified management: Through the external knowledge acquisition module, the system can automatically collect national standards, industry specifications, internal enterprise documents and third-party testing data, and realize multi-source knowledge fusion based on the information entropy weight allocation algorithm to form a unified construction engineering quality standard knowledge base, which significantly improves the centralized management and version update efficiency of standard information; Intelligent semantic segmentation and QA pair generation: The system introduces natural language processing technology to automatically perform semantic segmentation and question-answer pair generation on standard documents. Combined with deep learning models, it realizes intelligent structured processing of text, reduces the cost of manual interpretation, and ensures the accuracy and timeliness of standard information retrieval. Efficient vectorized retrieval and context generation: The vectorization technology, which integrates the M3E model and the OpenAIembedding model, transforms text into high-dimensional semantic vectors. Through Top-K similarity retrieval, hybrid retrieval, and context splicing strategies, it generates answers that are highly relevant to the user's query, significantly improving retrieval accuracy and intelligent question answering experience. Intelligent question answering and multi-task processing capabilities: Based on a finely tuned large language model, the system supports various tasks such as standard clause interpretation, compliance judgment, and historical case correlation analysis. It can output structured answers, hierarchical summaries, and multilingual results to meet diverse needs such as engineering management, compliance review, and technical decision-making.

[0015] In summary, this invention not only solves the technical bottlenecks in the management, retrieval, and intelligent application of construction engineering quality inspection and testing standards, but also achieves efficient management and intelligent services of standard information through multi-source data fusion, intelligent semantic processing, and security control, significantly improving the digitalization and intelligence level of engineering quality management. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the system flow of the present invention; Figure 2 This is a schematic diagram of the RAG process of the present invention; Figure 3 This is a schematic diagram of the knowledge base process of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1 like Figure 1-3 As shown, the present invention proposes an intelligent question-and-answer system for construction engineering quality inspection and testing standards based on FastGPT, including an external knowledge acquisition module, a document processing module, a vectorization module, a vector database, a user interaction module, a retrieval and context generation module, an intelligent question-and-answer module, a security and access control module, and a multi-source knowledge fusion module. The modules interact and collaborate with each other through an internal bus and API interface, enabling efficient management, intelligent retrieval, and professional answers of multi-source knowledge in the field of construction engineering quality inspection and testing standards.

[0019] The system adopts a layered architecture design. The bottom layer is responsible for data collection and storage, the middle layer performs semantic processing and vectorized modeling, and the upper layer completes intelligent question answering and secure access control, ensuring the scalability and stability of the system in complex engineering scenarios.

[0020] In this embodiment, the core function of the external knowledge acquisition module is to unify and access quality standard documents and test reports scattered across different industry organizations, government departments, and enterprises. The data access method is as follows: Automatic access: The system has a pre-set scheduled task that periodically accesses government standards databases and third-party testing agency data platforms through API interfaces to automatically download the latest quality standards and testing specifications; Manual Access: For non-public information within the enterprise, the system provides a visual upload interface, allowing project engineers to directly import PDF, Word, Excel, and HTML web page files into the system.

[0021] Access the following three types of data sources: Government Standards Database: Includes national and industry standards such as the "Unified Standard for Acceptance of Construction Quality of Building Engineering" (GB50300) and the "Code for Acceptance of Construction Quality of Concrete Structures" (GB50204) issued by the Ministry of Housing and Urban-Rural Development; Third-party testing organization documents: including the testing organization's test reports and testing method specifications; Internal company documents: including the construction company's quality management manual and internal process standards.

[0022] To enhance the intelligence of data access, the system introduces a dynamic sampling function during automatic data collection. The dynamic sampling function is as follows: in, Indicates the collection time. , , , These are the sampling attenuation coefficient, exponential attenuation rate, amplitude, and angular frequency, respectively. Intelligent balanced acquisition of data from both old and new standards is achieved by dynamically adjusting the sampling frequency.

[0023] The document processing module includes a QA splitting unit, a semantic segmentation unit, and a manual correction unit: QA splitting unit: Using a BERT-based question-answering generation model, clause-type sentences are automatically converted into "question-answer" pairs, reducing the secondary parsing cost during subsequent question-answering reasoning. For example, "reinforcing steel cover thickness requirement" is split into Q1: "What is the minimum thickness requirement for reinforcing steel cover?" and A1: "According to GB50204-2015, the minimum thickness of the concrete cover for stressed reinforcing steel is 25mm." Semantic segmentation unit: The system first identifies structured information such as titles, clause numbers, and technical parameter tables in the document through natural language word segmentation and syntactic analysis. Then, it uses semantic similarity calculation to aggregate adjacent sentence segments to ensure semantic coherence within each segment and relative semantic independence between segments. To quantify the rationality of semantic segmentation, a segmentation optimization objective function is introduced. The above segmentation optimization objective function is as follows: in, The total number of segments. Indicates the semantic angle between adjacent segments. Calculated from text vectors, The smaller the value of this function, the better the semantic continuity between the segments.

[0024] Manual correction unit: Domain experts manually review the automatically generated QA pairs and segmented results to ensure the accuracy of technical terms, parameter values, and standard clause references.

[0025] The vectorization module is based on the collaborative mechanism of the M3E vector model and the OpenAIembedding model, which maps each text unit to a high-dimensional vector representation.

[0026] During vector generation, a weighted fusion strategy is adopted. The formula for calculating the model fusion strategy vector is as follows: in, and These represent the vector outputs of the two models, respectively. λ∈[0,1] represents the model fusion weights, which the system can dynamically adjust according to the task.

[0027] Each text segment vector is set to 1536 dimensions to ensure sufficient semantic expressiveness while controlling storage space overhead. The system uses a weighted fusion strategy to select the optimal vector model combination based on the retrieval scenario, thereby improving the accuracy and recall of semantic matching.

[0028] The vector database uses PostgreSQL as its base database and integrates a vector retrieval engine, supporting the following features: Vector similarity retrieval: Top-K search using cosine similarity; Keyword and vector hybrid retrieval: Based on semantic retrieval, keyword Boolean query is combined to ensure that the recalled text units simultaneously satisfy semantic relevance and keyword exact matching; Contextual joint retrieval: Combining historical query records to achieve contextual understanding in continuous dialogue scenarios.

[0029] To optimize retrieval performance, a retrieval performance evaluation function is introduced, and retrieval performance is expressed by the following formula: in, Labels indicating the true relevance of search results Predict labels for the model, For indicator functions, To retrieve the number of results, For retrieval time, This is a time penalty factor.

[0030] In the user interaction module, users can input natural language queries through the web front-end or API calls, such as: "Please provide the seismic resistance level acceptance standard for buildings".

[0031] User input is first converted into a semantic vector, and then sent to the retrieval module for processing.

[0032] The semantic relevance of a user query to the candidate context is evaluated using the following formula: in, For query vector, For context vectors, To prevent positive constants with a denominator of zero.

[0033] The system supports voice input and multilingual input, automatically performs language detection and translation, and ensures effective question-and-answer communication in cross-language scenarios.

[0034] The retrieval and context generation module uses FlowController to dynamically orchestrate retrieval strategies. Automatically adjust the search scope; Candidates are ranked by contextual relevance; Remove invalid context; By piecing together logically coherent contextual information, the accuracy of subsequent question-answering model inputs is ensured.

[0035] To control the logic of context concatenation, a context reorganization scoring function is introduced. The context concatenation logic function is as follows: in, The number of context segments after reorganization. Indicates the first The relevance score of the paragraph to the context. Contexts with higher scores have higher priority during concatenation to ensure the integrity of the generated context logic.

[0036] The intelligent question-answering module includes the following sub-units: Model fine-tuning unit: Based on labeled data in the construction engineering field, transfer learning is performed on the large language model to enable it to parse professional clauses.

[0037] Multi-task learning unit: Joint execution of clause classification, compliance judgment and standard clause correlation analysis tasks.

[0038] Continuous learning unit: Regularly retrieves updated data from the government standards database and automatically performs incremental training of the model to ensure that the question-and-answer results are always consistent with the latest standards.

[0039] When a user queries "Standard for Testing Concrete Strength of High-Rise Buildings", the system can automatically return a hierarchical and structured answer, including: the original text of the standard clauses; an explanation of the applicable conditions of the clauses; and the source and version number of the standard.

[0040] In multi-task training, a joint loss function is used, which is: in, Classify losses according to the terms. To compensate for the losses incurred in compliance with regulations, For losses related to the provisions, , , These are task weight parameters to ensure the model's overall performance across multiple tasks.

[0041] The system supports multilingual output, hierarchical summaries, and tabular result display, making it easy for users to directly use for project management and compliance review.

[0042] In the security and access control module, this embodiment employs the following security strategies throughout the entire process of data access, transmission, and answer output: Data encryption: Data transmission encryption based on SSL / TLS; Access control is hierarchical: different user roles are assigned different query permissions; Operation log recording: All query and response processes are logged to meet the compliance requirements of the project.

[0043] The access control policy is defined by the following function: in, Indicates user permission levels, For resources The minimum required access level ensures that sensitive standards are only accessible to authorized engineers or administrators.

[0044] In the multi-source knowledge fusion module, when merging government, enterprise, and third-party data, the system automatically performs the following: Deduplication across different source standards; unified conversion of heterogeneous formats; cleanup of redundant knowledge.

[0045] Ultimately, a structured and unified knowledge base of engineering quality inspection and testing standards will be formed.

[0046] When integrating government standards, third-party testing data, and internal enterprise documents, the system uses information entropy weight allocation, and introduces an information entropy weight allocation function during the integration process: in, Indicates the first The information contribution of each data source The system automatically assigns weight ratios to documents from different sources within the fused knowledge base, based on their respective weights in the fusion results.

[0047] System operation method The system operation includes the following steps: Step 1: Acquire external knowledge data and format and store it. Step 2: Perform semantic segmentation to form multiple logically independent text units; Step 3: After vectorization, store the vector data in a vector database; Step 4: Receive the user's natural language query and vectorize it into a query vector; Step 5: Retrieve matching text units and reorganize them into contextual information; Step 6: Generate a Prompt and input it into the fine-tuned large language model; Step 7: Output structured, multilingual professional answers.

[0048] The above specific embodiments are merely several preferred embodiments of the present invention. Based on the technical solutions of the present invention and the relevant teachings of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

[0049] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention.

Claims

1. A smart question-and-answer system for construction engineering quality inspection and testing standards based on FastGPT, comprising an external knowledge acquisition module, a document processing module, a vectorization module, a vector database, a user interaction module, a retrieval and context generation module, and a smart question-and-answer module, characterized in that: The external knowledge acquisition module is used to automatically and manually access technical documents related to standards, specifications, and test reports in the construction engineering field. It supports the import of data in the following formats, including PDF, Word, Excel, and web page data, to ensure the comprehensiveness and authority of the knowledge source. The document processing module is used to preprocess the external knowledge. The preprocessing includes removing redundant information, text cleaning, semantic segmentation, and dividing the document content into multiple logically complete small text units (Chunks) to facilitate subsequent vectorization processing. The vectorization module is used to perform high-dimensional semantic vectorization processing on the text unit, extract semantic features using a deep language model, and generate corresponding vector representations. The vector database is used to store the vectorized text units, supports fast retrieval based on vector similarity, and dynamically returns the most relevant context information for query requests. The user interaction module is used to receive natural language query questions input by users and convert the query questions into semantic vectors to ensure matching with the vector database. The retrieval and context generation module is used to perform similarity retrieval in the vector database based on the query vector, call the text units related to the query, and dynamically reorganize their context information to form logically consistent context content. The intelligent question-answering module is used to combine the user query with the context information to form a unified prompt, and input the large language model (LLM) fine-tuned with data from the construction engineering field to automatically output a professional answer that conforms to the construction engineering quality inspection and testing standards.

2. The intelligent question-and-answer system for construction engineering quality inspection and testing standards based on FastGPT as described in claim 1, characterized in that: The document processing module further includes: The QA splitting unit automatically generates question-answer pairs based on the document's semantic structure, enhancing the targeting of subsequent searches; Semantic segmentation units are used to segment text based on the document's logical structure and semantic features, ensuring the semantic integrity of the segmented text units. The manual correction unit is used to manually correct and supplement the results of automatic processing to ensure the accuracy and completeness of professional knowledge.

3. The intelligent question-and-answer system for construction engineering quality inspection and testing standards based on FastGPT as described in claim 1, characterized in that: The vectorization module supports multi-model collaboration, including at least one of the M3E vector model and the OpenAIembedding model. The system can automatically select and weightedly fuse the outputs of different vector models based on the characteristics of the retrieval task to improve the accuracy and recall of the retrieval.

4. The intelligent question-and-answer system for construction engineering quality inspection and testing standards based on FastGPT as described in claim 1, characterized in that: The intelligent question-answering module includes: The model fine-tuning unit is used to perform transfer learning on a general language model based on labeled data in the construction engineering field, enabling it to recognize and correctly understand professional terms and standard clauses in the construction engineering field. The multi-task learning unit, based on the fine-tuned model, jointly performs multi-dimensional task training such as clause classification, compliance judgment, and standard clause correlation analysis to improve the model's comprehensive reasoning and question-answering capabilities in complex scenarios. The continuous learning unit is used to automatically collect and update knowledge data based on newly released industry standards, specifications or regulations, while performing incremental training of model parameters to ensure that the knowledge base and language model of the question answering system are always up-to-date.

5. The intelligent question-and-answer system for construction engineering quality inspection and testing standards based on FastGPT as described in claim 1, characterized in that: The vector database is built on PostgreSQL and supports a hybrid mode of vector similarity retrieval and keyword retrieval. It can also perform contextual joint retrieval by combining historical query records to improve the contextual continuity and professional completeness of the answers.

6. The intelligent question-and-answer system for construction engineering quality inspection and testing standards based on FastGPT as described in claim 1, characterized in that: The retrieval and context generation module uses FlowController to dynamically orchestrate retrieval strategies. The orchestration includes automatic adjustment of the retrieval scope, relevance ranking of candidate contexts, removal of invalid contexts, and final context concatenation, in order to improve question-answering accuracy and professional reliability while ensuring reasoning efficiency.

7. The intelligent question-and-answer system for construction engineering quality inspection and testing standards based on FastGPT as described in claim 1, characterized in that: The intelligent question-answering module supports multilingual formatting, hierarchical summary extraction, and structured result display when outputting answers, in order to meet the intelligent question-answering needs in cross-language and multi-professional scenarios.

8. The intelligent question-and-answer system for construction engineering quality inspection and testing standards based on FastGPT as described in claim 1, characterized in that: The question-and-answer method of the system includes the following steps: Step 1: Obtain relevant external knowledge data such as construction engineering standards, specifications, and test reports, and store them in a formatted manner; Step 2: Semantically segment the external knowledge to form multiple logically independent text units; Step 3: Vectorize the text units and store them in a vector database that supports similarity retrieval; Step 4: Receive the natural language query input by the user and vectorize it into a query vector; Step 5: Retrieve the text unit that best matches the query vector from the vector database and reorganize it into context information; Step 6: Concatenate the user query with the context information to generate a prompt, and input it into the finely tuned large language model; Step 7: Output structured answers related to construction project quality inspection and testing standards.

9. The intelligent question-and-answer system for construction engineering quality inspection and testing standards based on FastGPT as described in claim 8, characterized in that: The system further includes a security and access control module, which implements multi-level security policies for the access of external knowledge, the transmission of query data, and the output of answers. These policies include data encryption, access permission classification, and operation log recording to ensure the data security and compliance of the system in actual engineering projects.

10. The intelligent question-and-answer system for construction engineering quality inspection and testing standards based on FastGPT according to claim 1, characterized in that: The system further includes a multi-source knowledge fusion module, which is used to fuse multi-source data from government standard databases, third-party testing agency documents, and enterprise internal documents to solve the problems of heterogeneity and redundancy across standards and improve the universality and completeness of the knowledge base in multiple scenarios.

Images