AI knowledge base question-answering system for open caisson construction and electronic equipment

Abstract

The invention provides an AI knowledge base question-answering system for open caisson construction and electronic equipment, and relates to the technical field of intelligent construction and engineering informatization. The system comprises a knowledge base construction module used for constructing a domain-specific knowledge base comprising a vector index and a graph database; the intelligent question and answer module is used for receiving natural language questions of the user and generating professional answers with references; the scheme generation module is used for generating a construction technical scheme containing texts and multi-modal drawings based on the parameters input by the user; the risk early warning module is used for identifying construction risks based on the real-time monitoring data and triggering early warning and emergency response; data in various construction fields are integrated to construct an exclusive knowledge base, and knowledge is quickly and accurately recognized based on semantic analysis and mixed retrieval, so that the construction decision-making efficiency is improved; historical data are integrated by means of a scheme generation mechanism, a targeted scheme is generated for major risk events, the method adapts to complex and changeable construction environments, and personnel dependence is reduced.

Description

Technical Field

[0001] This invention relates to the field of intelligent construction and engineering information technology, and in particular to an AI knowledge base question-and-answer system and electronic device for caisson construction. Background Technology

[0002] Caisson construction involves multiple disciplines and processes, including soil mechanics, structural mechanics, hydrogeology, construction machinery, monitoring and testing, and safety management. Related knowledge exists not only in standards and specifications, design documents, construction organization plans, and monitoring data, but is also scattered across expert experience and construction logs, lacking a unified knowledge organization and structured management approach. Traditional retrieval methods, primarily relying on keywords and manual indexing, struggle to quickly and accurately acquire multi-source, heterogeneous knowledge, impacting the efficiency of construction decision-making.

[0003] While existing general-purpose language models possess powerful language understanding and generation capabilities, they are prone to insufficient knowledge coverage, low accuracy in responses, and the "AI illusion" phenomenon (generating answers that do not conform to reality) in highly specialized scenarios such as caisson construction. Once the model outputs incorrect information, it may lead to deviations in construction plans or safety hazards, making it difficult to meet the rigor and reliability requirements of the engineering construction field.

[0004] During the construction of caissons, traditional experience relies on manual planning, which is time-consuming and easily affected by the limitations of personal experience, resulting in insufficient targeting of the plan and difficulty in dynamically adapting to complex and ever-changing construction environments.

[0005] Significant risks are frequently encountered during the sinking of caissons. Current emergency response methods rely on manual judgment and lack intelligent analysis and dynamic response capabilities. Summary of the Invention

[0006] The purpose of this invention is to provide an AI knowledge base question-and-answer system and electronic device for caisson construction. The system comprises a knowledge base construction module, an intelligent question-and-answer module, a solution generation module, and a risk warning module. It integrates data from multiple construction fields to build a dedicated knowledge base and uses semantic parsing and hybrid retrieval to quickly and accurately identify knowledge, improving construction decision-making efficiency. Furthermore, by integrating historical data through a solution generation mechanism, it generates targeted solutions for major risk events, adapting to complex and ever-changing construction environments and reducing reliance on personnel.

[0007] In a first aspect, the present invention provides an AI knowledge base question-and-answer system for caisson construction, comprising: a knowledge base construction module, an intelligent question-and-answer module, a scheme generation module, and a risk warning module; the intelligent question-and-answer module is connected to the knowledge base construction module via data; the scheme generation module is connected to the knowledge base construction module and the intelligent question-and-answer module via data respectively; the risk warning module is connected to the knowledge base construction module and the intelligent question-and-answer module via data respectively, connected to an external construction monitoring system via signals, and connected to the scheme generation module via instructions. The knowledge base construction module is used to collect, clean, structure, and store multi-source heterogeneous data in the field of caisson construction, and to build a domain-specific knowledge base that includes vector indexes and graph databases. The intelligent question-answering module receives users' natural language questions, retrieves evidence fragments from a domain-specific knowledge base through semantic parsing and hybrid retrieval, and generates professional answers with citations. The solution generation module is used to extract content from a domain-specific knowledge base based on user-input parameters and populate it into a preset template to generate a construction technology solution containing text and multimodal drawings. The risk warning module is used to identify construction risks based on real-time monitoring data and trigger warnings and emergency responses.

[0008] In some preferred embodiments of the present invention, the knowledge construction module includes: The multi-source data acquisition unit is used to automatically extract raw data from industry data. Industry data includes: design drawings, technical specifications, construction logs, construction organization designs, special construction plans, typical cases, industry papers and patents; raw data includes: text, tabular and graphical data. The data cleaning unit is used to preprocess the raw data, including removing redundancy and error information and performing format conversion. The structured processing and semantic standardization unit is used to transform unstructured text data into structured triples using named entity recognition and relation extraction models, and to merge synonyms for caisson construction terminology; wherein, the structure of the structured triple is <first entity, relation, second entity>. The dual storage unit of vector and graph is used to vectorize text fragments and store them in vector indexes to achieve semantic recall, while importing structured triples into the graph database to support logical reasoning. The incremental update unit is used to perform online incremental updates to the domain-specific knowledge base using a timestamp and version number mechanism.

[0009] In some preferred embodiments of the present invention, the intelligent question-answering module adopts a multi-level hybrid retrieval strategy; the multi-level hybrid retrieval strategy includes vector retrieval, knowledge graph retrieval and keyword full-text retrieval performed sequentially; Vector retrieval is used to perform semantic similarity matching in a vector index based on the semantic vector of the user's question; Knowledge graph retrieval is used to find knowledge points in a graph database that have preset relationships and attributes with entities in the user's question; Keyword full-text search is used for precise matching based on keywords; The results of the mixed retrieval will be merged and reordered to form a final set of evidence fragments for answer generation; the working mechanism followed by the intelligent question answering module is retrieval first, and generation as a supplement; all answers generated by the intelligent question answering module are based on evidence fragments retrieved from the domain-specific knowledge base; the answers are accompanied by a citation tracing mechanism, including the normative clause number and / or construction case source of the cited knowledge.

[0010] In some preferred embodiments of the present invention, the intelligent question-answering module further includes: The interactive clarification unit is used to automatically generate two-option and / or open-ended clarification questions to guide users to supplement key information when the system's confidence level in a user's question is lower than a preset threshold. The confidence level is obtained by calculating the weighted score of four dimensions of signals between the recalled evidence fragments and the initially generated answers. The four dimensions of signals include: fragment coverage, answer overlap, relevance, and model self-evaluation probability.

[0011] In some preferred embodiments of the present invention, the scheme generation module specifically includes: The template-driven unit has a built-in standardized template library covering the entire process of caisson construction. The standardized template library includes at least the following standardized templates: construction organization design template, special construction plan template, monitoring and control template, and risk management template. Among them, special construction includes at least one of the following: caisson fabrication, caisson sinking, caisson bottom sealing, and real-time correction during the caisson sinking process. The parameter mapping unit is used to automatically map the geological conditions, hydrological parameters, caisson geometric dimensions and construction equipment requirements input by the user to template variables, and call the domain-specific knowledge base content in the knowledge base construction module to fill the template content in a targeted manner using a large language model; The drawing rendering unit is used to dynamically generate caisson structure diagrams, construction sequence diagrams, and monitoring point layout diagrams based on scalable vector graphics and network graphics library technology.

[0012] In some preferred embodiments of the present invention, the solution generation module is also used to feed the user-confirmed technical solution back to the knowledge base construction module as structured document data, so as to trigger incremental updates of the domain-specific knowledge base and form an application closed loop from solution generation to knowledge accumulation.

[0013] In some preferred embodiments of the present invention, the scheme generation module is also used to export the generated construction technology scheme and supporting drawings in multiple formats such as portable document format, engineering drawing format, and industrial basic format.

[0014] In some preferred embodiments of the present invention, the risk warning module includes: The monitoring data access and early warning unit is used to connect with the construction monitoring system, access real-time data from the positioning system, tilt angle, stress and strain, elevation, and water level, and trigger a risk warning when at least one sensor indicator exceeds a preset threshold. The risk identification unit is used to comprehensively analyze monitoring indicators that exceed the threshold and, in conjunction with risk case data in the knowledge base construction module, identify the risks of caisson construction. The emergency response plan generation unit, under user instructions, uses the risk cases, standards, and plan information in the knowledge base construction module to drive the plan generation module to generate a structured emergency response plan that includes a description of the risk event, emergency measures, and tracking procedures.

[0015] In some preferred embodiments of the present invention, the system further includes: The history module is used to write the original question, the evidence chain, the generated answer, the cited text, the user feedback, and the timestamp of each Q&A interaction into the database in a related data format, and generate a unique session identifier; at the same time, it records risk warning events and generated solutions, forming a traceable record of the entire life cycle of system operation.

[0016] In a second aspect, the present invention provides an electronic device including a processor and a memory, the memory storing computer-executable instructions that can be executed by the processor, the processor executing the computer-executable instructions to run the AI ​​knowledge base question-and-answer system for caisson construction provided in the first aspect above.

[0017] This invention brings the following beneficial effects: This invention provides an AI knowledge base question-answering system and electronic device for caisson construction, comprising: a knowledge base construction module, an intelligent question-answering module, a solution generation module, and a risk warning module; the intelligent question-answering module is connected to the knowledge base construction module via data; the solution generation module is connected to both the knowledge base construction module and the intelligent question-answering module via data; the risk warning module is connected to both the knowledge base construction module and the intelligent question-answering module via data, connected to an external construction monitoring system via signals, and connected to the solution generation module via instructions; the knowledge base construction module is used to collect, clean, structure, and store multi-source heterogeneous data in the field of caisson construction, constructing a domain-specific knowledge base including vector indexes and graph databases; the intelligent question-answering module is used to receive users' natural language questions and retrieve them from the domain-specific knowledge base through semantic parsing and hybrid retrieval. The system acquires evidence fragments and generates professional answers with citations; the solution generation module extracts content from a domain-specific knowledge base based on user input parameters and fills it into a preset template to generate construction technical solutions containing text and multimodal drawings; the risk warning module identifies construction risks based on real-time monitoring data and triggers warnings and emergency responses; an AI knowledge base question-and-answer system for caisson construction is built, including a knowledge base construction module, an intelligent question-and-answer module, a solution generation module, and a risk warning module. It integrates data from multiple construction fields to build a dedicated knowledge base and quickly and accurately identifies knowledge based on semantic parsing and hybrid retrieval, improving the efficiency of construction decision-making; relying on the solution generation mechanism to integrate historical data, it generates targeted solutions for major risk events, adapting to complex and ever-changing construction environments and reducing reliance on personnel. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This invention provides an architecture diagram of an AI knowledge base question-answering system for caisson construction. Figure 2 A flowchart for constructing a knowledge base is provided as an embodiment of the present invention; Figure 3 A flowchart of an intelligent question-answering process provided in an embodiment of the present invention; Figure 4 A flowchart for generating a solution is provided in an embodiment of the present invention; Figure 5 A risk warning flowchart provided as an embodiment of the present invention; Figure 6This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0020] Icons: 400 - Memory; 401 - Processor; 402 - Bus; 403 - Communication Interface. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0023] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0024] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of this invention is in use. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. In addition, the terms "first," "second," "third," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0025] Furthermore, terms such as "horizontal," "vertical," and "sag" do not imply that components must be absolutely horizontal or suspended, but rather that they can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," not that the structure must be completely horizontal, but can be slightly tilted.

[0026] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0027] Caissons, as a type of deep foundation that is "prefabricated on the ground and then sunk to the design elevation by their own weight," have been widely used in projects such as bridges across rivers and seas, wharves, and water intake and drainage pumping stations. In the past decade, with the continuous increase in span and water depth, the planar dimensions of caissons have developed from the early 10m range to the 70m range, and the burial depth has increased from 20m to over 80m.

[0028] Although BIM, finite element method, and geological 3D modeling technologies are widely used in structural calculations and construction simulations, and multi-source sensors such as hydrostatic levels, fiber optic gratings, GNSS, and sonar mud surface scanning can acquire data such as tilt, stress, mud surface elevation, and groundwater level in real time, and closed-loop control systems composed of hydraulic grab buckets, air suction mud, and ballast water automatic leveling equipment have enabled automatic correction of sinking attitude in some projects, the following technical problems still exist in the field of caisson construction: Knowledge fragmentation and inefficient retrieval: The knowledge in the field of caisson construction is highly specialized, and multidisciplinary and multi-source heterogeneous data are scattered in specifications, drawings, logs, monitoring data and expert experience, lacking a unified semantic system. Traditional keyword retrieval is difficult to overcome semantic gaps such as synonyms of professional terms, abbreviations and implicit conditions.

[0029] The general-purpose large model lacks professional adaptation: Most existing general-purpose large models are trained based on generalized corpora, which leads to misunderstandings of terms such as "blade reaction force coefficient" and "critical frictional resistance of sudden sinking", which can easily cause illusions. In addition, they cannot synchronize with the latest standard provisions in real time, which poses a risk of timeliness.

[0030] The generation of construction plans lacks specificity: In caisson construction, professional technical data such as construction plans need to take into account multiple factors such as geology, hydrology, structure, and on-site equipment. Traditional manual compilation is time-consuming and limited by individual experience. Traditional large models only support text generation and lack multimodal information such as drawings and flowcharts, making it difficult to achieve integrated output of "text-charts-drawings".

[0031] Lagging risk warning and emergency response: During the construction of caissons, emergencies such as "blockage, tilting, sand inrush, and piping" require minute-level response. Current handling methods rely on manual review of specifications or expert experience. Although the construction monitoring system can collect data in real time, it lacks intelligent analysis linked with the knowledge base, resulting in a break in the closed loop of warning-decision-response.

[0032] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0033] Example 1 This invention provides an AI knowledge base question-answering system for caisson construction, see [link to relevant documentation]. Figure 1 The illustrated embodiment of the present invention provides an AI knowledge base question-answering system architecture for caisson construction. The system includes: a knowledge base construction module, an intelligent question-answering module, a solution generation module, and a risk warning module. The intelligent question-answering module is connected to the knowledge base construction module via data. The solution generation module is connected to both the knowledge base construction module and the intelligent question-answering module via data. The risk warning module is connected to both the knowledge base construction module and the intelligent question-answering module via data, to an external construction monitoring system via signals, and to the solution generation module via instructions.

[0034] Specifically, the system constructs a professional and highly reliable intelligent question-answering platform through a modular architecture and retrieval-enhanced generation technology. The modules form an organic whole, with the knowledge base construction module providing data support as the system's foundation, the intelligent question-answering module serving as the core interaction interface, the solution generation module expanding the system's application capabilities, and the risk warning module extending the system's capabilities to real-time construction monitoring scenarios. This interconnected architecture ensures the smooth transmission of data, signals, and instructions between the modules, achieving a complete closed loop from knowledge acquisition and intelligent question-answering to solution generation and risk response.

[0035] The knowledge base construction module (K-Builder) is the foundation of the entire system. It is responsible for structuring the scattered knowledge in the field of caisson construction and storing it in a vector-knowledge graph dual-storage database. This includes extracting entities, relationships, and vector representations from various specifications, drawings, and logs.

[0036] When a user poses a natural language question through the user interface (Web / Mobile), the intelligent question answering module (QA-Core) is activated. QA-Core first performs semantic parsing and intent recognition on the question, then initiates a hybrid retrieval process, recalling the most relevant evidence fragments from the knowledge base. These evidence fragments are then fed into a finely tuned large language model to generate an accurate and professional answer. The generated answer will include citation tags, allowing users to trace back to the original source. Simultaneously, the user's question, the system's response, and the citation chain are recorded in the history module for subsequent dialogue management and system optimization.

[0037] When a user needs to develop a technical solution, the solution generation module (Plan-Gen) is invoked. After the user inputs key parameters, Plan-Gen uses a large model to extract relevant content from the knowledge base, automatically fills in a preset template, and dynamically renders and generates related drawings. After the generated solution is confirmed by the user, it can be fed back into the knowledge base as a new knowledge source for incremental updates.

[0038] At the construction site, the Risk-Warn module continuously monitors data from various sensors. Once an abnormality is detected in the monitored indicators, the module immediately issues an alert and, based on risk cases and specifications in the knowledge base, calls upon the QA-Core and Plan-Gen modules to assist on-site personnel in quickly generating corrective solutions, thus achieving a closed loop from knowledge to decision-making to on-site application.

[0039] The knowledge base construction module is used to collect, clean, structure, and store multi-source heterogeneous data in the field of caisson construction, and to build a domain-specific knowledge base that includes vector indexes and graph databases.

[0040] Specifically, through a systematic data processing workflow, the professional knowledge scattered across standards, design documents, construction organization plans, monitoring data, expert experience, and construction logs in the field of caisson construction is unified and structured for management. This solves the problem that traditional retrieval methods, which rely primarily on keywords and manual indexing, struggle to quickly and accurately acquire multi-source heterogeneous knowledge, providing high-quality, structured knowledge support for the entire system.

[0041] Furthermore, in some preferred embodiments of the present invention, the knowledge construction module includes: a multi-source data acquisition unit, used to automatically extract raw data from industry data; wherein, industry data includes: design drawings, technical specifications, construction logs, construction organization designs, special construction plans, typical cases, industry papers and patents; raw data includes: text, tabular and graphical data; a data cleaning unit, used to preprocess the raw data, wherein, preprocessing includes: removing redundant and erroneous information and performing format conversion; a structured processing and semantic standardization unit, used to use named entity recognition and relation extraction models to convert unstructured text data into structured triples, and to merge synonyms for caisson construction professional terms; wherein, the structure of the structured triple is <first entity, relation, second entity>; a vector and graph dual storage unit, used to vectorize text fragments and store them in a vector index to achieve semantic recall, and simultaneously import the structured triples into a graph database to support logical reasoning; and an incremental update unit, used to perform online incremental updates to the domain-specific knowledge base through a timestamp and version number mechanism.

[0042] For details, see Figure 2The illustrated embodiment of the present invention provides a knowledge base construction flowchart. All functions of this system are built upon a high-quality professional knowledge base. This knowledge base is built and dynamically maintained by the K-Builder module. This module serves as the knowledge entry point, using a series of technical means to transform scattered and heterogeneous knowledge about caisson construction into a structured and searchable format.

[0043] First, K-Builder possesses multi-source data acquisition and processing capabilities. The collected construction-related documents, such as design drawings (DWG, PDF), national and industry standards (e.g., "Construction Specifications for Caissons and Pneumatic Caissons GB / T 51130-2016", "Technical Standards for Caisson and Caisson Construction DG / TJ 08-2084-2023"), construction logs, geological survey reports, expert review reports, and historical case studies, undergo preprocessing, including OCR recognition of drawing text, format conversion, and redundant information removal, to ensure data quality.

[0044] Next, by constructing and maintaining a professional dictionary in the field of caisson construction, the synonyms and abbreviations in the field of caisson knowledge are semantically standardized. For example, "caisson", "caisson", and "open caisson" are unified as "caisson", which avoids the same concept being treated as multiple different entities and improves data consistency and query accuracy.

[0045] Secondly, the cleaned data undergoes structuring processing. Based on the Transformer's Named Entity Recognition (NER) and Relation Extraction (RE) models, unstructured text content is parsed into <entity, relation, entity> triples. For example, from a technical description of a correction method, after automatic segmentation and sentence division, the large model can automatically extract "jack counterweight" as an entity, "correction means" as a relation, and "caisson tilt" as another entity, thus establishing the <caisson tilt, correction means, jack counterweight> triple. In this way, various scattered professional terms in the caisson field can be linked together for subsequent logical analysis.

[0046] Furthermore, data is stored using both vector and knowledge graph methods. All text fragments are segmented and vectorized, then stored in a vector database (such as FAISS) to support efficient semantic similarity retrieval. Simultaneously, the structured triples are used to construct a knowledge graph, stored in a graph database (such as Neo4j) to support complex logical reasoning and relational queries. This dual-storage architecture combines vector and knowledge graph approaches, leveraging the strengths of both to ensure the system can quickly respond to user semantic queries while also supporting rigorous logical deduction.

[0047] Finally, K-Builder has an incremental update mechanism. If there are new construction logs or specification revisions, the knowledge base can be updated to ensure its timeliness and accuracy.

[0048] For example, the knowledge building function includes the following steps: Step 1: Obtaining and cleaning the original files.

[0049] Input: PDF scanned version of "Construction Organization Plan for the North Tower Caisson of the XX Yangtze River Bridge".

[0050] OCR: Detects paragraphs, tables, and figures, and converts them into common text formats.

[0051] Redundant information cleanup: Automatically deletes noisy information such as headers, footers, "Page ×× of ××", and copyright notices.

[0052] Step 2, semantic standardization.

[0053] Use the domain dictionary of the well to perform bidirectional maximum matching.

[0054] “Ceiling box” → “caisson”; “blade angle” → “blade foot”; “grab bucket soil extraction” → “grab bucket soil digging”.

[0055] Step 3, semantic slicing.

[0056] The organizational plan should be divided into segments, neither too fragmented nor too long.

[0057] If a single section is longer than 300 characters, the information should be further segmented by line or by semicolons, etc. Step 4, Entity-Relation Extraction (NER-RE).

[0058] Taking a certain section of Article 4.3.2 of the "Caison Construction Organization Plan" as an example: When the cutting edge of the caisson enters the ⑤2 silty clay layer, it is prone to sudden sinking due to the low strength and high water content of the soil in this layer. An air suction dredger should be turned on to reduce the water head in the well, and a 200-ton counterweight platform should be symmetrically loaded on the top of the well to control the sinking speed to not exceed 0.3 m / d.

[0059] The NER module automatically extracted the following entity results: E1:⑤2 Silty clay layer; E2: Cutting edge; E3: Sudden sinking; E4: Air suction sludge machine; E5: Water head in well; E6: Counterweight platform; E7: Sinking speed; E8: 0.3 m / d.

[0060] Further, the RE module is run to extract relations, adding relations to the above entities to form triples: R1: ⑤2 Silty clay layer — (prone to) → sudden settlement; R2: Air suction sludge machine — (used to reduce) → water head in the well; R3: Counterweight platform — (control) → Sinking speed; R4: Sinking velocity — (threshold) → 0.3 m / d; In summary, this section of content yielded a total of 4 triplet information through NER-RE processing.

[0061] Step 5, vector writing.

[0062] The generated fragments after slicing are fed into the BERT model to generate dense vectors, which are then written to the FAISS database and indexed.

[0063] Step 6: Write the knowledge graph.

[0064] The structured triple information is imported into the Neo4j database to form a knowledge graph.

[0065] Step 7, Knowledge Base Management and Incremental Updates.

[0066] Generate a UUID fingerprint for the entire "xx Construction Organization Plan", which records information such as its release date and validity status.

[0067] When the system encounters a "revision" with the same name as this organizational scheme, the system compares the fingerprint differences of the files and automatically triggers steps 1 to 6 above to achieve incremental updates.

[0068] The intelligent question-answering module receives users' natural language questions, retrieves evidence fragments from a domain-specific knowledge base through semantic parsing and hybrid retrieval, and generates professional answers with citations.

[0069] Specifically, the intelligent question-answering module addresses the issues of insufficient knowledge coverage, low accuracy, and the "AI illusion" phenomenon in highly specialized scenarios such as caisson construction using general-purpose language models. It constructs a reliable question-answering mechanism by pre-training the basic BERT model on over 200,000 industry-specific corpora, including caisson specifications, construction logs, and technical Q&A. This allows the model to fully grasp the contextual semantics of specialized terms such as "cutting edge," "correction," and "bottom sealing," ensuring accurate understanding of user questions.

[0070] Furthermore, in some preferred embodiments of the present invention, the intelligent question-answering module employs a multi-level hybrid retrieval strategy. This strategy includes sequential vector retrieval, knowledge graph retrieval, and keyword full-text retrieval. Vector retrieval is used to perform semantic similarity matching in a vector index based on the semantic vector of the user's question. Knowledge graph retrieval is used to find knowledge points in a graph database that have preset relationships and attributes with entities in the user's question. Keyword full-text retrieval is used for precise matching based on keywords. The results of the hybrid retrieval are then fused and reordered to form a final set of evidence fragments for answer generation. The intelligent question-answering module follows a retrieval-first, generation-supplemented working mechanism. All answers generated by the intelligent question-answering module are based on evidence fragments retrieved from a domain-specific knowledge base. The answers include a citation tracing mechanism, including the normative clause number and / or construction case source of the cited knowledge.

[0071] Specifically, the three-tiered hybrid retrieval strategy simultaneously executes vector retrieval, knowledge graph retrieval, and BM25 full-text retrieval, and finally merges and re-ranks the results. Vector retrieval performs rapid semantic similarity matching in the vector index based on the semantic vector of the question, recalling semantically relevant document fragments; knowledge graph retrieval searches for complex relationship paths between entities in the graph database; and keyword full-text retrieval ensures accurate matching based on keywords, preventing the omission of important information due to semantic misunderstanding biases. This hybrid retrieval strategy maximizes the recall of all evidence related to the question, significantly improving the recall and precision of the retrieval.

[0072] Furthermore, in some preferred embodiments of the present invention, the intelligent question-answering module further includes: an interactive clarification unit, used to automatically generate two-option and / or open-ended clarification questions to guide the user to supplement key information when the system's confidence level in the user's question is lower than a preset threshold; wherein, the confidence level is obtained by calculating the weighted score of four-dimensional signals between the recalled evidence fragment and the initially generated answer, the four-dimensional signals including: fragment coverage, answer overlap, relevance, and model self-evaluation probability.

[0073] For details, see Figure 3 The illustrated embodiment of the present invention provides an intelligent question-answering flowchart. Once the knowledge base is established, the intelligent question-answering module (QA-Core) begins to play its core role, providing users with intelligent question-answering and decision support. All of its answers must rely on the evidence provided by the knowledge base.

[0074] When on-site technicians raise a natural language question, such as, "How to control uneven settlement in soft soil foundations?", QA-Core first performs semantic parsing to accurately understand the user's true needs and extract key information such as "soft soil foundation" and "uneven settlement." To achieve this capability, the system has pre-trained its basic BERT on over 200,000 industry-specific data points, including caisson specifications, construction logs, and technical Q&As, ensuring it fully grasps the contextual semantics of professional terms such as "edge," "correction," and "bottom sealing." During the parsing phase, the model performs three tasks in parallel: intent recognition, determining whether the user wants "control measures," "cause analysis," or "emergency handling"; slot filling, automatically extracting key fields such as geological conditions, construction stage, and caisson type; and context completion, logically restoring omitted subjects, times, or locations, such as completing the question as "When constructing caissons in soft soil foundations, what control measures can be used to control uneven settlement?"

[0075] After semantic parsing, QA-Core performs confidence testing through two steps: "retrieval" and "reading." "Retrieval" involves recalling the Top-K segments of text (K is typically 3-5) from the knowledge base. "Reading" refers to outputting the answer internally using the generative model, while simultaneously having the model provide the probability of whether the answer is sufficiently supported by the recalled segments. This probability is calculated as a weighted score of the four-dimensional signals of the recalled segments and the answer: conf = w1·chunk_coverage + w2·answer_span_overlap + w3·passage_relevance_score + w4·model_self_check. The four-dimensional signals specifically refer to segment coverage, answer overlap, relevance, and self-assessment probability. Recommended empirical values ​​for w1 to w4 are 0.25, 0.25, 0.3, and 0.2, respectively. After normalization, output values ​​between 0 and 1. When the system's confidence level (c) for a user's question is lower than a preset threshold (0.8 is recommended), QA-Core can automatically generate two-option or open-ended follow-up questions, generate clarification questions, and guide the user to supplement information.

[0076] Once the question is defined, the system enters a three-level hybrid retrieval process. QA-Core first performs vector retrieval (intensive retrieval), using the semantic vector of the question to quickly retrieve all semantically relevant document fragments from the knowledge base. Simultaneously, it performs relational retrieval in the knowledge graph, searching for knowledge nodes such as "control measures" or "emergency plans" related to "uniform settlement of caissons." Finally, it initiates BM25 full-text retrieval (sparse retrieval) to ensure accurate keyword-based matching and prevent the omission of important information due to semantic comprehension biases. By executing intensive retrieval, relational retrieval, and sparse retrieval separately, and then fusing and reordering the results, this hybrid retrieval strategy maximizes the retrieval of all evidence related to the question. (Whether it's simultaneous or sequential retrieval needs to be clarified – it's performed simultaneously, and the results are then fused and reordered).

[0077] The retrieved evidence fragments are then fed into a lightweight, fine-tuned large language model. This model strictly adheres to a "retrieval-first, generation-supplement" working mechanism, serving only as a tool for language organization and logical restructuring, rather than an independent source of knowledge. It synthesizes, refines, and organizes multiple evidence fragments to generate a clearly structured and logically rigorous answer. To completely eliminate "AI illusions," each answer must include citation traceability, clearly indicating the source of the cited knowledge, such as the code clause number or construction log date. Users can click on these citations to directly jump to the original document for verification, ensuring the reliability and compliance of the answer.

[0078] After the answer is generated, users can perform three types of feedback with one click in the same dialog interface: "accept", "modify" or "reject". The system will store the feedback results in the historical version library and continuously optimize the model through fine-tuning and reinforcement learning.

[0079] For example, this embodiment uses several specific question-and-answer scenarios to illustrate the workflow of the intelligent question-and-answer module (QA-Core).

[0080] Q: What is the minimum required concrete strength grade for the cutting edge of the caisson? How long does it need to be cured after pouring before sinking can begin? answer: According to the "Code for Construction of Caissons and Pneumatic Caissons GB / T 51130-2016": ① The minimum strength grade of the cutting edge concrete must be C30. If the caisson depth exceeds 50m or the geological layer is gravel, it should be increased to C35; ② After the cutting edge concrete is poured, the standard curing time shall not be less than 14 days, and the compressive strength of the test block cured under the same conditions must reach 100% of the design strength before the sinking construction can be started.

[0081] (Reference: GB / T 51130-2016 5.2.3, 6.1.4). Q: What precautions should be taken when excavating a caisson in a pebble layer? A: (Clarification and follow-up question): Are you asking about the selection of excavation equipment, or about soil extraction control? A. Selection of excavation equipment B. Control of soil extraction depth Question (supplementary): A.

[0082] answer: The main challenges in caisson excavation in pebble-soil layers are low excavation efficiency and high wear on the cutting edges. Regarding the selection of excavation equipment, the following measures are recommended for caisson construction: Select heavy-duty grabs: Heavy-duty hydraulic grabs or grabs with crushing functions should be used to overcome the high density of the gravel layer.

[0083] Avoid using suction equipment: Pure air suction or water jet suction equipment is not suitable for large-diameter pebble layers.

[0084] Combined with blasting or soil loosening: For densely cemented pebble layers, it is advisable to combine micro-blasting or high-pressure water jetting techniques to loosen the soil without affecting structural safety.

[0085] Wear monitoring: During construction, the grab teeth and cutting edges should be inspected more frequently to prevent loss of control of the sinking posture due to severe wear.

[0086] (Reference: Construction Organization Plan for the Caisson Project of the North Channel Bridge of the Zhangjinggao Yangtze River Bridge (August 15, 2023)). The solution generation module is used to extract content from a domain-specific knowledge base based on user-input parameters and populate it into a preset template to generate a construction technology solution containing text and multimodal drawings.

[0087] Specifically, the scheme generation module solves the problems of time-consuming manual scheme preparation, limitations due to personal experience, and the fact that traditional large models only support text generation and lack multimodal information such as drawings and flowcharts. By comprehensively considering multiple factors such as geological conditions, hydrological conditions, structural form, and construction equipment input by the user, it generates practical schemes for specific working conditions, achieving integrated output of "text-charts-drawings".

[0088] Furthermore, in some preferred embodiments of the present invention, the scheme generation module specifically includes: a template driving unit, which has a built-in standardized template library covering the entire process of caisson construction; the standardized template library includes at least the following standardized templates: construction organization design template, special construction scheme template, monitoring and control template, and risk management template; wherein, special construction includes at least one of the following: caisson fabrication, caisson sinking, caisson bottom sealing, and real-time correction during the caisson sinking process; a parameter mapping unit, which is used to automatically map the geological conditions, hydrological parameters, caisson geometric dimensions, and construction equipment requirements input by the user to template variables, and call the domain-specific knowledge base content in the knowledge base construction module, and use a large language model to fill the template content in a targeted manner; and a drawing rendering unit, which is used to dynamically generate caisson structure diagrams, construction step sequence diagrams, and monitoring point layout diagrams based on scalable vector graphics and network graphics library technology.

[0089] Specifically, the template-driven unit's built-in standardized template library covers the entire process of caisson construction. After the user selects a scheme type, the system automatically recalls the corresponding outline directory and parameter list. The parameter mapping unit maps key information such as caisson dimensions, geological conditions, and construction environment input by the user to template variables, and intelligently fills in matching technical requirements, process flows, material and equipment lists, and safety control points from the knowledge base. The drawing rendering unit, based on SVG and WebGL technology, dynamically generates multimodal drawings such as caisson structure drawings with reinforcement and reserved hole markings, phased construction sequence diagrams, and sensor deployment location diagrams based on parameters.

[0090] Furthermore, in some preferred embodiments of the present invention, the solution generation module is also used to feed the user-confirmed technical solution back to the knowledge base construction module as structured document data, so as to trigger incremental updates of the domain-specific knowledge base and form an application closed loop from solution generation to knowledge accumulation.

[0091] Specifically, the feedback mechanism allows user-confirmed excellent solutions to be written into the knowledge building module as new "original documents," triggering incremental index updates. This enables the system to continuously learn from practice, forming a virtuous cycle from "knowledge learning" to "practical application" and then to "knowledge feedback," constantly enriching and optimizing the knowledge base content and providing more referable practical cases for subsequent question answering and solution generation.

[0092] Furthermore, in some preferred embodiments of the present invention, the scheme generation module is also used to export the generated construction technology scheme and supporting drawings in multiple formats such as portable document format, engineering drawing format, and industrial basic format.

[0093] Specifically, multi-format output support ensures the usability of system-generated content in actual projects. Portable document formats guarantee universal readability, engineering drawing formats meet design modification needs, and industrial basic formats enable data exchange with mainstream BIM software such as Revit and ArchiCAD, truly achieving digital delivery and information-based construction management.

[0094] See Figure 4 The illustrated embodiment of the present invention provides a flowchart for generating a scheme. When a user needs to prepare a specific construction scheme, the user first enters the category and name of the scheme on the interface, such as "caisson fabrication construction scheme". The large model will immediately retrieve the pre-set standardized template based on the category name information, including the corresponding outline directory and the list of required parameters. After the user adjusts and confirms, the user is prompted to supplement key information such as caisson size, geological conditions, and construction environment. After the parameters are confirmed, the system will then accurately retrieve the technical requirements, process flow, material and equipment list and safety control points that completely match the input conditions from the knowledge base, automatically fill them into the reserved positions in the template, and output a structurally complete and personalized specific construction scheme text. The user can complete secondary adjustments and modifications and export online on the same page.

[0095] Furthermore, Plan-Gen can dynamically render and generate related multimodal drawings based on parameters, such as caisson structure diagrams and construction sequence diagrams, and supports export in multiple formats (such as PDF and DWG), truly achieving integrated output of "text and drawings".

[0096] Finally, once certain excellent solutions are identified and locked in, these solutions are written back into the K-Builder module as new "original documents," triggering incremental updates. This allows the system to continuously learn from practice, forming a virtuous cycle from "knowledge learning" to "practical application" and then to "knowledge feedback."

[0097] For example, this embodiment uses the example of on-site technicians needing to prepare a "caisson construction plan" to demonstrate the specific workflow and output results of the plan generation module (Plan-Gen).

[0098] User input: Compile the "Special Construction Plan for the Fabrication and Sinking of the #3 Main Pier Caisson".

[0099] System output (The Plan-gen system determines the required solution category as "caisson construction and sinking," calls the corresponding template from the built-in standardized template library, generates a preliminary directory structure of the solution, and prompts the user for confirmation and adjustments): [#3 Main Pier Caisson Fabrication and Sinking Special Construction Plan Catalog:] 1. Project Overview 2. Basis for compilation 3. Caisson structural parameters and geological conditions (Caisson structural parameters and geological parameters need to be supplemented) 4. Main construction methods for caisson fabrication 5. Selection of Caisson Sinking Technology and Parameters (Sinking technology and equipment need to be supplemented) 6. Caisson Correction Measures 7. Monitoring, Control and Risk Early Warning 8. Safety and environmental protection measures Confirm this directory? User input: Confirm System output: Please provide additional information regarding the caisson's structural and geological parameters, as well as the sinking process and equipment. User input (upload information form 1): Table 1

[0100] System output (excerpt from the complete solution): ... Chapter 5 Selection of Caisson Sinking Technology Considering the large size of the caisson and the presence of sand layers and confined water at its base, this plan adopts a heavy-duty hydraulic grab bucket combined with air suction for foundation clearing and soil removal to ensure efficient removal of pebbles and sand. To successfully cross the confined water layer, detailed dewatering or compressed air measures need to be developed in advance. Throughout the sinking process, over-excavation of the cutting edge is the main means of controlling the frictional resistance of the caisson wall and the sinking attitude; the over-excavation depth should be dynamically adjusted based on the actual frictional resistance monitoring values ​​and the sinking speed.

[0101] ... The accompanying drawings generated by the system: [Structural plan and section of the caisson (with reinforcement bars and markings of reserved holes)] Caisson construction phased construction sequence diagram Monitoring layout map during caisson sinking (GNSS and tilt sensor locations) The risk warning module is used to identify construction risks based on real-time monitoring data and trigger warnings and emergency responses.

[0102] Specifically, the risk warning module addresses major risk events commonly encountered during the caisson sinking process, such as "jamming," "tilting," "sand inrush," "piping," and "uneven settlement," resolving the shortcomings of existing emergency response methods that rely on manual judgment and lack intelligent analysis and dynamic response capabilities. By deeply integrating real-time monitoring data with a knowledge base, it enables early risk identification and rapid response.

[0103] Furthermore, in some preferred embodiments of the present invention, the risk warning module includes: a monitoring data access and warning unit, used to associate with the construction monitoring system, access real-time data of the positioning system, tilt angle, stress strain, elevation, and water level, and trigger a risk warning after at least one sensor indicator exceeds a preset threshold; a risk identification unit, used to comprehensively analyze the monitoring indicators exceeding the threshold and, in conjunction with the risk case data in the knowledge base construction module, identify the risks of caisson construction; and an emergency response plan generation unit, used, under user instruction, to drive the plan generation module to generate a structured emergency response plan containing a risk event description, emergency measures, and tracking procedures based on risk cases, specifications, and plan data in the knowledge base construction module.

[0104] Specifically, the monitoring data access and early warning unit connects to GNSS, tilt sensors, stress gauges, and other equipment to collect key indicators such as settlement velocity, wellbore friction, and attitude tilt angle in real time, with a sampling frequency of down to the second level. An early warning is immediately triggered when an indicator exceeds a threshold (e.g., settlement velocity > 50 mm / h, attitude tilt angle > 0.1°). The risk identification unit comprehensively analyzes multiple indicator anomaly patterns and, combined with risk cases in the knowledge base, quickly identifies risk types such as sudden subsidence, deflection, and structural cracking. After obtaining user confirmation, the emergency response plan generation unit drives the plan generation module to quickly generate an emergency response plan based on real-time sensor data and knowledge base content. This plan includes specific measures such as emergency shutdown instructions, restoration of cutting edge supports, and structural safety checks, achieving a response time within minutes.

[0105] For details, see Figure 5 The illustrated embodiment of the present invention provides a risk warning flowchart. The risk warning module (Risk-Warn) aims to extend the value of the knowledge base to real-time safety management at construction sites. As the system's "eyes" and "alarm system," it achieves real-time monitoring and early warning of potential risks through data interface with on-site sensors.

[0106] This module continuously receives real-time monitoring data from the construction monitoring system, including data from GNSS sensors, tilt sensors, and stress gauges. This data is analyzed in real time, and when any one or more indicators exceed a preset threshold, Risk-Warn immediately triggers an early warning mechanism, notifying relevant personnel via system pop-ups or other means.

[0107] Upon issuing an alert, Risk-Warn immediately invokes QA-Core to assess the current risk event. QA-Core searches its knowledge base for relevant risk cases and contingency plans related to the current anomaly (such as "sudden sinking" or "deviation") using metrics from the anomaly sensor data. It then synthesizes these pieces of evidence and outputs a judgment and recommendations regarding the current situation.

[0108] After receiving confirmation from user feedback, Risk-Warn invokes the Plan-Gen module and, combined with data from relevant sensors, quickly generates a preliminary emergency response plan, providing on-site personnel with scientific and timely decision-making advice. In this way, Risk-Warn achieves a seamless connection from "data monitoring" to "risk identification" and then to "decision support," elevating the value of the knowledge base from static querying to dynamic decision support, significantly improving construction safety.

[0109] For example, this embodiment uses the risk of "sudden sinking of a caisson" as an example to illustrate the specific implementation process of the risk warning module (Risk-Warn).

[0110] At a certain moment, the construction monitoring system detected an anomaly in the key indicators shown in Table 2 below. The data was transmitted to the system, and an early warning mechanism was triggered immediately: Table 2

[0111] System output: Based on preliminary assessments of monitoring data, the caisson has experienced sudden subsidence (or instantaneous settlement). The direct cause of this sudden subsidence is likely: The supporting force of the soil under the foot of the cutting edge (such as medium and fine sand) is lost instantly after over-excavation.

[0112] 2. After the frictional resistance between the well wall and the soil is overcome, the self-weight of the well and the water and soil pressure cause rapid settlement.

[0113] Key impact: Sudden sinking may cause abrupt changes in the stress on the caisson structure, leading to the risk of structural cracking or tilting.

[0114] Please confirm: Is it necessary for the system to generate a preliminary emergency response plan at this time? (Y / N?) The on-site personnel affirmed the system's assessment and confirmed the need for system support to generate a preliminary emergency response plan.

[0115] System output: [Preliminary Plan Outline for Emergency Response to Sudden Sinking] 1. Risk Event Description and Impact Analysis (Automatic Import of Monitoring Data) 2. Emergency Response Organization and Responsibilities 3. Emergency Response Measures for Sudden Sinking Incidents 4. Follow-up tracking and resumption of work procedures 5. Resource Guarantee "Please confirm that the above list meets the on-site requirements. Once confirmed, the content will be automatically populated to generate an emergency response plan." User input: Confirm System output (excerpt from the preliminary emergency response plan for sudden sinking): ... 3. Emergency Response Measures for Sudden Sinking Incidents Objective: Immediately stop the caisson's settlement, stabilize its attitude, and check its structural safety.

[0116] Operating procedures: Emergency Stop Order: The on-site commander shall immediately issue a stop order to halt all excavation, sludge removal, over-excavation, and dewatering operations within the well.

[0117] Restoring the support of the cutting edge: Quickly throw granular materials such as gravel or coarse sand near the cutting edge at the bottom of the well to rapidly restore the supporting force of the cutting edge on the soil and inhibit settlement.

[0118] Ballast balance check: Check the ballast condition of the caisson water tank to ensure that the ballast is evenly distributed. Any local counterweight adjustment that may increase the risk of tilting is strictly prohibited when the caisson is unstable.

[0119] Structural safety inspection: After stabilization, immediately conduct a comprehensive visual inspection and non-destructive testing of key parts such as the inner wall of the caisson, partition walls, and cutting edge, and record any cracks or damage.

[0120] Subsequent processing: After the structure and attitude have stabilized, prepare a detailed structural inspection report and a resumption of work and sinking plan, and submit them to the supervision and design units for approval.

[0121] ... Furthermore, in some preferred embodiments of the present invention, the system further includes: a history record module, used to write the original question, retrieval evidence chain, generated answer, cited text, user feedback and timestamp of each question-and-answer interaction into the database in an associated data format, and generate a unique session identifier; at the same time, it records risk warning events and generated solutions, forming a traceable record of the entire life cycle of system operation.

[0122] Specifically, the history module stores all interaction data in JSON-LD format, generating unique session IDs to support dialogue retracing. The front-end provides a "History Conversation" floating window, allowing users to easily resume conversations from any historical context. It supports four-dimensional searching by project, time period, keyword, and user ID, as well as single or batch export to PDF, Markdown, and Word formats. By recording user feedback on answers ("acceptance," "modification," or "rejection"), this module provides a data foundation for continuous fine-tuning and reinforcement learning, constantly optimizing model performance and reducing the hallucination rate. Simultaneously, the complete operation log provides reliable data support for engineering quality traceability and accident analysis.

[0123] This invention addresses the problems of knowledge dispersion and inefficient retrieval in the field of caisson construction. Addressing the issue that existing caisson construction knowledge is scattered across multiple heterogeneous sources such as specifications, drawings, logs, and expert experience, and that traditional keyword retrieval suffers from a "semantic gap" and low efficiency, this invention achieves a breakthrough through a knowledge base construction module (K-Builder). This module automatically collects heterogeneous data such as design drawings (DWG / PDF), technical specifications (e.g., "Construction Specification for Caissons and Pneumatic Caissons GB / T 51130-2016"), construction logs, and geological survey reports. After data cleaning and preprocessing, the information is structured using Transformer's Named Entity Recognition (NER) and Relation Extraction (RE) models to achieve semantic association. Simultaneously, a dual-storage architecture of "vector + knowledge graph" is adopted. The vector library supports millisecond-level semantic similarity recall, while the knowledge graph enables complex logical reasoning, avoiding missed and false detections in traditional retrieval. Combined with an incremental update mechanism using timestamps and version numbers, this ensures that the knowledge base is synchronized with engineering practice, resolving the lag in traditional static knowledge management.

[0124] This invention adapts general-purpose large-scale models to specific industries to minimize "AI illusions." Addressing the issues of insufficient knowledge coverage in the field of caissons, the susceptibility to "AI illusions," and the lack of traceability in existing general-purpose large-scale models, this invention constructs a reliable mechanism through an intelligent question-answering module (QA-Core): a basic BERT model is retrained based on caisson industry corpus to ensure accurate understanding of specialized terminology; a three-level hybrid retrieval system ensures comprehensive evidence, while an interactive clarification mechanism ensures high-quality answers; furthermore, all answers are generated based on knowledge base evidence and annotated with citation sources, allowing users to trace the original documents, significantly reducing the incidence of "AI illusions" and meeting the requirements of engineering rigor.

[0125] This invention intelligently generates multimodal construction plans. Addressing the problems of traditional manual caisson construction plans being time-consuming, limited by experience, and only capable of text output, this invention optimizes the process through a plan generation module (Plan-Gen). It includes built-in standardized templates for all processes, such as caisson fabrication, sinking, bottom sealing, and correction. After user input, the system automatically maps template variables, extracts matching technical requirements, process flows, and equipment lists from the knowledge base, and generates a plan adapted to specific working conditions. It dynamically generates visual charts such as caisson structure diagrams and construction sequence diagrams, solving the problem of abstract and difficult-to-understand traditional text-based plans. Excellent plans confirmed by the user are also fed back into the knowledge base, triggering incremental updates and forming a virtuous cycle of "plan generation - practical verification - knowledge accumulation."

[0126] This invention avoids delayed emergency response and completes a closed loop of risk warning and emergency decision-making. Addressing the problems of existing caisson construction risk management relying on manual judgment, disconnect between monitoring data and knowledge base, and slow emergency plan generation, this invention constructs a closed loop through a risk warning module (Risk-Warn) working in collaboration with other modules: It analyzes indicators in real time by linking with the construction monitoring system, providing rapid warnings when thresholds are exceeded, thus solving the problem of delayed risk detection. After an anomaly occurs, the system automatically calls the intelligent question-and-answer module (QA-Core) to identify the risk type using the risk knowledge base, and simultaneously triggers the plan generation module (Plan-Gen) to generate an emergency response plan based on real-time data. The entire process from "data monitoring → risk warning → type identification → plan generation" is automated, eliminating the need for manual review of specifications or waiting for expert guidance, enabling rapid response to risk events, significantly reducing the probability of risks escalating into accidents, and ensuring construction safety and progress.

[0127] This invention provides an AI knowledge base question-answering system for caisson construction, comprising: a knowledge base construction module, an intelligent question-answering module, a solution generation module, and a risk warning module; the intelligent question-answering module is connected to the knowledge base construction module via data; the solution generation module is connected to both the knowledge base construction module and the intelligent question-answering module via data; the risk warning module is connected to both the knowledge base construction module and the intelligent question-answering module via data, connected to an external construction monitoring system via signals, and connected to the solution generation module via instructions; the knowledge base construction module is used to collect, clean, structure, and store multi-source heterogeneous data in the field of caisson construction, constructing a domain-specific knowledge base including vector indexes and graph databases; the intelligent question-answering module is used to receive users' natural language questions and retrieve data from the domain-specific knowledge base through semantic parsing and hybrid retrieval. The system generates professional answers with citations based on evidence fragments; a solution generation module extracts content from a domain-specific knowledge base based on user input parameters and fills it into a preset template to generate construction technical solutions containing text and multimodal drawings; a risk warning module identifies construction risks based on real-time monitoring data and triggers warnings and emergency responses; an AI knowledge base question-and-answer system for caisson construction is built, including a knowledge base construction module, an intelligent question-and-answer module, a solution generation module, and a risk warning module. It integrates data from multiple construction fields to build a dedicated knowledge base and quickly and accurately identifies knowledge based on semantic parsing and hybrid retrieval, improving the efficiency of construction decision-making; relying on the solution generation mechanism to integrate historical data, it generates targeted solutions for major risk events, adapting to complex and ever-changing construction environments and reducing reliance on personnel.

[0128] Example 2 This invention also provides an electronic device for running the AI ​​knowledge base question-answering system for caisson construction provided in the above embodiments; see also Figure 6The diagram shown is a structural schematic of an electronic device provided by an embodiment of the present invention. The electronic device includes a memory 400 and a processor 401. The memory 400 is used to store one or more computer instructions, and the one or more computer instructions are executed by the processor 401 to run the AI ​​knowledge base question-and-answer system for caisson construction provided in the above embodiment.

[0129] Furthermore, Figure 6 The electronic device shown also includes a bus 402 and a communication interface 403. The processor 401, the communication interface 403 and the memory 400 are connected via the bus 402.

[0130] The memory 400 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 403 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 402 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0131] Processor 401 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 401 or by instructions in software form. Processor 401 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a readily available storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 400, and processor 401 reads information from memory 400 and, in conjunction with its hardware, completes the steps of the method described in the foregoing embodiments.

[0132] This invention also provides a computer-readable storage medium storing computer-executable instructions. When these computer-executable instructions are called and executed by a processor, the computer can execute instructions for an AI knowledge base question-and-answer system for caisson construction, which will not be elaborated further here.

[0133] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and / or device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0134] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0135] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An AI knowledge base question-answering system for caisson construction, characterized in that, include: The module includes a knowledge base construction module, an intelligent question answering module, a solution generation module, and a risk warning module. The intelligent question-answering module is connected to the knowledge base construction module via data; The scheme generation module is connected to the knowledge base construction module and the intelligent question answering module via data respectively; the risk warning module is connected to the knowledge base construction module and the intelligent question answering module via data respectively, connected to the external construction monitoring system via signal, and connected to the scheme generation module via command. The knowledge base construction module is used to collect, clean, structure, and store multi-source heterogeneous data in the field of caisson construction, and to build a domain-specific knowledge base that includes vector indexes and graph databases. The intelligent question-answering module is used to receive users' natural language questions, obtain evidence fragments from the domain-specific knowledge base through semantic parsing and hybrid retrieval, and generate professional answers with citations. The solution generation module is used to extract content from the domain-specific knowledge base based on user-input parameters and fill it into a preset template to generate a construction technology solution containing text and multimodal drawings. The risk warning module is used to identify construction risks based on real-time monitoring data and trigger warnings and emergency responses.

2. The AI ​​knowledge base question-answering system for caisson construction according to claim 1, characterized in that, The knowledge construction module includes: A multi-source data acquisition unit is used to automatically extract raw data from industry data; wherein, the industry data includes: design drawings, technical specifications, construction logs, construction organization designs, special construction plans, typical cases, industry papers and patents; the raw data includes: text, tabular and graphic data; A data cleaning unit is used to preprocess the raw data, wherein the preprocessing includes: removing redundant and erroneous information and performing format conversion; The structured processing and semantic standardization unit is used to transform unstructured text data into structured triples using a named entity recognition and relation extraction model, and to merge synonyms for caisson construction terminology; wherein the structure of the structured triple is <first entity, relation, second entity>. The dual storage unit of vector and graph is used to vectorize text fragments and store them in a vector index to achieve semantic recall, while importing the structured triples into a graph database to support logical reasoning. The incremental update unit is used to perform online incremental updates to the domain-specific knowledge base using a timestamp and version number mechanism.

3. The AI ​​knowledge base question-answering system for caisson construction according to claim 2, characterized in that, The intelligent question-answering module employs a multi-level hybrid retrieval strategy, which includes sequential vector retrieval, knowledge graph retrieval, and keyword full-text retrieval. The vector retrieval is used to perform semantic similarity matching in the vector index based on the semantic vector of the user's question; The knowledge graph retrieval is used to search the graph database for knowledge points that have preset relationships and attributes with entities in the user's question; The keyword full-text search is used for precise matching based on keywords; The results of the hybrid retrieval will be merged and reordered to form a final set of evidence fragments for answer generation; the intelligent question-answering module follows a working mechanism of retrieval priority followed by generation supplementation; all answers generated by the intelligent question-answering module are based on evidence fragments retrieved from the domain-specific knowledge base; the answers are accompanied by a citation tracing mechanism, including the normative clause number and / or construction case source of the cited knowledge.

4. The AI ​​knowledge base question-answering system for caisson construction according to claim 3, characterized in that, The intelligent question-answering module also includes: An interactive clarification unit is used to automatically generate two-option and / or open-ended clarification questions to guide users to supplement key information when the system's confidence level in a user's question is lower than a preset threshold. The confidence level is obtained by calculating the weighted score of four dimensions of signals between the recalled evidence fragments and the initially generated answers. The four dimensions of signals include: fragment coverage, answer overlap, relevance, and model self-evaluation probability.

5. The AI ​​knowledge base question-answering system for caisson construction according to claim 1, characterized in that, The scheme generation module specifically includes: The template driving unit has a built-in standardized template library covering the entire process of caisson construction; the standardized template library includes at least the following standardized templates: construction organization design template, special construction plan template, monitoring and control template, and risk management template; wherein, the special construction includes at least one of the following: caisson fabrication, caisson sinking, caisson bottom sealing, and real-time correction during the caisson sinking process; The parameter mapping unit is used to automatically map the geological conditions, hydrological parameters, caisson geometric dimensions and construction equipment requirements input by the user to template variables, and call the domain-specific knowledge base content in the knowledge base construction module to fill the template content in a targeted manner using a large language model; The drawing rendering unit is used to dynamically generate caisson structure diagrams, construction sequence diagrams, and monitoring point layout diagrams based on scalable vector graphics and network graphics library technology.

6. The AI ​​knowledge base question-answering system for caisson construction according to claim 5, characterized in that, The solution generation module is also used to feed the user-confirmed technical solution back to the knowledge base construction module as structured document data, so as to trigger the incremental update of the domain-specific knowledge base and form an application closed loop from solution generation to knowledge accumulation.

7. The AI ​​knowledge base question-answering system for caisson construction according to claim 1, characterized in that, The scheme generation module is also used to export the generated construction technology scheme and supporting drawings in multiple formats such as portable document format, engineering drawing format, and industrial basic format.

8. The AI ​​knowledge base question-answering system for caisson construction according to claim 1, characterized in that, The risk warning module includes: The monitoring data access and early warning unit is used to connect with the construction monitoring system, access real-time data from the positioning system, tilt angle, stress and strain, elevation, and water level, and trigger a risk warning when at least one sensor indicator exceeds a preset threshold. The risk identification unit is used to comprehensively analyze monitoring indicators that exceed the threshold and, in conjunction with the risk case data in the knowledge base construction module, identify the risks of caisson construction. The emergency response plan generation unit is used, under user instructions, to drive the plan generation module to generate a structured emergency response plan containing a risk event description, emergency measures, and tracking procedures based on the risk cases, standards, and plan information in the knowledge base construction module.

9. The AI ​​knowledge base question-answering system for caisson construction according to claim 1, characterized in that, The system also includes: The history module is used to write the original question, the evidence chain, the generated answer, the cited text, the user feedback, and the timestamp of each Q&A interaction into the database in a related data format, and generate a unique session identifier; at the same time, it records risk warning events and generated solutions, forming a traceable record of the entire life cycle of system operation.

10. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer-executable instructions that can be executed by the processor, the processor executing the computer-executable instructions to run the AI ​​knowledge base question-answering system for caisson construction as described in any one of claims 1 to 9.

Images