BIM (Building Information Modeling) visual security technology disclosure method based on large language model

Through large language model and BIM visualization technology, the potential risks in building construction are automatically identified and intuitively displayed, and the problems of insufficient information transmission and understanding deviation in traditional safety briefing methods are solved, achieving more efficient and accurate safety technical briefing.

CN118939769APending Publication Date: 2024-11-12广东建科创新技术研究院有限公司

Patent Information

Application Number
CN202410973108.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

Traditional construction safety briefing methods rely on manual experience, are time-consuming and labor-intensive, and are inadequately transmitted information, which can easily lead to risk identification omissions and understanding deviations, and lack intuitiveness.

Method used

Using BIM visual security technology briefing method based on large language model, by building a security risk database and custom prompt word templates using RAG technology, engineering features are extracted, potential risks are automatically identified, and they are visually displayed in the BIM model, combined with labeling, Color coded and 3D graphics.

Benefits of technology

It improves the efficiency, comprehensiveness and accuracy of safety technical briefing, helps construction workers to understand the risk sources more intuitively, and improves the efficiency and comprehensiveness of risk identification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118939769A_ABST
    Figure CN118939769A_ABST
Patent Text Reader

Abstract

A BIM visual security technology disclosure method based on a large language model comprises the following steps that 1, a security risk database is constructed, and the security risk database is formed by cutting and vectorizing general files related to security risks; 2) questioning the big language model based on the RAG technology by using the customized cue word template, and extracting engineering characteristics of the project from a special file related to an actual project by using the big language model; the cue word template is a combination of a series of summarized problems assisting in extracting the engineering characteristics of the project; 3) inputting the engineering characteristics into the large language model, and requiring the large language model to retrieve risks possibly caused by the engineering characteristics and corresponding processing measures in comparison with the security risk database; and 4) establishing a building model by using a BIM visualization technology, and enabling the extracted risk to correspond to a corresponding position in the building model. The safety technical disclosure efficiency and effect in the building construction process can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of construction engineering production safety management and artificial intelligence technology, and specifically relates to a BIM visualization safety technology disclosure method based on a large language model. Background Art

[0002] In traditional construction safety management, safety briefing is an important link, which aims to ensure that construction personnel understand and master the necessary safety knowledge and skills to prevent and reduce the occurrence of safety accidents. However, the traditional form of safety briefing is too simple, usually requiring manual analysis of risk sources, consulting specifications to formulate disposal measures, and then relying on verbal explanations or distributing some paper materials to conduct safety technical briefings for production personnel. On the one hand, the identification of safety risk sources in this way depends entirely on personal experience and cognitive scope, which is not only time-consuming and labor-intensive, but also may result in omissions and subjective biases. On the other hand, this method of briefing is not intuitive enough, which can easily lead to insufficient information transmission and deviations in understanding. Summary of the invention

[0003] The present invention aims to provide a method for improving the efficiency and effect of safety technology disclosure during the construction process.

[0004] The technical problem of the present invention is solved by the following technical solution: A BIM visualization safety technology disclosure method based on a large language model, comprising the following steps:

[0005] Step 1) constructing a safety risk database, wherein the safety risk database is formed by cutting and vectorizing common documents related to safety risks such as production safety management regulations, safety procedures, construction technical specifications, laws and regulations;

[0006] Step 2) Using a customized prompt word template to ask questions to the large language model based on RAG technology, the large language model is allowed to extract the engineering characteristics of the project from the survey and design documents, construction plan, major risk source list and other special documents related to the actual project; the prompt word template is a combination of a series of questions that are summarized to assist in extracting the engineering characteristics of the project;

[0007] Step 3) inputting the engineering features into the large language model, requiring it to retrieve the risks that may be caused by the engineering features and the corresponding treatment measures by comparing with the security risk database;

[0008] Step 4) Use BIM visualization technology to establish a building model, and correspond the extracted risks to corresponding positions in the building model.

[0009] Large Language Models (LLMs) are an important development in the field of artificial intelligence in recent years. These models can understand and generate natural language text through deep learning and natural language processing technology. With the improvement of computing power and the enrichment of data sets, large language models have shown great application potential, including text generation, semantic understanding, machine translation, etc. The present invention optimizes large language models based on RAG technology. The RAG mechanism is particularly suitable for scenarios that need to deal with continuous updating of information. Through RAG, large language models can directly access the latest information without retraining in order to generate more reliable outputs.

[0010] In the field of construction, the management of safety risks is of vital importance. Traditional risk identification relies on manual experience and cognitive scope. This method has limitations, such as low efficiency of information induction and is likely to be incomplete. In the present invention, by applying a large language model based on RAG technology, a large amount of text can be quickly analyzed, and potential risks can be automatically identified and extracted, which is conducive to improving the efficiency and comprehensiveness of risk identification. BIM visualization technology can be used to digitally express the physical and functional characteristics of a building, so that the participants of the construction project can intuitively see the content involved in the project. The present invention combines the risk information identified by the large language model with the BIM model to intuitively display the source of risk, which is conducive to enhancing people's recognition and understanding of the source of risk, thereby improving the efficiency, comprehensiveness and accuracy of safety technology briefings as a whole.

[0011] In addition, the present invention first refines the project characteristics, and then allows the large language model to search for risks and treatment measures based on the project characteristics, thereby avoiding the problem that the output results of the large language model are less comprehensive and less targeted due to reasons such as the lack of directionality in questions.

[0012] The prompt word template suggests exploring the engineering characteristics of the project from five dimensions: personnel, management, equipment and materials, construction technology, and environment. This organizational method is more conducive to ensuring the comprehensiveness of the collected engineering characteristics.

[0013] The specific steps for cutting and vectorization in step 1) are as follows:

[0014] Document cutting

[0015] The material document to be analyzed is cut into first-generation document blocks, and then the first-generation document blocks are cut into second-generation document blocks, so as to form a hierarchical structure with multiple layers (more than two layers) of parent-child relationships. All child document blocks under the same parent document block are aggregated to form the parent document block;

[0016] Vectorization

[0017] The document blocks at each level of the hierarchical structure are vectorized and stored in a vector database to construct a security risk database.

[0018] Step 3) belongs to the retrieval generation stage of RAG technology, which uses automatic merge retrieval optimization technology, as follows:

[0019] After the engineering characteristics of the project are extracted using a large language model based on RAG technology, they are vectorized and then used as an input source to calculate the correlation with the minimum sub-document block in the vector database. The n minimum sub-document blocks with the highest correlation are selected to determine whether the number or proportion of highly correlated sub-document blocks contained in their parent document blocks reaches a threshold. If so, the parent document block is provided to the large language model to identify potential risk sources in the project. Otherwise, the highly correlated sub-document blocks are directly provided.

[0020] If the hierarchical structure has a large number of levels, the parent document block that reaches the threshold may be taken as a new highly relevant document block to further determine whether a parent document block at a higher level should be provided to the large language model.

[0021] For large language models based on RAG technology, document segmentation is a commonly used technical route, which can improve the speed of vector calculation, but it may also cause the possibility of missing document block context information. Based on the traditional technical route, the present invention automatically determines whether to return a child document block or a merged parent document block, providing an optimal coordination solution in terms of computing speed and information richness.

[0022] The method, after identifying potential risks, categorizes and prioritizes these risks.

[0023] This step can help the construction team more clearly understand which risks are the most urgent and need to be focused on and dealt with. At the same time, by classifying risks, it can provide a basis for subsequent risk management strategies.

[0024] When the risk information extracted by the large language model is associated with the corresponding position in the BIM model in step 4), the method displays the risk source in the BIM model in different ways including annotation, color coding and three-dimensional graphics. The applicability of different display methods is as follows:

[0025] Marking: suitable for pointing out specific risk points;

[0026] Implementation method: Add text annotations at the corresponding locations of the BIM model to directly indicate the type of risk that exists there;

[0027] Color coding: suitable for distinguishing different types or levels of risks and dividing risk areas;

[0028] Implementation method: Assign different colors to different types of risks;

[0029] Three-dimensional graphics: suitable for displaying complex risk structures or dynamically changing risk situations;

[0030] Implementation method: Use three-dimensional graphics to represent the sources of risks and possible transmission paths.

[0031] Beneficial effects:

[0032] 1) The present invention designs a complete set of processes to realize intelligent analysis of project files, automatically extract risks and intuitively display them in the BIM model, so that construction personnel can understand safety requirements more intuitively, comprehensively and accurately, and is conducive to improving the efficiency and effectiveness of safety technology disclosure;

[0033] 2) The present invention first extracts the project characteristics, and then allows the large language model to search for risks and treatment measures based on the project characteristics, thereby avoiding the problem that the output results of the large language model are less comprehensive and targeted due to the lack of directionality of the questions;

[0034] 3) The present invention adopts the technology of automatic merge retrieval optimization on the traditional RAG technical route, and provides an optimal coordination solution in terms of operation speed and information richness;

[0035] 4) The present invention extracts engineering features by setting prompt word templates, which is conducive to more convenient and comprehensive feature extraction. In addition, the present invention explores the engineering features of the project from five dimensions: personnel, management, equipment and materials, construction technology, and environment. This organizational method can better ensure the comprehensiveness of the collected engineering features. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 The present invention is a flowchart of the main steps of the BIM visualization safety technology disclosure method based on the large language model. DETAILED DESCRIPTION

[0037] The present invention aims to improve the efficiency and effect of safety technology disclosure during construction, thereby providing a BIM visualization safety technology disclosure method or system based on a large language model, such as Figure 1 As shown, the method includes the following steps, or the system is configured to perform the following steps:

[0038] Step 1) Construct a safety risk database, which is formed by cutting and vectorizing common documents related to safety risks, such as production safety management regulations, safety procedures, construction technical specifications, laws and regulations, etc. This part belongs to the data preparation stage of RAG technology.

[0039] 1.1 Document cutting

[0040] The relevant documents such as the safety production management regulations, safety procedures, construction technical specifications, laws and regulations are cut according to the set hierarchical structure to form multiple document blocks with parent-child relationships. For example, the document is first cut into 1,000 tokens to form the first-generation child document blocks, and then the first-generation child document blocks are cut into 500 tokens to form the second-generation child document blocks. The lower-level document blocks are the child document blocks of their upper-level document blocks. The child document blocks at the same level are independent of each other. All the child document blocks under the same parent document block are aggregated to form its parent document block.

[0041] 1.2 Vectorization

[0042] The document blocks at all levels after cutting (such as the second-generation sub-document blocks formed in step 1.1, indicating that it has three levels) are vectorized, such as using an embedding model for vectorization, and stored in a vector database to form a security risk database.

[0043] Step 2) Use the customized prompt word template to ask questions to the large language model based on RAG technology, so that the large language model can extract the engineering characteristics of the project from the special documents related to the actual project, such as the survey and design documents, construction plan, list of major risk sources, etc.

[0044] The prompt word template is a combination of a series of questions that help extract the engineering characteristics of a project. The prompt word template suggests exploring the engineering characteristics of a project from five dimensions: personnel, management, equipment and materials, construction technology, and environment.

[0045] Among them, personnel-related matters include but are not limited to special operations, operations at heights, hot work, operations in confined spaces, temporary electricity use, etc.;

[0046] Management related information includes but is not limited to organizational structure, responsibility system, emergency plan, etc.;

[0047] Equipment and materials related include but are not limited to formwork (slip formwork, climbing formwork, flying formwork, tunnel formwork, concrete formwork, high-rise formwork), lifting and hoisting (tower cranes, construction elevators, material hoists, etc.), scaffolding (types of scaffolding, hanging baskets, unloading platforms, operating platforms), construction machinery (shield machines, pipe jacking machines, drilling machines, cranes, excavators, crushers, rollers, concrete pump trucks, bulldozers, concrete tankers, concrete placing machines, cutting machines, electric welders, tamping machines, etc.), flammable and explosive materials and items, etc.;

[0048] Construction technology related projects include but are not limited to foundation pit projects (deep foundation pit projects, basements), dark excavation projects, steel structure, grid and cable membrane structure installation projects, manual bored pile projects, concrete prefabricated component installation projects, high slope cutting projects, high side slope projects, high filling projects, demolition projects, etc.

[0049] Environmental factors include but are not limited to geological conditions (soft soil foundation), groundwater conditions, surrounding environment (roads, underground pipelines, rail transit, rivers), working environment (tunnels, civil air defense projects, high temperature, conductive dust, humidity), extreme weather, winter construction, flood season and rainy season, etc.

[0050] The prompt word template can be organized in many ways. It is not necessary to set questions completely according to the five categories provided above. You can establish a suitable classification system based on the actual project. For example, if you are only concerned about the new materials and new technologies used in the project, you can set questions from the two perspectives of equipment materials and construction technology. However, generally speaking, exploring the engineering characteristics of the project from the five perspectives recommended in the present invention can collect engineering characteristics more comprehensively.

[0051] Step 3) Input the engineering features into the large language model and require it to search the risks and corresponding treatment measures that may be caused by the engineering features against the security risk database. This part belongs to the retrieval generation stage of the RAG technology. In particular, this stage adopts the automatic merge retrieval optimization technology.

[0052] Automatic merge search optimization

[0053] After extracting the engineering characteristics of the project using the large language model based on RAG technology, it is vectorized and then used as an input source to calculate the relevance with the minimum sub-document block in the vector database. The n minimum sub-document blocks with the highest relevance are selected to determine whether the number or proportion of highly relevant sub-document blocks contained in their parent document blocks reaches a threshold. If so, the parent document block is provided to the large language model for identifying potential risks and corresponding treatment measures in the project. Otherwise, the highly relevant sub-document blocks are directly provided. If there are many levels cut out, it can be further determined whether the upper-level document blocks should be provided to the large language model, and the judgment logic is similar.

[0054] For large language models based on RAG technology, document segmentation is a commonly used technical route, which can improve the speed of vector calculation, but it may also cause the possibility of missing document block context information. Based on the traditional technical route, the present invention automatically determines whether to return a child document block or a merged parent document block, providing an optimal coordination solution in terms of computing speed and information richness.

[0055] As a preferred solution, the present invention classifies and prioritizes the potential risks after identifying them. This step can help the construction team more clearly understand which risks are the most urgent and need to be focused on and dealt with. At the same time, by classifying the risks, it can provide a basis for subsequent risk management strategies.

[0056] Traditional risk source identification methods mainly rely on manual reading of construction plans, empirical analysis, and review of historical data, which results in low efficiency. The present invention utilizes the rapid processing capabilities of large language models to analyze large amounts of text information, automatically identify and extract potential risks and corresponding treatment measures, which not only improves the efficiency of risk identification, but also helps ensure comprehensiveness.

[0057] In addition, the present invention first refines the project characteristics, and then allows the large language model to search for risks and treatment measures based on the project characteristics, thereby avoiding the problem that the output results of the large language model are less comprehensive and less targeted due to reasons such as the lack of directionality in questions.

[0058] Examples of steps 1) to 3) above:

[0059] like Figure 1 As shown in the step 1, the safety risk database contains many descriptions of the risks that may be caused by soft soil and the corresponding treatment measures.

[0060] like Figure 1 As shown in step 2, according to the environment-related question setting in the prompt word template, ask the large language model (LLM): What are the geological characteristics of the foundation of this project? LLM answers: The base of this project is mainly thick fluidized silt soil, which belongs to soft soil foundation. This is one of the engineering characteristics of the project. Subsequently, the large language model can extract the possible risks and corresponding treatment measures in the safety risk database based on the engineering characteristics, such as Figure 1 The steps are shown in three parts.

[0061] Step 4) Building a building model using BIM visualization technology and mapping the extracted risks to corresponding locations in the building model. This step includes:

[0062] BIM model construction: Based on the specific content of the construction plan, use BIM software to build a three-dimensional model containing information such as building structure, construction equipment and working environment.

[0063] Risk source data integration: The risk information obtained from the large language model analysis is matched with the specific locations in the BIM model to form a complete list of risk sources.

[0064] Combining risk sources with buildings allows for intuitive display of risk sources, which helps improve the accuracy of people's knowledge and understanding of risk sources. More importantly, construction workers can quickly, clearly and comprehensively understand the possible risks at different locations of the building model, thereby achieving better safety briefing results.

[0065] In addition, when the risk information extracted from the large language model is associated with the corresponding position in the BIM model, the risk source can be displayed in the BIM model in a variety of different ways, such as annotation, color coding or three-dimensional graphics. The applicability of different display methods is as follows:

[0066] Marking: Applicable to pointing out specific risk points, such as a specific component, equipment or construction area.

[0067] Implementation method: Add text annotations to the corresponding locations of the BIM model to directly indicate the risk type at that location, such as "high fall risk area", "electrical hazard", etc. This method is simple, intuitive and easy to understand.

[0068] Color coding: suitable for distinguishing different types or levels of risks, and dividing risk areas.

[0069] Implementation method: Assign different colors to different types of risks, such as red for high-risk areas, yellow for medium-risk areas, and blue for low-risk areas. Through the change of colors, construction workers can quickly identify risk areas and risk levels.

[0070] Three-dimensional graphics: suitable for displaying complex risk structures or dynamically changing risk situations.

[0071] Implementation method: Use three-dimensional graphics, such as arrows, icons or other symbols, to indicate the source of risk and possible transmission path. For example, a red arrow can be used to show the flow from low-risk foundation pit drainage to high-risk foundation pit inrush.

[0072] Using the BIM model with risk source annotation, the construction personnel are given safety briefings. The key to this step is:

[0073] Interactive safety education: Through the interactive function of the BIM model, construction personnel can understand the location, nature and possible impact of risk sources from multiple angles and levels.

[0074] Detailed safety regulations explanation: Combined with the detailed safety regulations and operating guidelines provided by the large language model, comprehensive safety education is provided to construction workers to ensure that they understand and can comply with all safety requirements.

[0075] Simulated emergency drills: Simulate various possible safety accident scenarios in the BIM model, allowing construction workers to conduct emergency response drills in a virtual environment to improve their emergency handling capabilities.

[0076] It has been verified that the method or system of the present invention is more effective than the traditional manual Methods, efficiency and effects can be greatly improved.

Claims

1. A BIM visualization safety technology disclosure method based on a large language model, characterized in that: The steps include: Step 1) constructing a security risk database, wherein the security risk database is formed by cutting and vectorizing common files related to security risks; Step 2) using a customized prompt word template to ask questions to the large language model based on RAG technology, so that the large language model can extract the engineering characteristics of the project from the dedicated files related to the actual project; the prompt word template is a combination of a series of questions that are summarized to assist in extracting the engineering characteristics of the project; Step 3) inputting the engineering features into the large language model, requiring it to retrieve the risks that may be caused by the engineering features and the corresponding treatment measures by comparing with the security risk database; Step 4) Use BIM visualization technology to establish a building model, and correspond the extracted risks to corresponding positions in the building model.

2. The BIM visualization safety technology disclosure method according to claim 1 is characterized in that: The general documents include production safety management regulations, safety regulations, construction technical specifications, laws and regulations.

3. The BIM visualization safety technology disclosure method according to claim 1 is characterized in that: The special documents include the survey and design documents, construction plans, and lists of major risk sources for specific projects.

4. The BIM visualization safety technology disclosure method according to claim 1 is characterized in that: The prompt word template explores the engineering characteristics of the project from five dimensions: personnel, management, equipment and materials, construction technology and environment.

5. The BIM visualization safety technology disclosure method according to claim 1 is characterized in that: The specific steps for cutting and vectorization in step 1) are as follows: Document cutting The material document to be analyzed is cut into first-generation document blocks, and then the first-generation document blocks are cut into second-generation document blocks, so as to form a hierarchical structure with multiple layers of parent-child relationships. All child document blocks under the same parent document block are aggregated to form the parent document block; Vectorization The document blocks at each level of the hierarchical structure are vectorized and stored in a vector database to construct a security risk database.

6. The BIM visualization safety technology disclosure method according to claim 5 is characterized in that: Step 3) belongs to the retrieval generation stage of RAG technology, which uses automatic merge retrieval optimization technology, as follows: After the engineering characteristics of the project are extracted using a large language model based on RAG technology, they are vectorized and then used as an input source to calculate the correlation with the minimum sub-document block in the vector database. The n minimum sub-document blocks with the highest correlation are selected to determine whether the number or proportion of highly correlated sub-document blocks contained in their parent document blocks reaches a threshold. If so, the parent document block is provided to the large language model to identify potential risk sources in the project. Otherwise, the highly correlated sub-document blocks are directly provided.

7. The BIM visualization safety technology disclosure method according to claim 6 is characterized in that: If the hierarchical structure has a large number of levels, the parent document block that reaches the threshold is taken as a new highly relevant document block to further determine whether a parent document block at a higher level should be provided to the large language model.

8. The BIM visualization safety technology disclosure method according to claim 1 is characterized in that: The method, after identifying potential risks, categorizes and prioritizes these risks.

9. The BIM visualization safety technology disclosure method according to claim 1 is characterized in that: When the risk information extracted by the large language model is associated with the corresponding position in the BIM model in step 4), the method displays the risk source in the BIM model in different ways including annotation, color coding and three-dimensional graphics. The applicability of different display methods is as follows: Marking: suitable for pointing out specific risk points; Implementation method: Add text annotations at the corresponding locations of the BIM model to directly indicate the type of risk that exists there; Color coding: suitable for distinguishing different types or levels of risks and dividing risk areas; Implementation method: Assign different colors to different types of risks; Three-dimensional graphics: suitable for displaying complex risk structures or dynamically changing risk situations; Implementation method: Use three-dimensional graphics to represent the sources of risks and possible transmission paths.

Citation Information

Patent Citations

  • Project construction safety risk source identification method based on BIM visualization

    CN117852883A

  • Construction safety risk prompting method and system based on BIM and large language model

    CN118071153A

Cited By

  • Material attribute acquisition method and system based on large language model and BIM

    CN119516287A

  • Image recognition modeling method and system

    CN121765817A