A management method of a BIM model component library and coding

Abstract

The application provides a BIM model component library and coding management method, relating to the field of digital construction. The method adopts a web page and plug-in dual-end architecture, realizing integrated integration of component library semantic alignment, coding configuration and intelligent recommendation. Through digital model separation technology, component information and model files are independently stored and associated by a unique ID, a multi-stage coding system including design, construction and completion stages is constructed, ensuring stage data isolation and independent update. A heterogeneous graph neural network is introduced to construct a heterogeneous graph structure containing component types, functional relationships and stage attributes, and through a double attention mechanism, high-order semantic relationships between components are mined to realize intelligent component recommendation, automatic coding assignment and consistency checking. Dynamic data synchronization and version control are supported, effectively solving problems such as unclear component relationships, unintelligent coding recommendation and multi-stage inconsistency in traditional BIM systems, improving BIM modeling efficiency and data standardization level, and providing intelligent support for the digital transformation of the construction industry.

Description

Technical Field

[0001] This invention relates to the field of digital construction, specifically a management method for a BIM model component library and its coding. Background Technology

[0002] With the widespread application of Building Information Modeling (BIM) technology in the design, construction, and operation and maintenance phases of building engineering, the utilization rate of BIM models has significantly increased. In the process of BIM application, component libraries, as the fundamental resources for model construction, have a significant impact on project quality and collaborative efficiency due to their standardization and efficient management. Unified component coding is key to achieving information interconnection, data sharing, and full lifecycle management. However, currently, most BIM component libraries and coding systems are independent, lacking integrated management methods. This leads to problems such as inconvenient component searching, inconsistent coding, and high maintenance costs, limiting the overall application efficiency of BIM technology.

[0003] Currently, most common component libraries and coding plugins on the market are two separate plugins. Component libraries can only call components, but cannot assign codes to the called components; coding plugins can only code, but cannot call components. This separation of functions makes component searching inconvenient. Personnel at different stages, such as design and construction, find it difficult to quickly locate the required components from a large and scattered pool of resources. Furthermore, different coding methods are used in each stage, leading to confusion during data transfer and integration. In addition, separate component libraries and coding systems require separate investment of manpower and resources for management and updates, significantly increasing maintenance costs and limiting the overall application efficiency of BIM technology. Therefore, there is an urgent need for an intelligent tool that integrates component library management and coding system management to improve BIM modeling efficiency and data standardization. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a management method for BIM model component library and coding, which solves the semantic mismatch problem in the cross-stage evolution process of existing BIM models, realizes efficient management and standardized coding of components, and improves BIM modeling efficiency and data standardization management level.

[0005] The solution provided by this invention is a management method for BIM model component libraries and codes, comprising the following steps:

[0006] S1. Establish a dual-end collaborative architecture. Configure the Heterogeneous Graph Neural Networks (HGNN) model and multi-stage encoding rules on the web page. The plug-in end is integrated into the BIM software and synchronizes the parameters on the web page. The two ends interact in real time through API. S2. Component feature extraction and node mapping: Extract spatial topological and semantic features of components from the BIM model, and map and associate the features with the corresponding component nodes in the heterogeneous graph structure through a unique ID code. S3. Construction and definition of heterogeneous graph structure: Based on the physical connections and logical dependencies between components, establish polymorphic edges and introduce stage attribute nodes to construct a heterogeneous information network. S4. Semantic feature fusion: Capture cross-stage semantic features through meta-path, extract and fuse features using a dual attention mechanism network, and calculate the matching probability between component features and multi-stage coding system. S5, cross-stage consistency alignment and dynamic update, based on matching probability triggering automatic encoding assignment, dynamic update for local model changes, to achieve semantic alignment and version traceability of cross-stage data.

[0007] Preferably, in step S2, spatial topological features and non-geometric attributes are extracted from the BIM model database and component information database, respectively, and mapped together as multi-class node features in a heterogeneous information network using a unique ID code. The method also includes a data-model separation technique that logically links the component information database, BIM model database, and heterogeneous graph database using a unique ID code. Specifically, the component information database stores classification, attribute, and version information; the BIM model database stores geometric and topological data; and the heterogeneous graph database stores node, edge, meta-path, and high-order relation features learned by the HGNN. These three databases are stored independently and updated synchronously.

[0008] Preferably, in step S3, the construction of the heterogeneous graph structure includes: Treat BIM components, codes, project phases, professional fields, and relationships as different types of nodes; Physical connections, logical dependencies, stage affiliation, and professional affiliation relationships between components are treated as different types of edges.

[0009] Preferably, in step S4, the metapath includes: Phase flow path: Design phase - Dependency edge - Construction phase - Dependency edge - Completion phase, used for consistency verification of multi-phase coding; The “component-encoding-stage” path: component-binding edge-encoding-stage affiliation edge-stage, is used for automatic recommendation of stage encoding. Based on the “component-encoding-stage” path, the dual attention network can output the encoding with the highest matching degree with the current stage attribute and the corresponding probability according to the feature mapping relationship of the component in different stages. The “Specialty-Component-Connection” path is: Specialty-Specialty Affiliation Edge-Component-Connection Edge-Component-Specialty Affiliation Edge-Specialty, used for collaborative recommendation of cross-specialty components.

[0010] Preferably, in step S4, the dual attention mechanism network includes: a semantic-level attention module for learning the importance weights of different meta-paths for semantic reasoning of the target component; by performing fully connected transformation and softmax normalization on the aggregated features of each meta-path, calculating their attention weights and performing weighted fusion to obtain semantic-level fused features; and a node-level attention module for learning the influence weights of different neighboring nodes on the aggregation of target node features on the same meta-path; by calculating the difference between the target node features and the features of each of its neighboring nodes, obtaining the attention weights of each neighboring node through a fully connected layer and the LeakyReLU activation function, performing weighted aggregation and concatenating it with the target node's own features to obtain node-level fused features.

[0011] Preferably, in step S5, the dynamic update specifically involves: when the plug-in detects a local change in the BIM model, the change information is associated with the corresponding node and edge in the heterogeneous graph through a unique ID code, triggering HGNN to incrementally recalculate the node and edge weights of the affected local subgraph, and updating the relevant coding recommendation results.

[0012] Preferably, it also includes a consistency monitoring step: calculating the matching probability between the current code and feature of the component based on the dynamic update result; marking it as a suspicious code and issuing a warning when the matching probability is lower than a preset threshold; and tracing the cause of the anomaly based on the heterogeneous graph and recommending correction suggestions.

[0013] Beneficial effects of this invention: 1. Improved the overall efficiency of BIM modeling and data management. Through a dual-end collaborative architecture of web and plug-in terminals, component calling, coding rule configuration, intelligent recommendation and assignment are completed in one unified platform, avoiding the operation of switching between different independent systems and significantly reducing manual operation steps and time costs. 2. More accurate and intelligent component and code recommendation has been achieved. By introducing HGNN and dual attention mechanism customized for the BIM field, the system can deeply mine high-order semantic relationships such as spatial topology, professional affiliation and stage dependence between components, which greatly improves the accuracy and context relevance of component recommendation and code assignment. 3. Ensures the consistency and traceability of data across project phases. Based on the separation of data and model and the association of unique IDs, combined with a coding system and phase flow meta-path designed for multiple phases, it realizes independent storage, isolated updates and accurate synchronization of data at each phase, avoids data conflicts and semantic mismatch issues at multiple phases, and ensures the integrity and consistency of data throughout the entire lifecycle. 4. Reduced complexity and labor costs of system maintenance and data verification. The dynamic synchronization and incremental update mechanism reduces the need for full data recalculation, improves system response speed and reduces computational overhead. The system has automated coding consistency monitoring and anomaly warning functions, which can proactively detect and prompt suspicious codes and provide correction suggestions. Attached Figure Description

[0014] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a structural diagram of a BIM model component library and coding management method according to the present invention; Figure 2 This is a flowchart of the component uploading and digital-to-analog separation process of the present invention; Figure 3 This is a flowchart of the multi-stage coding system of the present invention; Figure 4 This is a flowchart of the heterogeneous graph neural network module of the present invention. Detailed Implementation

[0015] To better understand the purpose, system architecture, and functional implementation of this embodiment, the embodiments and features in the embodiments of this application can be combined with each other without conflict. The exemplary embodiments disclosed in this application will be described below with reference to the accompanying drawings, which include specific technical details disclosed in this embodiment to aid understanding; however, these details should be considered exemplary rather than restrictive.

[0016] Example 1 Figure 1 This is a structural diagram of a BIM model component library and coding management method in this invention.

[0017] like Figure 1 As shown, the management method 100 for BIM model component library and coding adopts a layered design, which is divided into method deployment, business logic and data collection, including steps S110~S130.

[0018] In operating the S110, global parameter configuration and HGNN model training are performed through the web interface, while real-time feature extraction and inference are executed through the plugin interface.

[0019] In operating S120, the business logic is mainly divided into four parts: the model-data separation module, the encoding and assignment module, the multi-stage management module, and the heterogeneous graph neural network module.

[0020] In operation S130, database construction mainly includes four types of databases: construction information database, BIM model database, coding rule base, and heterogeneous map database.

[0021] According to an embodiment of the present invention, in operation S110, the web page supports component library management, coding rule configuration, data query and historical version tracing, while the plug-in is integrated with mainstream BIM software such as Revit and MicroStation to realize automated processing of component calls and coding assignment, improve BIM modeling efficiency and collaboration quality. The integrated platform completes component calls and coding assignment in one step, reduces cross-system operation steps and improves efficiency. The introduction of unified coding rules and multi-stage coding systems (such as design coding, construction coding and as-built coding) ensures that personnel at different stages can quickly locate components and obtain accurate information, significantly reducing the risk of confusion in data transmission and integration.

[0022] According to an embodiment of the present invention, in operation S120, the model-data separation module in the business logic is responsible for storing component information and BIM model files independently and establishing a connection through a unique ID code. This design ensures that component information and model data do not interfere with each other. Even if the model changes, only the model data of the corresponding stage needs to be updated, while information in other stages remains stable. The coding and assignment module is based on a preset five-level coding rule and supports manual adjustment and batch assignment. The coding of each stage is bound to the model data, and information isolation is achieved through label switching, ensuring the independence and consistency of data throughout the entire lifecycle. The multi-stage management module supports independent coding and information isolation for the design, construction, and completion stages. When the BIM model changes, only the model data of the corresponding stage is updated, while information in other stages remains unchanged, avoiding data conflicts caused by model adjustments. The heterogeneous graph neural network module treats the entire BIM component library as a large-scale heterogeneous information network, and uses HGNN to learn and mine the deep and complex semantic relationships between these nodes in the heterogeneous information network.

[0023] According to an embodiment of the present invention, in operation S130, the component information database stores the component entry information table, including data such as classification, attributes, and version. This information is used for semantic alignment and querying of components, ensuring the standardization and normalization of components. The BIM model database stores BIM model files, including geometric data and topological relationships. The BIM model database and the construction information database are associated through a unique ID code to achieve separation of data and model. The coding rule library stores coding rules, supporting manual adjustment and batch assignment. The coding rule library ensures the consistency between component information and coding data, improving the level of data standardization. The heterogeneous graph database stores nodes, edges, meta-paths, and relational features learned by the HGNN, used to implement the training process of the heterogeneous graph neural network module, and stores the output results of the heterogeneous graph neural network module.

[0024] Figure 2 This is a flowchart of the component uploading and digital-to-analog separation process of the present invention.

[0025] like Figure 2As shown in Flowchart 200 of the component upload and model separation process, when a user uploads a component, they must simultaneously submit a component entry information table, including data such as classification, attributes, and version, as well as a BIM model file. This serves as the basic input for subsequent processing. The uploaded component entry information table and BIM model file are parsed to prepare for generating a unique association identifier. A unique ID code is generated through model separation technology to establish the association between component information and model files. By using the ID code, component information and BIM model files are stored in the component information database and BIM model database respectively, achieving a model architecture where component information and model data are stored independently yet related. The separated data is verified and synchronized to ensure data consistency and accuracy between the two. Simultaneously, the ID code is not only the primary key of the database but also a node identifier in the heterogeneous graph. Through a dynamic synchronization mechanism, when the model's attributes change on the plug-in side, the update vector is sent back to the HGNN in real time via the ID code, triggering the recalculation of the weights of the local subgraph. This incremental update mechanism avoids the computational redundancy of full model recalculation and achieves second-level alignment of cross-stage data.

[0026] Figure 3 This is a flowchart of the multi-stage coding system of the present invention.

[0027] like Figure 3 As shown, a five-level coding planning structure is adopted. Project coding is used to classify and encode projects in a basic way. Professional coding is used to encode projects according to different professional fields. Section coding further subdivides projects into different parts for coding. Component type coding encodes various component types in the project. Sequence code assigns a unique sequence number to each specific component or part, forming a complete coding hierarchy system and providing a unified rule framework for coding at each stage.

[0028] Multi-stage coding binding is implemented, generating design stage codes, construction stage codes, and completion stage codes for the same component based on a five-level coding plan. The design stage code is bound to the design stage model and parameters, the construction stage code is bound to the construction stage model and parameters, and the completion stage code is bound to the final acceptance model and actual data. The same component is associated with the three sets of codes through a unique ID, enabling multi-stage data coexistence. Users can switch stages by tags, and the system automatically loads the corresponding codes and models.

[0029] The design model, construction model, and as-built model are three specific models generated based on the above coding and association. It supports switching between different stages of coding and model data by tag. When the model changes, only the data of the current stage is updated to avoid cross-stage information conflicts caused by model changes. The models of each stage are physically stored separately and are associated with the coding only by a unique ID. After the coding or model is updated, it is synchronized to all associated modules in real time by the unique ID.

[0030] Figure 4 This is a flowchart of the heterogeneous graph neural network module of the present invention.

[0031] like Figure 4 As shown, based on a dual-end architecture of web-based and plugin-based architecture, a heterogeneous graph neural network module is added to the existing component library and coding management module to achieve intelligent recommendation and coding optimization. HGNN parameter configuration, HGNN inference, and a heterogeneous graph database are added to the existing interaction layer, logic layer, and data layer. In the BIM component library, multiple types of data are abstracted into a heterogeneous graph. Complex relationships between components are captured through node, edge, and meta-path definitions. Node types include component nodes, coding nodes, stage nodes, professional nodes, and relationship nodes. Component nodes include geometric features, attribute features, and version features; coding nodes include a five-level coding structure and stage labels; stage nodes include stage attributes such as time intervals, participants, and business rules; professional nodes include professional categories and domain rules; and relationship nodes include connection relationships and dependency relationships. The semantic relationships between nodes are distinguished by edge type, including binding edges, stage affiliation edges, professional affiliation edges, connection edges, and dependency edges. Binding edges represent the binding relationship between a component and its corresponding unique ID code; stage affiliation edges indicate that a component / code belongs to a certain project stage; professional affiliation edges indicate that a component belongs to a certain professional field; connection edges represent the physical connection relationship between components; and dependency edges represent the dependency relationship between project stages. Metapaths are paths connecting different types of nodes in a heterogeneous graph, used to capture higher-order relationships across stages and professions. Core metapaths include stage flow paths, "component-code-stage" paths, "profession-component-connection" paths, and coding rule paths. The path structure, semantic meaning, and application scenarios of metapaths are shown in Table 1 below.

[0032] Table 1 Metapath Definition Table

[0033] Based on the defined heterogeneous graph structure, relationships between components are learned through semantic-level and node-level attention mechanisms. This includes data preprocessing and feature initialization. Data preprocessing involves feature extraction of component node features, coded node features, and stage / professional node features. Component node features are extracted from the BIM model library for geometric features and from the component library for attribute features. Coded node features are decomposed into five fields based on the five-level coding, with each field represented by an embedding vector. Stage labels are encoded using one-hot encoding. Stage / professional node features use predefined embedding vectors based on domain knowledge. Graph structure initialization stores node features, edge types, and meta-paths in the heterogeneous graph database, using an adjacency matrix to record the connections between nodes. Initial weights are assigned based on the rules defined in the edge type definition.

[0034] The semantic-level attention mechanism primarily learns the importance of meta-paths, studying the weights of different meta-paths in component relationship reasoning. For each meta-path, it calculates the aggregated features of nodes along the path, constructing a meta-path attention network. The aggregated features of all meta-paths are input, and the attention weights of each meta-path are output through fully connected layers and a softmax activation function. Finally, the aggregated features of each meta-path are weighted and summed according to their attention weights to obtain the semantic-level fused features. The node-level attention mechanism primarily learns the importance of neighboring nodes, studying the influence of different neighboring nodes on the target node within the same meta-path. It iterates through all neighboring nodes of the target node, extracting the differences between neighboring features and target node features, constructing a node attention network. First, the feature differences are input, and the attention weights of neighboring nodes are calculated through fully connected layers and a LeakyReLU activation function. The neighboring features are weighted and summed according to their attention weights, and then concatenated with the target node's own features to obtain the node-level fused features.

[0035] The semantic alignment of the BIM component library is achieved through a heterogeneous graph neural network module, and intelligent coding recommendation is implemented on the plug-in side. When a user calls a component in Revit / MicroStation, the plug-in side completes feature extraction by obtaining the set features, stage, and discipline of the current component in real time, locates the component node in the heterogeneous graph, retrieves the associated meta-path to complete the graph query, inputs the component features and meta-path, and the model outputs the matching code to complete HGNN inference. Combined with a five-level coding rule base, codes that do not conform to the current stage / discipline are filtered out, and the final recommendation result is returned. After the user confirms, the plug-in side writes the recommended code into the component attributes and updates it synchronously to the web database.

[0036] To ensure consistency across stages, multi-stage coding is optimized. First, for the coding differences between the design, construction, and completion stages, HGNN is used to achieve dynamic coding adaptation. Based on the meta-path of the stage transition path, HGNN learns the mapping relationship between codes of different stages. When a component is transferred from the design stage to the construction stage, the model automatically recommends the construction code and marks the difference fields that need to be manually confirmed. If the model changes during the construction stage, HGNN updates only the construction code and associated nodes through the "component-code-stage" path, while the design / completion codes remain unchanged, thus avoiding cross-stage conflicts.

[0037] By leveraging the collaboration between the web interface and the plug-in interface for intelligent component recommendation, when users search for components on the web interface or model them on the plug-in interface, the system recommends related components based on HGNN. When HGNN passes through the meta-path of "profession-component-connection", it mines frequently paired components and adjusts the recommendation weights based on the user's historical call records. The connection relationship of the recommended components is displayed in a graph on the web interface, and the location of the recommended components is highlighted in the BIM software on the plug-in interface.

[0038] For code consistency verification, in the web-based review module, HGNN assists in detecting coding anomalies. The model calculates the matching probability between the current code and the component features and stage attributes. If it is lower than the threshold, it is marked as a suspicious code. The cause of the anomaly is traced through the "code-stage-component" path, and the corrected code is recommended based on historical correct cases.

[0039] Finally, dynamic synchronization and version control are implemented. When a new component is uploaded or the coding rules are modified, the heterogeneous graph database automatically updates node / edge information, triggering incremental training of the HGNN to update the heterogeneous graph structure. The coding rules and HGNN parameters configured on the web interface are synchronized to the plugin interface in real time via API, ensuring consistent recommendation results on both ends and achieving dynamic synchronization. The heterogeneous graph state and HGNN model parameters are periodically saved, supporting regression to historical versions. Through the above process, this invention semantically aligns the heterogeneous graph neural network module with the BIM component library, significantly improving component recommendation accuracy and coding consistency verification efficiency, significantly reducing manual intervention costs, and providing intelligent support for BIM applications throughout the building lifecycle.

[0040] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0041] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A management method for BIM model component library and coding, characterized in that, Includes the following steps: S1. Establish a dual-end collaborative architecture. The web page is configured with the HGNN model and multi-stage coding rules, while the plug-in is integrated into the BIM software and synchronizes the parameters of the web page. The two ends interact in real time through API. S2. Component feature extraction and node mapping: Extract spatial topological and semantic features of components from the BIM model, and map and associate the spatial topological and semantic features with the corresponding component nodes in the heterogeneous graph structure through a unique ID code. S3. Construction and definition of heterogeneous graph structure: Based on the physical connections and logical dependencies between components, establish polymorphic edges and introduce stage attribute nodes to construct a heterogeneous information network. S4. Semantic feature fusion: Capture cross-stage semantic features through meta-path, extract and fuse features using a dual attention mechanism network, and calculate the matching probability between component features and multi-stage coding system. S5, cross-stage consistency alignment and dynamic update, based on matching probability triggering automatic encoding assignment, dynamic update for local model changes, to achieve semantic alignment and version traceability of cross-stage data.

2. The management method for BIM model component library and coding according to claim 1, characterized in that, In step S2, spatial topological features and non-geometric attributes are extracted from the BIM model database and component information database, respectively, and mapped together as multiple types of node features in a heterogeneous information network using a unique ID code.

3. The management method for BIM model component library and coding according to claim 2, characterized in that, It also includes digital model separation technology that uses a unique ID code to logically link the component information database, BIM model database and heterogeneous diagram database; The component information database stores classification, attribute, and version information; the BIM model database stores geometric and topological data; and the heterogeneous graph database stores the node, edge, meta-path, and high-order relation features learned by HGNN. The three databases are stored independently and updated synchronously.

4. The management method for BIM model component library and coding according to claim 1, characterized in that, In step S3, the construction of the heterogeneous graph structure includes: Treat BIM components, codes, project phases, professional fields, and relationships as different types of nodes; Physical connections, logical dependencies, stage affiliation, and professional affiliation relationships between components are treated as different types of edges.

5. The management method for BIM model component library and coding according to claim 1, characterized in that, In step S4, the metapath includes: Phase flow path: Design phase - Dependency edge - Construction phase - Dependency edge - Completion phase, used for consistency verification of multi-phase coding; The "component-coding-stage" path is: component-binding edge-coding-stage affiliation edge-stage, used for automatic recommendation of stage codes; The "Specialty-Component-Connection" path is: Specialty-Specialty Affiliation Edge-Component-Connection Edge-Component-Specialty Affiliation Edge-Specialty, used for collaborative recommendation of cross-specialty components.

6. The management method for BIM model component library and coding according to claim 5, characterized in that, Based on the "component-encoding-stage" path, the dual attention network can output the encoding with the highest matching degree to the attribute of the current stage and the corresponding probability, according to the feature mapping relationship of the component at different stages.

7. The management method for BIM model component library and coding according to claim 1, characterized in that, In step S4, the dual attention mechanism network includes: A semantic-level attention module is used to learn the importance weights of different meta-paths for semantic reasoning of the target component; The node-level attention module is used to learn the influence weights of different neighboring nodes on the feature aggregation of the target node on the same metapath.

8. The management method for BIM model component library and coding according to claim 7, characterized in that, The semantic-level attention module performs fully connected transformation and softmax normalization on the aggregated features of each meta-path, calculates their attention weights, and performs weighted fusion to obtain semantic-level fused features. The node-level attention module calculates the difference between the target node's features and the features of its neighboring nodes, obtains the attention weights of each neighbor through a fully connected layer and the LeakyReLU activation function, performs weighted aggregation, and concatenates it with the target node's own features to obtain the node-level fused features.

9. The management method for BIM model component library and coding according to claim 1, characterized in that, In step S5, the dynamic update is specifically as follows: when the plug-in detects a local change in the BIM model, it associates the change information with the corresponding node and edge in the heterogeneous graph through a unique ID code, triggers HGNN to incrementally recalculate the node and edge weights of the affected local subgraph, and updates the relevant coding recommendation results.

10. A management method for a BIM model component library and coding according to claim 1 or 9, characterized in that, It also includes a consistency monitoring step: calculating the matching probability between the current code and features of the component based on the dynamic update results; marking it as a suspicious code and issuing a warning when the matching probability is lower than a preset threshold; and tracing the cause of the anomaly based on the heterogeneous graph and recommending correction suggestions.

Images