Building construction data acquisition management method and system

Abstract

The invention relates to the technical field of data processing, and discloses a building construction data acquisition management method and system, and the device comprises the steps: S1, extracting the information of construction members from a building information model, and constructing a graph structure of member nodes and association relationship edges between the members; s2, collecting multi-source sensor data corresponding to the component, and performing time synchronization, format unification and unit conversion processing on the data to generate standardized data; and S3, constructing a multi-dimensional tensor model based on the standardized data, wherein the multi-dimensional tensor model is used for organizing data of the component at different time points, data modals and construction stage dimensions. By completing graph structure reasoning and node representation updating of a construction state and analyzing a residual error between an estimation tensor and original data, an edge connection relation or a perception configuration parameter in a graph structure is further dynamically adjusted, so that self-adaptive data acquisition management oriented to a whole construction period is realized.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for collecting and managing construction data. Background Technology

[0002] The construction data acquisition and management method and system is a data processing system for the entire construction process. The data acquisition and management method includes a series of standardized processes for acquiring, filtering, and integrating various types of information in the construction process. It is mainly used to comprehensively, in real time, and accurately collect multi-dimensional information such as equipment operating status, material usage, construction progress, quality inspection data, and safety monitoring data during the construction process. Through systematic management, it realizes the storage, classification, analysis, and sharing of data, providing data support for construction units in progress control, resource allocation, quality assurance, safety prevention, and decision-making, and helping to promote the construction process more scientifically and efficiently.

[0003] The discrepancies and omissions in multi-source data collected across time points, modal types, and construction stages, coupled with the lack of a unified data organization method, make it difficult to accurately identify and dynamically manage the status of construction components. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for collecting and managing construction data, which solves the problem that multi-source construction data is difficult to model uniformly due to differences in time, mode, and stage, thus affecting the accurate identification and dynamic management of component status.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for collecting and managing construction data, comprising: S1. Extract information about construction components from the building information model and construct a graph structure of component nodes and edges relating to the relationships between components; S2. Collect multi-source sensor data corresponding to the component, and perform time synchronization, format unification and unit conversion on the data to generate standardized data; S3. Construct a multidimensional tensor model based on standardized data to organize the data of components at different time points, data modes, and construction stages. S4. Complete the tensor model to obtain the estimated tensor, and calculate the residuals corresponding to each component based on the difference between the estimated tensor and the original tensor. S5. Input the data of each component in the estimated tensor as node features into the graph structure, and use the graph neural network to perform information propagation and node representation update in the graph structure to obtain the state information of each component. S6. Adjust the edge connection relationship in the residual adjustment diagram of the component, or adjust the acquisition parameter configuration of the corresponding component sensor according to the residual adjustment diagram. S7. Update the status information of the component to the corresponding component in the building information model.

[0006] By adopting the above technical solution, unified organization and modeling of multi-source heterogeneous construction data in three dimensions—time point, data modality, and construction stage—was achieved, and a structured multi-dimensional tensor data model was constructed. Then, through tensor completion and residual calculation, the problem of difficulty in unified modeling of multi-source construction data due to differences in time, modality, and stage was solved, which in turn affected the accurate identification and dynamic management of component status.

[0007] Preferably, in step S1, constructing the graph structure includes extracting the three-dimensional geometric information, component type information, and construction sequence information of the components from the building information model, and determining the edge connection relationship between each component node in the graph structure based on the spatial connection relationship, construction logic relationship, or structural dependency relationship between the components. Each component corresponds to a node, and each edge represents the association between two components.

[0008] Preferably, in step S2, the multi-source sensor data collected includes image data, displacement data, strain data, temperature data, and vibration data; The data processing includes time alignment of the collected data, interpolation or resampling of data from different modalities according to timestamps, and conversion of the data into a unified structure through unit conversion and format unification, as standardized data output.

[0009] Preferably, in step S3, the constructed multidimensional tensor model is a four-dimensional tensor, with components, time, data modes, and construction stages as the four dimensions. The construction process includes filling the data values ​​corresponding to each component at each time point, each mode, and each construction stage in the standardized data into the corresponding positions of the tensor to form a tensor unit T(i,j,k,l), where i represents the component index, j represents the time index, k represents the mode index, and l represents the construction stage index.

[0010] Preferably, in step S4, the tensor completion process includes: The Tucker decomposition method is applied to the constructed tensor model to reconstruct the low-rank tensor, resulting in the estimated tensor. The residual is calculated by taking the difference between the original tensor T(i,j,k,l) ​​and the estimated tensor. The residual is calculated by taking the Frobenius norm of the tensor quantum block for each component at all times, modes, and stages, expressed as: Where T i Represents a tensor slice of component i. This represents the estimated value of component i.

[0011] Preferably, in step S5, the graph neural network includes an input layer, a graph convolutional layer, and an output layer. The input layer receives the data vector of each component in the estimation tensor as node features; the graph convolutional layer performs adjacent node feature aggregation and node representation update based on the edge connection relationship in the graph structure; and the output layer maps the updated node representation to the state information of the component.

[0012] Preferably, in step S6, if the residual value of a component exceeds a preset threshold, one of the following operations is performed: S601. Add a new edge connection to the component node in the diagram structure, wherein the edge connection points to other nodes related to its construction relationship; S602. Adjust the sensor acquisition configuration parameters corresponding to the component node, including sampling frequency, sampling time window length or sampling mode type.

[0013] Preferably, in step S7, the component status information includes a status label, construction progress percentage, or structural health level; The update operation includes writing status information into the attribute fields corresponding to the components in the building information model, while maintaining compatibility with the original data structure of the building information model.

[0014] Preferably, the building information model is a three-dimensional information model. The status information of the components is integrated into the model in the form of attribute values ​​through the application interface. The status information of the components can be expressed in the visualization interface of the building information model through the graphical display of the component attribute values, including status classification, type labeling or component filtering.

[0015] A construction data acquisition and management system, the system comprising the following modules: The component information parsing module is used to parse the attribute information of construction components from the building information model, and establish a graph data structure representing component nodes and their relational edges based on the geometric layout relationship between components, construction procedure logic, or structural connection dependency; the multi-source data integration module is used to collect construction site data obtained from multiple heterogeneous sensors, and perform time unification, format compatibility and unit consistency processing on the data, and output a standard data set in a unified format. The temporal multimodal modeling module is used to generate multi-level tensor data volumes based on standardized data, with components, time, modality, and construction stage as dimensions, to realize the structured expression of cross-modal data throughout the construction cycle; The tensor inference module is used to fill in missing values ​​in the constructed tensors, generate estimated tensors, and extract component residual indices for error analysis based on the difference between the estimated values ​​and the original data. The graph computation modeling module is used to map the component features in the estimated tensor to the node attributes in the graph structure, and combine the graph neural network model to perform graph convolution operations and node embedding updates to generate an expression result that reflects the current state of the component. The feedback adjustment module is used to update the node connection relationship in the graph structure based on the component residual information, or to trigger the dynamic adjustment of sensor sampling parameters to optimize the data acquisition strategy. The model synchronization module is used to inject the acquired component status results into the component objects of the building information model in the form of attribute values, thereby completing the integration and display of the construction status in the 3D information model.

[0016] This invention provides a method and system for collecting and managing construction data. It has the following beneficial effects: 1. In this invention, a unified multidimensional tensor model is constructed by collecting data of construction components at different time points, data modalities, and construction stages. This model is then combined with tensor completion technology to perform missing value inference. The estimated data is then input into a graph neural network to complete the graph structure inference and node representation update of the construction state. By analyzing the residual between the estimated tensor and the original data, the edge connection relationship or perception configuration parameters in the graph structure are further dynamically adjusted. This enables adaptive data acquisition and management for the entire construction cycle, solving the problem that multi-source construction data is difficult to model uniformly due to differences in time, modality, and stage, which affects the accurate identification and dynamic management of component states.

[0017] 2. In this invention, through a systematic data standardization process, the original sensing data can be synchronized in time, converted in format and consistent in unit, and output a dataset with a unified structure to adapt to subsequent modeling processing. This solves the multi-source data barrier and enables image data, displacement data, strain information and other data to collaboratively express the state of construction components, thereby improving the robustness and adaptability of the data processing system.

[0018] 3. In this invention, after obtaining the component status information through graph neural network, it is written into the corresponding component object field in the building information model in the form of attributes, and the status label, progress expression or health level are integrated and updated through the model interface. This realizes the semantic enhancement of the three-dimensional model, strengthens the linkage between the BIM model and the data collected on site, and constructs a data flow path that integrates perception, computing and expression, providing a more intelligent and intuitive decision-making basis for the whole construction process management. Attached Figure Description

[0019] Figure 1 This is a flowchart of a construction data acquisition and management method according to the present invention; Figure 2 This is a flowchart illustrating the graphical structure construction of a construction data acquisition and management method according to the present invention. Figure 3 This is a flowchart of tensor completion and residual calculation for a construction data acquisition and management method according to the present invention; Figure 4 This is a flowchart illustrating the feedback adjustment mechanism of a construction data acquisition and management method according to the present invention. Figure 5 This is a flowchart of the building information model update process for a building construction data acquisition and management method according to the present invention. Figure 6 This is an architecture diagram of a construction data acquisition and management system according to the present invention. Detailed Implementation

[0020] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1 In a first embodiment of the present invention, the present invention provides a method for collecting and managing construction data, such as... Figures 1-5 As shown, it includes: S1. Extract information about construction components from the building information model and construct a graph structure of component nodes and edges representing relationships between components. Further, in step S1, constructing the graph structure includes extracting the three-dimensional geometric information, component type information, and construction sequence information of the components from the building information model, and determining the edge connection relationships between each component node in the graph structure based on the spatial connection relationship, construction logic relationship, or structural dependency relationship between components. Each component corresponds to one node, and each edge represents the association between two components.

[0022] Specifically, the information of construction components is first extracted from the building information model as the initial entry point for construction data management. The building information model is a three-dimensional building model built by the existing BIM platform, which contains rich structured information such as component attributes, spatial layout, and construction process. By parsing the model data, component entities related to the construction stage are extracted, and a set of nodes with components as the basic unit is constructed. During the process of extracting component information, the information obtained includes the three-dimensional geometric information of the component, such as size, position, and orientation; component type information, such as wall, beam, column, and foundation; and construction sequence information, such as the order of construction procedures or work logic path. This information can be read through attribute fields or associated forms in the model, and can be obtained by calling API interfaces or directly parsing from IFC format files. Based on the extracted component information, a construction drawing structure is further constructed. Each component in the drawing structure corresponds to a node in the drawing, and the node identifier is consistent with the component number in the BIM model to maintain data consistency and traceability. The edges in the drawing structure represent the relationships between components. The basis for constructing this edge set includes: spatial connection relationships, such as two components having geometric contact or nesting relationships; construction logic relationships, such as the construction of one component must be carried out after another component; or structural dependency relationships, such as an upper component depending on the load-bearing capacity of a lower component. The connection method of the edges can be undirected or directed, depending on the semantic type being expressed. For example, for the construction process sequence, directed edges can be used to represent the order of events; for structural connections, undirected edges can be used to represent the equivalence of load-bearing relationships between components. The above-mentioned graph structure can be constructed automatically by traversing the connection records between components, construction schedules, or structural topology information in the building information model, or it can be assisted by manual rule settings.

[0023] S2. Collect multi-source sensor data corresponding to the component, and perform time synchronization, format unification and unit conversion on the data to generate standardized data; Furthermore, in step S2, the acquired multi-source sensor data includes image data, displacement data, strain data, temperature data, and vibration data; Data processing includes time alignment of the collected data, interpolation or resampling of data from different modalities according to timestamps, and conversion of the data into a uniform structure through unit conversion and format unification, as standardized data output.

[0024] Specifically, for the established construction component diagrams, based on the physical state and monitoring needs of each component during construction, multi-source sensor data corresponding to each component is collected as key data input for construction status modeling and management. The multi-source sensor data includes various types such as image data, displacement data, strain data, temperature data, and vibration data, which are used to reflect various aspects of information such as component appearance changes, structural deformation, load response, environmental conditions, and dynamic behavior. Among them, image data can be collected by fixed or mobile camera equipment, displacement data and strain data can be collected by structural monitoring equipment such as laser rangefinders, fiber optic sensors, or resistance strain gauges, respectively, while temperature and vibration data are usually monitored in real time by thermal elements, accelerometers, or smart IoT terminals. The aforementioned sensors are distributed across different components or construction sites, and their sampling frequencies, data formats, and unit standards often differ significantly. Furthermore, some modal data are acquired asynchronously. To ensure unified modeling and fusion processing of subsequent data, it is necessary to standardize the acquired multi-source data. First, time alignment processing is performed on all sensor data. By analyzing the timestamps of various types of data, interpolation, resampling, or sliding window methods are used to map asynchronously sampled data to a unified time series framework. Specific methods can include linear interpolation, spline interpolation, or the nearest neighbor time window aggregation method, so that all data have corresponding multimodal observations at the same time point. Secondly, the raw data is formatted and converted to different units. This means that the data output by various sensors, such as the pixel matrix of an image, the acceleration sequence of vibration, and the Celsius value of temperature, are converted into a structured numerical form. For example, image data can be converted into vector representation using feature extraction algorithms, and other physical quantities can be converted according to the standard unit system. This ensures that the data of all components in each mode meet the same numerical domain standard, which facilitates unified modeling and comparative analysis. After the above processing, the generated standardized data can be organized into structured data tables or multidimensional arrays and stored using components, modalities, and time points as index systems. This standardized data not only ensures the structural consistency of cross-modal information but also provides format specifications and a complete data foundation for subsequent tensor construction and graph neural network input.

[0025] S3. Construct a multidimensional tensor model based on standardized data to organize the data of components at different time points, data modes, and construction stages. Furthermore, in step S3, the constructed multidimensional tensor model is a four-dimensional tensor, with components, time, data modes, and construction stages as the four dimensions respectively. The construction process includes filling the corresponding positions of the tensor with the data values ​​of each component at each time point, each mode, and each construction stage in the standardized data to form tensor units T(i,j,k,l), where i represents the component index, j represents the time index, k represents the mode index, and l represents the construction stage index.

[0026] Specifically, based on the standardized data obtained in step S2, a multidimensional tensor model is constructed to achieve unified fusion and structured expression of monitoring data of construction components at different time points, different sensor modes and construction stages, which can be used for subsequent state reasoning and data analysis operations. The tensor model is a four-dimensional tensor used to systematically organize construction data. Its four dimensions correspond to the component dimension i, the time dimension j, the data modality dimension k, and the construction stage dimension l, respectively. Among them, the component dimension is used to distinguish different construction components, the time dimension is used to characterize the time point of data collection, the modality dimension represents different types of sensor data sources such as images, displacement, strain, temperature, vibration, etc., and the construction stage dimension is used to reflect the different procedures or stages in the construction process, such as foundation construction, structural construction, decoration and installation, etc.

[0027] The process of constructing tensors includes the following key steps: First, standardized data is classified according to component ID, and data belonging to the same component are merged into the same tensor index i; combining timestamp information and a unified timeline, data of the same component at different time points are mapped to time index j, and filled into the corresponding modality index k according to the identifier of the acquisition mode; through the progress information, process number, or stage identifier in the planning management system during the construction process, each piece of data is assigned to the corresponding construction stage and filled into the construction stage index l; under the above four-dimensional index positioning, each tensor unit T(i,j,k,l) ​​stores the observation value corresponding to a certain component at a certain time, under a certain mode, and at a certain construction stage. To ensure the integrity and consistency of the tensor structure, cells with missing data can be pre-initialized with null values, zero padding, or placeholder flags to provide input templates for subsequent completion algorithms. For non-numerical or complex data in multimodal data, they can be converted into tensorable numerical vectors through feature encoding or dimensionality reduction methods such as principal component analysis and convolutional feature extraction to ensure the universality and adaptability of tensor construction.

[0028] S4. Complete the tensor model to obtain the estimated tensor, and calculate the residuals corresponding to each component based on the difference between the estimated tensor and the original tensor. Furthermore, in step S4, the tensor completion process includes: The Tucker decomposition method is applied to the constructed tensor model to reconstruct the low-rank tensor, resulting in the estimated tensor. The residual is calculated by taking the difference between the original tensor T(i,j,k,l) ​​and the estimated tensor. The residual is calculated by taking the Frobenius norm of the tensor quantum block for each component at all times, modes, and stages, expressed as: Where T i Represents a tensor slice of component i. This represents the estimated value of component i.

[0029] Specifically, due to the inevitable missing, abnormal or inconsistent issues that occur in the actual collection of multi-source data, in order to improve the integrity of the overall data structure and the usability of subsequent modeling, the multidimensional tensor model constructed in the aforementioned step S3 is completed to obtain the estimated tensor, and further residual information at the component level is calculated based on the difference between the original tensor and the estimated tensor. In specific implementation, this embodiment uses the Tucker decomposition method to perform low-rank tensor reconstruction on the original tensor. Tucker decomposition is a typical tensor decomposition method that can express high-order tensors as the product between the core tensor and the factor matrices of each dimension. By retaining the main components, it effectively compresses and reconstructs the original tensor information. In the application of this invention, the main reason for choosing Tucker decomposition is that it has strong data fitting ability and the ability to express high-dimensional structures, and is suitable for processing four-dimensional tensors with multiple structures of components, time, modes and construction stages. The decomposition process first involves dimensional analysis and rank setting of the input original tensor T to determine the target rank of each factor matrix. Then, optimization algorithms such as alternating least squares or higher-order singular value decomposition are used to jointly fit the core tensor and factor matrices. Finally, the estimated tensor is reconstructed. Used to replace missing terms in the original tensor, while performing structurally consistent fitting on the known data; In obtaining the estimated tensor Then, the residual index corresponding to each component is calculated by the difference between the estimated value and the original value. The residual is used to characterize the degree of difference between the current component's data performance and modeling estimate in the whole cycle, multimodal, and each stage, and is used to adjust the acquisition configuration and graph structure in the future.

[0030] S5. Input the data of each component in the estimated tensor as node features into the graph structure, and use the graph neural network to perform information propagation and node representation update in the graph structure to obtain the state information of each component. Further, in step S5, the graph neural network includes an input layer, a graph convolutional layer, and an output layer. The input layer receives the data vector of each component in the estimation tensor as node features; the graph convolutional layer performs adjacent node feature aggregation and node representation update based on the edge connection relationship in the graph structure; and the output layer maps the updated node representation to the state information of the component.

[0031] Specifically, after completing tensor completion and calculating component residuals, component-level feature inputs are further constructed based on the data in the estimated tensors, and state modeling is carried out in conjunction with the graph structure to realize the expression and updating of the construction state of each component. In order to effectively capture the spatial dependencies, construction logic relationships, and interactive effects between multi-source data features among construction components, this embodiment uses a graph neural network as the core modeling method, and completes the propagation of node features and state representation learning in the graph structure through this network structure. In terms of specific implementation, the structure design of the graph neural network includes three basic parts: input layer, graph convolutional layer, and output layer. The input layer is used to receive the initial feature vector of each component. The feature vector comes from the estimated tensor obtained in step S4. That is, the observation data of each component under multiple time, modality and construction stage dimensions are compressed or flattened as necessary to generate the feature vector corresponding to the component. Each component node corresponds to a unique initial vector input in the graph structure to ensure the one-to-one correspondence between the data and the graph node. The graph convolutional layer, as the core of the network, is used for the aggregation and updating of node features. Based on the graph structure constructed in step S1, the edge connections between component nodes represent spatial adjacency, process dependence, or structural coupling. In this layer, in each round of propagation, the component node not only retains its own features but also integrates the feature information of its neighboring nodes to form a new embedded representation. This process is achieved through graph convolution, using weighted summation or transformation operations based on edge weights or connection attributes. The output layer further maps the node representation updated by the graph convolutional layer to the state information of the component. The state information is a multi-dimensional vector or classification result, which may include the component's current construction state label, functional state representation, or other feature information reflecting its physical, structural, or progress attributes. This state result can be used for feedback adjustment, model updates, or visualization. By introducing graph neural networks, the structural relationships between components and the spatial correlation of data are utilized, enabling the model to not only model based on the observation features of a single component, but also to enhance node representation based on neighborhood information, thereby improving the accuracy and robustness of the construction status representation. In practical deployments, the structure depth, convolution method, and feature dimensions of graph neural networks can all be configured according to project requirements or data scale, exhibiting good flexibility and adaptability.

[0032] S6. Adjust the edge connection relationship in the residual adjustment diagram of the component, or adjust the acquisition parameter configuration of the corresponding component sensor according to the residual adjustment diagram. Furthermore, in step S6, if the residual value of a component exceeds a preset threshold, one of the following operations is performed: S601. Add a new edge connection to the component node in the graph structure. The edge connection points to other nodes related to its construction relationship. S602. Adjust the sensor acquisition configuration parameters corresponding to the component node. The parameters include sampling frequency, sampling time window length or sampling mode type.

[0033] Specifically, after obtaining the state information of each component and its corresponding residual value, the graph structure and sensor acquisition strategy are dynamically adjusted based on the residual information, thereby improving the adaptability and robustness of the construction data acquisition system. Through the feedback adjustment mechanism, the residual value is used as a measure of data integrity or modeling deviation to optimize the relationship between components or the sensing configuration in a targeted manner. The residual value reflects the degree of deviation between the estimated tensor and the original observation data, that is, the credibility or consistency index of the component data under the current modeling framework. In practical applications, when the residual value of a component exceeds the preset deviation threshold, the system will consider that the state expression of the component has potential inaccuracies. At this time, one of the following two types of adjustment operations will be triggered to enhance the model structure or improve the data source input: Dynamic adjustment of graph structure edge connections: When a component node has high residuals, it may be due to insufficient or irrelevant information of its neighboring nodes connected in the current graph structure, resulting in poor information propagation effect. To address this, new edge connections can be introduced into the graph structure for the node of this component to supplement its semantic or structural relationships with other components. For example, based on construction sequence, spatial proximity, or historical co-occurrence patterns, other nodes closely related to the construction of this component can be selected to establish new edge connections, thereby enhancing its information context. This operation does not change the topological core of the original graph structure but is performed as an incremental structural optimization. Sensor acquisition configuration parameter adjustment: If the high residual is caused by insufficient quality of sensing data, the configuration strategy of the sensors associated with this component can be optimized and adjusted. Specifically, the system can automatically modify its sampling parameters, such as increasing the sampling frequency, extending the time window length, or adding key modes, such as expanding from a single strain sensor to strain + temperature dual-mode acquisition. This adjustment can be applied to the field sensing devices through the control interface to achieve synchronous update and distribution of configuration parameters. The above two types of operations can be executed according to the system strategy, or priority or triggering conditions can be set in actual deployment to adapt to different construction stages and data importance scenarios. At the same time, the residual threshold can be set through historical data statistics, adaptive adjustment or expert rules, which has a certain degree of flexibility. Through the above feedback adjustment mechanism, this embodiment realizes a closed-loop mechanism in which the modeling results act in reverse on the data acquisition and graph construction process, avoiding the degradation of model performance caused by long-term reliance on fixed structures and static sampling strategies. This results in the construction of a construction data acquisition and management system with dynamic adaptive capabilities, effectively supporting intelligent sensing and status recognition tasks in complex construction environments.

[0034] S7. Update the status information of the component to the corresponding component in the building information model.

[0035] Furthermore, in step S7, the component status information includes status labels, construction progress percentage, or structural health level; The update operation involves writing status information into the attribute fields corresponding to the components in the building information model, while maintaining compatibility with the original data structure of the building information model.

[0036] Specifically, after completing the modeling and representation of the status information of construction components, in order to achieve effective integration between the building information model and construction perception data, the component status is further integrated and updated. Specifically, component status information is written into the corresponding building information model to achieve visualized expression and continuous integrated management of construction status; Component status information, based on project management and structural monitoring requirements, shall include at least one or more of the following forms of information: Status label: used to identify the current status stage of the component in the construction process, such as not started, under construction, completed, and abnormal, etc. Construction progress percentage: Based on the comparison between the construction plan and the actual progress, a quantitative indicator of progress completion is assigned to each component to facilitate overall project control and phase coordination. Structural health level: Based on the state results obtained from multi-source sensor data modeling, output the health assessment results of the components, such as graded labels such as good, minor damage, and severe damage; The aforementioned status information will be written to the BIM model through the system's update interface. To ensure the consistency and compatibility of BIM data, this embodiment adopts an attribute injection approach for integration: First, the system identifies the unique identifier of the component in the BIM model, such as a GUID or component code, and establishes a one-to-one correspondence between construction perception data and BIM entity objects; the system maps the aforementioned status information to the attribute fields of the corresponding components in the model, and completes the data writing using extended attribute tables, user-defined fields, or plug-in interfaces, avoiding damage to the original BIM data structure and maintaining compatibility with existing model standards; In practical applications, the above update operations are usually executed automatically through middleware or BIM platform plugins, supporting three modes: timed synchronization, event triggering, or user command. The updated status information can also be expressed in the visualization interface of the BIM platform, for example: using color coding to represent different component status levels; displaying construction progress information with floating labels in the model; and combining filtering functions to filter and locate components in different statuses.

[0037] Through the above methods, this embodiment realizes the dynamic integration and visualization of component status in the 3D BIM model, enabling construction managers to grasp the status evolution of each component in real time in a unified model environment, thereby improving the information transparency and decision-making efficiency of the entire construction process.

[0038] Furthermore, the Building Information Model (BIM) is a three-dimensional information model. The status information of components is integrated into the model in the form of attribute values ​​through the application interface. The status information of components can be expressed in the visualization interface of the BIM through the graphical display of component attribute values, including status classification, type labeling, or component filtering.

[0039] Specifically, Building Information Modeling (BIM) is a three-dimensional information model constructed using standardized modeling methods. It can express the geometric shape, attribute information, and semantic relationships of building components and serves as a digital carrier for the integration of information throughout the entire construction lifecycle. To achieve collaborative management and interactive display between the perceived state of components and the 3D model, this embodiment introduces component state information into the model, specifically using an attribute value integration method for data writing; The state information of the components comes from the graph neural network modeling results and is usually represented in the form of structured labels, classification fields or numerical fields. To ensure the integrity of the model structure and the efficiency of data access, the state information is written through a standardized application interface. The interface can be in the form of a plug-in, a platform-provided API, or a custom model extension function, and has the characteristics of good compatibility and strong adaptability. During the writing process, the system first locates the target component's entity object in the 3D information model. This location is based on the component's unique identifier, such as a globally unique ID or component number, ensuring a one-to-one correspondence between status data and model components. The system binds the status information to the user-defined attribute fields or extended fields of the component object. This binding process does not require changing the original structure of the model or rebuilding the model system. In this way, dynamic mounting of status information can be achieved while maintaining the consistency of the model's original geometric and engineering semantics. The integrated status information is not only updated at the data layer but also supports graphical representation in the building information model's visualization interface; specifically, it includes the following common display methods: Status level display: Based on the status level of the component, different colors, transparency or materials are used to render the component. For example, a healthy state is green and an abnormal state is red. Type labeling display: Floating status labels such as under construction, completed, and requiring inspection are displayed on components to enhance readability; Component filtering function: Users can filter or group components based on their status information to quickly locate components in specific states within the model, facilitating construction management or operation and maintenance deployment; Through the visualization integration and status graphic display mechanism of the above-mentioned three-dimensional information model, data-driven model enhancement is realized, providing a more intuitive and efficient status feedback channel for the construction site.

[0040] Example 2: In complex construction site environments, numerous heterogeneous sensors are deployed, resulting in significant inconsistencies in data collection across time distribution, data modalities, and construction stages. Furthermore, the construction dependencies between different components are difficult to directly map into a unified data structure. This hinders traditional construction monitoring systems from achieving multi-dimensional data fusion modeling and intelligent status identification, particularly when data is incomplete or partially lost. The lack of an effective mechanism to infer component status and dynamically adjust collection strategies limits the level of intelligence in construction status perception and management. To address these issues, this invention provides a building construction data acquisition and management system, such as... Figure 6 As shown. The specific implementation process of this system is as follows: A construction data acquisition and management system, comprising the following modules: The component information parsing module is used to parse the attribute information of construction components from the building information model, and establish a graph data structure representing component nodes and their relational edges based on the geometric layout relationship between components, construction procedure logic, or structural connection dependency; the multi-source data integration module is used to collect construction site data obtained from multiple heterogeneous sensors, and process the data execution time to be unified, format compatible, and unit consistent, and output a standard data set in a unified format. The temporal multimodal modeling module is used to generate multi-level tensor data volumes based on standardized data, with components, time, modality, and construction stage as dimensions, to realize the structured expression of cross-modal data throughout the construction cycle; The tensor inference module is used to fill in missing values ​​in the constructed tensors, generate estimated tensors, and extract component residual indices for error analysis based on the difference between the estimated values ​​and the original data. The graph computation modeling module is used to map the component features in the estimated tensor to the node attributes in the graph structure, and combine the graph neural network model to perform graph convolution operations and node embedding updates to generate an expression result that reflects the current state of the component. The feedback adjustment module is used to update the node connection relationship in the graph structure based on the component residual information, or to trigger the dynamic adjustment of sensor sampling parameters to optimize the data acquisition strategy. The model synchronization module is used to inject the acquired component status results into the component objects of the building information model in the form of attribute values, thereby completing the integration and display of the construction status in the 3D information model.

[0041] Specifically, the component information parsing module identifies the spatial arrangement relationship, construction sequence information, or structural connection logic of components in the model, and expresses the coupling relationship between the component and its upstream and downstream components in the form of a graph structure, providing the input basis for the subsequent node definition and edge connection relationship of the graph neural network. The multi-source data integration module is configured with preprocessing logic for multimodal sensor data such as images, temperature, and strain, ensuring that the acquired data can maintain consistency in terms of time, units, and format, and outputting standardized data formats for subsequent processing modules to use; The temporal multimodal modeling module organizes data according to a preset dimensional structure, including components, time, modality, and stage, to generate a unified data representation model for the entire construction process. This tensor structure provides the system with a structured input data interface, supporting complete data modeling in terms of temporal continuity and modal intersection. The tensor inference module identifies missing or sparse regions in the constructed tensor structure and performs tensor completion and residual calculation. The residual information serves as a key basis for the system to provide feedback on data consistency and observation integrity, and is used to drive the subsequent graph structure optimization and acquisition strategy adjustment mechanism of the system. The graph computation modeling module maps component-level data in the completed tensor to graph structure node features, and performs graph convolution and node representation update operations in conjunction with the graph neural network framework to generate the current state representation of each component. By utilizing structural dependency information in the graph, it effectively integrates local component features and their adjacency influences to achieve context-aware state reasoning capabilities. The feedback adjustment module is used to dynamically adjust the graph structure or acquisition configuration based on residual information. It supports expanding or shrinking edge relationships in the graph, enhancing the system's ability to model potential influence paths between components. It can also strategically reconfigure sensor parameters to optimize data acquisition quality. The model synchronization module is used to inject the status results of each component into the original building information model in the form of attribute values. It supports integration with the BIM platform through an interface, and can realize the synchronous update of status fields while keeping the original component structure unchanged. It also supports the intuitive presentation of construction status information through the model visualization interface, realizing closed-loop data management.

[0042] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A construction work data collection management method characterized by, The method comprises the following steps: S1, extracting information of construction components from a building information model, constructing a graph structure of component nodes and edges of inter-component association; S2, collecting multi-source sensor data corresponding to the components, and performing time synchronization, format unification and unit conversion processing on the data to generate standardized data; S3, constructing a multi-dimensional tensor model based on the standardized data, for organizing data of the components in different time points, data modalities and construction stage dimensions; S4, performing completion processing on the tensor model to obtain an estimated tensor, and calculating the residual of each component according to the difference between the estimated tensor and the original tensor; S5, inputting the data of each component in the estimated tensor as node features into the graph structure, and performing information propagation and node representation update in the graph structure through a graph neural network to obtain state information of each component; S6, adjusting the edge connection relationship in the graph structure according to the residual of the component, or adjusting the acquisition parameter configuration of the corresponding component sensor; S7, updating the state information of the component to the corresponding component in the building information model.

2. The construction data collection management method according to claim 1, wherein, In step S1, the construction of the graph structure comprises extracting three-dimensional geometric information, component type information and construction sequence information of the components from the building information model, and determining the edge connection relationship between the nodes of each component in the graph structure based on the spatial connection relationship, construction logic relationship or structural dependency relationship between the components, wherein each component corresponds to a node, and each edge represents the association between two components.

3. The method of claim 1, wherein, In step S2, the collected multi-source sensor data includes image data, displacement data, strain data, temperature data and vibration data; The data processing includes time alignment processing of the collected data, interpolation or resampling of the data of different modalities according to the time stamp, and conversion of the data into a unified structure through unit conversion and format unification as standardized data output.

4. The method of claim 1, wherein, In step S3, the constructed multi-dimensional tensor model is a four-dimensional tensor, with components, time, data modalities and construction stages as four dimensions; the construction process includes filling the data value of each component in the standardized data at each time point, each modality and each construction stage into the corresponding position of the tensor to form a tensor unit T(i,j,k,l), wherein i represents the component index, j represents the time index, k represents the modality index, and l represents the construction stage index.

5. The method of claim 1, wherein, In step S4, the tensor completion processing includes: Applying the Tucker decomposition method to the constructed tensor model for low-rank tensor reconstruction, an estimated tensor The residual is computed by the difference between the original tensor T(i,j,k,l) and the estimated tensor, the residual is computed as the Frobenius norm of the tensor sub-blocks of each component at all times, modes and phases, denoted as where T i denotes the tensor slice of component i, denotes the estimate of component i.

6. The method for construction data collection management according to claim 1, wherein, In step S5, the graph neural network includes an input layer, a graph convolution layer and an output layer, wherein the input layer receives the data vector of each component in the estimated tensor as the node feature; the graph convolution layer performs adjacent node feature aggregation and node representation update based on the edge connection relationship in the graph structure; and the output layer maps the updated node representation to the state information of the component.

7. The method of claim 1, wherein, In step S6, if the residual value of a component exceeds a preset threshold, one of the following operations is performed: S601, adding a new edge connection to the component node in the graph structure, the edge connection pointing to other nodes related to the construction relationship thereof; S602, adjusting the sensor acquisition configuration parameters corresponding to the component node, the parameters including sampling frequency, sampling time window length or sampling modality type.

8. The method of claim 1, wherein, In step S7, the component state information includes a state label, a construction progress percentage, or a structure health level; The updating operation includes writing the state information into an attribute field corresponding to the component in the building information model, and keeping compatible with the original data structure of the building information model.

9. The method of claim 1, wherein, The building information model is a three-dimensional information model, and the state information of the component is integrated into the model in the form of attribute values through an application interface, and can be expressed in the visual interface of the building information model through graphical display of component attribute values, including state classification, type labeling, or component screening.

10. A construction data acquisition management system characterized by comprising: The system comprises the following modules: A component information analysis module is configured to analyze attribute information of construction components from the building information model, and to establish a graph data structure representing component nodes and their relationship edges based on geometric layout relationships between components, construction process logic, or structural connection dependencies; A multi-source data integration module is configured to collect construction site data obtained by a plurality of heterogeneous sensors, and to perform time unification, format compatibility, and unit consistency processing on the data to output a standard data set in a unified format; A time-series multi-modal modeling module is configured to generate a multi-order tensor data body with components, time, modalities, and construction stages as dimensions based on the standardized data, to realize structured expression of construction full-cycle, cross-modal data; A tensor reasoning module is configured to perform missing value completion on the constructed tensor to generate an estimated tensor, and to extract component residual indicators for error analysis based on differences between the estimated values and the original data; A graph computing modeling module is configured to map component features in the estimated tensor to node attributes in the graph structure, to perform graph convolution operation and node embedding update in combination with a graph neural network model, and to generate an expression result reflecting the current state of the component; A feedback adjustment module is configured to update node connection relationships in the graph structure based on component residual information, or to trigger dynamic adjustment of sensor sampling parameters to optimize data acquisition strategies; A model synchronization module is configured to inject the obtained component state result into the component object of the building information model in the form of attribute values, to complete integration and display of the construction state in the three-dimensional information model.

Images