Intelligent building structure construction method based on point cloud data and digital twinning

Abstract

The invention discloses an intelligent building structure construction method based on point cloud data and digital twinning, and the method comprises the steps: obtaining a building point cloud through three-dimensional laser scanning, carrying out the high-precision registration and preprocessing, carrying out the component feature extraction and parameterized reverse modeling through an RANSAC algorithm, and generating a high-fidelity BIM model; then, the model is deeply fused with GIS, finite element and Internet of Things data under a unified space-time reference, and a dynamically updated digital twinning scene is constructed. Finally, a virtual simulation platform is developed based on the scene, and real-time simulation, visualization and intelligent decision support of the construction progress and mechanical response are achieved. According to the method, the problems of insufficient precision and difficulty in multi-source data fusion of a traditional modeling method are solved, accurate mapping and interaction of a building structure from a physical entity to a digital virtual body are realized, and the intelligent level and decision scientificity of a construction process are remarkably improved.

Description

Technical Field

[0001] This invention relates to the fields of building information technology and intelligent construction technology, specifically to an intelligent construction method for building structures that integrates three-dimensional laser scanning, reverse modeling, digital twin and virtual simulation technologies. Background Technology

[0002] Currently, the construction industry is transforming and upgrading towards digitalization, industrialization, and intelligentization. Although Building Information Modeling (BIM) technology is widely used, its modeling process heavily relies on design drawings. For existing buildings, historical buildings, or complex structures with deviations during construction, it is difficult to quickly and accurately construct models that reflect the actual physical state. In addition, traditional construction management relies on two-dimensional drawings and experience-based judgment, lacking accurate perception and forward-looking prediction of dynamic changes in the construction process.

[0003] 3D laser scanning technology can rapidly acquire high-precision, high-resolution spatial 3D information (point cloud) of building surfaces non-contactly, making "real-world modeling" possible. However, the automated conversion of massive point cloud data into semantic BIM models that can be analyzed and simulated remains a technical challenge. Meanwhile, multi-source heterogeneous data generated during construction, such as progress, quality, and mechanical properties, are isolated from each other, forming "information silos" that cannot effectively support collaborative decision-making.

[0004] Digital twin technology, as a bridge connecting the physical and information worlds, provides a framework for solving the aforementioned problems. However, current applications of digital twins in the construction field mostly focus on the operation and maintenance phase, while their application in the construction phase is still in its infancy, generally suffering from low model fidelity, insufficient data fusion depth, inadequate dynamic update mechanisms, and weak simulation and decision-making capabilities.

[0005] Therefore, there is an urgent need for a systematic approach that integrates high-precision reverse modeling, deep data fusion, and high-fidelity simulation to achieve accurate mapping and two-way interaction between building structures from "physical entities" to "digital virtual entities," thereby driving intelligent decision-making and management in the construction process. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent construction method for building structures based on point cloud data and digital twins, so as to achieve:

[0007] a) Rapid and high-precision digital reverse reconstruction of building structures (especially complex spatial structures and existing structures);

[0008] b) Break down the barriers between multiple data sources such as point cloud, BIM, mechanical analysis, and IoT monitoring to achieve deep semantic fusion and dynamic association;

[0009] c) Construct a virtual simulation platform that can reflect the construction status in real time, support process simulation and intelligent decision-making, and improve construction quality, safety and efficiency.

[0010] To achieve the above objectives, the present invention adopts the following technical solution:

[0011] A smart construction method for building structures based on point cloud data and digital twins includes the following steps:

[0012] S1. High-precision point cloud data acquisition and preprocessing of building structure: The target building structure is scanned at multiple stations using a 3D laser scanner to obtain complete on-site point cloud data. Then, the on-site point cloud data is processed sequentially with automatic registration, noise reduction filtering and lightweighting.

[0013] S2. Reverse modeling of building structure digital twin: Based on preprocessed point cloud data, feature extraction and parametric modeling algorithms are used to automatically identify and reconstruct the main structural components of the building to generate a high-fidelity building information model.

[0014] S3. Construction of digital twin scenarios through multi-source heterogeneous data fusion: The reverse-generated building information model is unified with geographic information system data, finite element analysis model and Internet of Things sensor data in terms of spatiotemporal reference and semantic integration to construct a dynamically updated digital twin scenario.

[0015] S4. Virtual simulation and intelligent decision-making for building structure construction: Based on the digital twin scenario, the construction progress and structural mechanical response are simulated and visualized in real time. By comparing and analyzing the simulation data and the measured data, the construction plan is dynamically optimized and intelligent decision support is provided.

[0016] Furthermore, in step S1,

[0017] The automatic registration adopts a strategy that combines target-assisted coarse registration with the iterative nearest point algorithm to unify the multi-site cloud into the same coordinate system. Specifically, it includes: using a coarse registration algorithm based on least squares fitting for initial positioning; and using an iterative nearest point algorithm for fine registration. The iterative nearest point algorithm uses the sum of the squares of the distances between each point in the source point cloud and the nearest point in the target point cloud after rotation and translation as the registration error. By iteratively optimizing the rotation matrix and translation vector, the overall registration error is controlled within ±2mm.

[0018] The noise reduction uses a statistical outlier removal algorithm to remove identified noise points. Specifically, it includes: first, calculating the average distance from each point in the point cloud to its preset number of nearest neighbors; if the average distance is greater than the sum of the products of the mean of the average distances of all points, the threshold coefficient, and the standard deviation, then the point is identified as a noise point and removed.

[0019] The lightweighting process employs a voxel meshing method to uniformly downsample the denoised point cloud.

[0020] Further, step S2 specifically includes:

[0021] 2.1) Based on spatial geometric features and Euclidean clustering algorithm, the point cloud is automatically segmented into point sets corresponding to different structural components such as columns, beams and slabs;

[0022] 2.2) The RANSAC algorithm is used to fit the central axis and cross-sectional contour of the segmented component point cloud. The optimal model parameters are found to maximize the number of data points that meet the condition that the distance from the point to the geometric model to be fitted is less than a preset threshold.

[0023] 2.3) Optimize the spatial relationship of component connection nodes based on the nearest neighbor fitting algorithm; the optimization objective is to minimize the sum of squared distances between the endpoints of adjacent component axes after rigid body transformation.

[0024] 2.4) After extracting the geometric parameters of the component, including length, orientation, and cross-sectional dimensions, a parametric three-dimensional solid model is generated using the cross-section sweep method.

[0025] Furthermore, step S3 specifically includes:

[0026] 3.1) Construct a unified spatial reference transformation model based on a spherical coordinate system, integrate digital elevation models and orthophotos through the Cesium engine, realize the positional alignment of data from different sources, and build a three-dimensional virtual geographic environment with a real geographic background;

[0027] 3.2) The Building Information Model is parsed and lightweighted using the IFC standard, converted into glTF / 3D Tiles format, and optimized using the level of detail method to achieve efficient loading, rendering, and visualization in the Web environment;

[0028] 3.3) Establish a unified data standard and interface, and realize node-level semantic association and dynamic mapping between finite element analysis results and geometric models based on the structured data parsing engine of Parquet columnar storage.

[0029] Furthermore, step S4 specifically includes:

[0030] 1) A virtual simulation platform based on the Cesium engine and a B / S architecture was developed, integrating a construction progress simulation module and a stress-displacement simulation module;

[0031] The construction progress simulation module associates the construction plan with the model components to realize 4D dynamic simulation of the construction process and compare and analyze it with the actual progress; the stress-displacement simulation module integrates a lightweight finite element analysis engine and completes mechanical performance simulation through the finite element control equations composed of structural stiffness matrix, nodal displacement vector, and nodal load vector.

[0032] 2) WebGL technology is used to achieve multi-dimensional data fusion rendering and immersive interaction of point clouds with hundreds of millions of points, large-scale BIM models and mechanical cloud maps, and to map mechanical simulation results to digital twin models in real time;

[0033] 3) Establish a closed-loop control process: data acquisition - model correction - simulation prediction - command feedback.

[0034] The twin model and simulation results are dynamically updated based on IoT measured data; by comparing the simulation predictions with the measured values, potential risks are diagnosed, and data-driven decision-making suggestions are provided for adjusting the construction plan.

[0035] The design concept of this invention is as follows:

[0036] This method first acquires building point clouds through 3D laser scanning. After high-precision registration and preprocessing, it uses algorithms such as RANSAC to extract component features and perform parametric reverse modeling to generate a high-fidelity BIM model. Then, this model is deeply integrated with GIS, finite element, and IoT data under a unified spatiotemporal reference to construct a dynamically updated digital twin scene. Finally, a virtual simulation platform is developed based on this scene to achieve real-time simulation, visualization, and intelligent decision support for construction progress and mechanical response. This invention solves the problems of insufficient accuracy and difficulty in multi-source data fusion in traditional modeling methods, realizing accurate mapping and interaction of building structures from physical entities to digital virtual entities, significantly improving the intelligence level and scientific decision-making of the construction process.

[0037] Compared with the prior art, the present invention has the following significant advantages:

[0038] 1. High modeling accuracy and high degree of automation: Through point cloud-driven reverse modeling technology, it overcomes the dependence of traditional BIM modeling on drawings and can accurately reflect the actual state of the structure. In particular, for irregular components and construction deviations, the automated algorithm greatly improves the modeling efficiency.

[0039] 2. Strong data fusion depth: It realizes deep fusion of multi-dimensional and multi-scale data from geometric model (BIM), physical properties (FEA) to real-time status (IoT), so that the digital twin is no longer a static model, but a dynamic entity carrying rich information.

[0040] 3. Scientific and intuitive decision support: The complex construction process and structural mechanical behavior are simulated and presented in a visual way, which lowers the decision-making threshold and enables managers to proactively identify problems and optimize solutions, effectively improving construction safety and economy.

[0041] 4. Full lifecycle support: The high-precision digital twin model and its data fusion framework created by this invention not only serve the construction phase, but can also be seamlessly transferred to the operation and maintenance phase, laying a solid foundation for the full lifecycle management of buildings. Attached Figure Description

[0042] Figure 1 Flowchart for reverse modeling and digital twin generation of building structure point cloud data;

[0043] Figure 2 A flowchart for constructing digital twin scenarios and fusing multi-source heterogeneous data;

[0044] Figure 3 A flowchart illustrating the intelligent construction simulation and decision-making closed-loop control process for building structures based on digital twins;

[0045] Figure 4 This is a schematic diagram showing the normal state of the structure in the embodiment, with the key column C1_F4 in a normal state.

[0046] Figure 5 This is a schematic diagram illustrating a structural emergency state in the embodiment, where the critical column C1_F4 is in a dangerous state.

[0047] Figure 6 This is a schematic diagram showing the structure returning to its normal state in the embodiment, with the key column C1_F4 in a normal state.

[0048] Figure 7 This is a comprehensive report on the digital twin simulation of the multi-layer frame structure in the embodiments. Detailed Implementation

[0049] The present invention will be further described below with reference to the accompanying drawings.

[0050] A smart construction method for building structures based on point cloud data and digital twins includes the following steps:

[0051] S1. High-precision point cloud data acquisition and preprocessing of building structures

[0052] A terrestrial 3D laser scanner is used to perform multi-station scanning of the target building structure (this method focuses on buildings under construction, but can also be applied to existing buildings) to obtain complete site point cloud data. The preprocessing process includes:

[0053] Registration: A strategy combining target-assisted coarse registration and the iterative nearest-point algorithm is employed to unify multi-site cloud data into the same coordinate system. Fine registration utilizes the iterative nearest-point algorithm, with the objective function being:

[0054] (1)

[0055] In the formula: Registration error; It is a rotation matrix; It is a translation vector; Points in the source point cloud; The nearest point in the target point cloud; To determine the number of matching point pairs.

[0056] Through iterative optimization and This ensures that the overall registration error is controlled within ±2mm.

[0057] Noise Reduction and Lightweighting: A statistical outlier removal algorithm is used to filter noise points. The criteria for this algorithm are as follows:

[0058] (2)

[0059] like Then determine Noise points are identified and removed.

[0060] In the formula: For point To its The average distance between the nearest neighbors; This is the mean of the average distances between all points; Standard deviation; This is the threshold coefficient (usually taken as 1~2).

[0061] Subsequently, a voxel meshing method was used to uniformly downsample the point cloud, which significantly reduced the amount of data while preserving geometric features.

[0062] S2, Reverse Modeling of Building Structure Digital Twin

[0063] This step is crucial for moving from "point" to "volume" and from "geometry" to "semantics," specifically:

[0064] Component segmentation and recognition: Based on spatial geometric features (such as normal vectors and curvature) and clustering algorithms (such as Euclidean clustering), the point cloud is automatically segmented into different sets of structural component points (such as columns, beams, and slabs).

[0065] Feature extraction and parametric modeling: For the segmented component point cloud, the RANSAC algorithm is used to fit its central axis and cross-sectional profile. The goal of RANSAC is to maximize the set of interior points, and its mathematical model is as follows:

[0066] (3)

[0067] In the formula: These are the optimal model parameters; Point cloud data points; The geometric model to be fitted (such as a plane or cylinder; the geometric model refers to the shape corresponding to the structural component). This is the distance from the point to the model; This is the distance threshold.

[0068] After extracting the key geometric parameters of the component (such as length, direction, and cross-sectional dimensions), a parametric three-dimensional solid model is generated using cross-section sweeping technology.

[0069] When using point cloud data to acquire component axes in a digital twin model, the phenomenon of missing or non-intersecting axis points at the connection nodes of each component is encountered (mainly due to two reasons: firstly, the axes of real structural components cannot intersect at a single point; secondly, the point cloud data at the node connections is missing during the point cloud scanning process, thus preventing the extraction of axis information at the nodes). The connection relationship is optimized through nearest-neighbor point cloud fitting, with the goal of minimizing the distance between the endpoints of the axes at the nodes.

[0070] (4)

[0071] In the formula: , These are the endpoints of the axes of adjacent components; These are the rigid body transformation parameters.

[0072] S3, Digital Twin Scenario Construction Based on Multi-Source Heterogeneous Data Fusion

[0073] Build a twin scenario that supports dynamic data-driven approaches:

[0074] Virtual geographic environment construction: Based on the Cesium engine, integrating digital elevation models (DEM) and orthophotos (DOM), a 3D base map scene with real geographic coordinates is constructed.

[0075] Model Lightweighting and Integration: The reverse-generated BIM model undergoes lightweighting processing through format conversion (e.g., from IFC to glTF / 3D Tiles), including geometric topology simplification, redundant component merging, and texture compression optimization. Tools such as CesiumLab are then used to efficiently convert it to the glTF / 3D Tiles format suitable for Web 3D rendering. Simultaneously, Level of Detail (LMD) technology is incorporated. LOD (Level of Detail) is an existing computer graphics optimization technique that creates multiple model versions of the same object with different geometric complexities and automatically selects the appropriate version for rendering based on viewpoint distance or screen space occupancy, thereby significantly improving rendering performance while maintaining visual realism. For example, in building visualization, when the observer is far away, the system loads only a basic outline model composed of a few polygons; as the observer approaches, it sequentially switches to a medium-detail model containing components such as windows and balconies, until a high-precision model containing interior structural and decorative details is loaded. In digital twin platforms, this technology is a key guarantee for enabling real-time browsing of large-scale architectural scenes. LOD dynamically schedules model representations of different precisions based on viewpoint distance and scene importance, and improves data loading efficiency through spatial indexing and tile organization, ultimately achieving smooth integration and high-performance rendering of large-scale architectural digital twin models in the Cesium engine.

[0076] Multi-source data fusion: Establish a unified data standard and interface to bind and map nodal stress / displacement data from finite element analysis and real-time monitoring data from IoT sensors to corresponding components in the BIM model through spatial location and semantic relationships; adopt columnar storage formats (such as Parquet) to achieve efficient data parsing and dynamic mapping.

[0077] S4. Virtual Simulation and Intelligent Decision-Making for Building Structure Construction

[0078] Based on the aforementioned digital twin, a virtual simulation platform is developed to enable intelligent applications:

[0079] Construction progress simulation: The construction plan is linked with the model components to realize 4D dynamic simulation of the construction process and compare and analyze it with the actual progress.

[0080] Mechanical performance simulation and visualization: Integrating a lightweight finite element analysis engine or interfacing with external analysis software, the mechanical simulation results (such as stress contour plots and deformation animations) are rendered in real time in the twin model. The finite element governing equations can be simplified as follows:

[0081] (5)

[0082] In the formula: Here is the structural stiffness matrix; The nodal displacement vector; This represents the nodal load vector.

[0083] Intelligent Decision Support: Constructing a closed loop of "perception-decision-execution". The platform diagnoses potential risks (such as excessive deformation and stress concentration) by comparing simulation predictions with sensor measurements, and provides data-driven decision suggestions for adjusting construction plans (such as support unloading sequence and component hoisting path).

[0084] like Figure 1 As shown, this fully demonstrates the entire reverse modeling process from raw point clouds to a structured, parametric Building Information Model (BIM). This process forms the geometric and semantic foundation of the digital twin.

[0085] 1. Point Cloud Registration: This module corresponds to step S1 in the manual. Its function is to unify discrete point clouds obtained from multiple stations into a single global coordinate system through target-assisted coarse registration and an iterative nearest-point algorithm. The process is as follows: first, initial positioning (coarse registration) based on least squares is performed; then, the rotation matrix R and translation vector T are iteratively optimized to minimize the distance error E between corresponding points, achieving spatial alignment with millimeter-level accuracy.

[0086] 2. Object Point Cloud Denoising: This module corresponds to the denoising and lightweighting processing in section S1 of the manual. Its function is to remove noisy data such as flying points and outliers generated during the scanning process. The processing employs a statistical outlier removal algorithm, calculating the average distance from each point to its k nearest neighbors. Points that deviate significantly from the overall distribution are identified as noise and filtered out to purify the data.

[0087] 3. Point Cloud Clustering and Segmentation: This module corresponds to step S2.1 in the instruction manual. Its function is to automatically decompose the denoised overall point cloud into point sets corresponding to different structural components (such as columns, beams, and slabs) based on spatial location and geometric features. The processing combines Euclidean clustering with analysis based on geometric features such as normal vectors and curvature to achieve preliminary identification and separation of components.

[0088] 4. Component Cross-section and Axis: This module corresponds to step S2.2 in the instruction manual. Its function is to extract key geometric parameters defining the shape and position of the segmented component point cloud. The processing employs robust fitting algorithms such as RANSAC to identify and fit the component's central axis and cross-sectional profile (e.g., rectangle, circle) from the point cloud, and calculate parameters such as length, direction, and dimensions.

[0089] 5. Connecting Nodes: This module corresponds to step S2.3 in the instruction manual. Its function is to handle the complex geometric relationships in the intersection areas of components, ensuring the correct model topology. The processing is based on nearest-neighbor point clouds, optimizing the connection relationships of node regions by minimizing the distance between the endpoints of adjacent component axes, and performing rigid body transformations to achieve precise component splicing.

[0090] 6. Structural Feature Extraction: This module runs through S2 and represents the information enhancement stage of reverse modeling. Its function is to extract structured information from the identified geometric shapes for analysis and management. The process includes: assigning semantic types to components (e.g., "KZ-1 concrete column"), associating material properties, and establishing hierarchical and connection relationships between components.

[0091] 7. Model Building: This module corresponds to step S2.4 in the instruction manual. Its function is to ultimately generate a three-dimensional solid digital twin model that can be used for calculation and simulation. The processing is based on the extracted parameters (sections, axes), and uses parametric modeling methods such as section sweeping to generate a BIM model (such as IFC format) with complete geometric and semantic information, laying the foundation for subsequent integration and application.

[0092] like Figure 2 As shown, this demonstrates how to deeply integrate a reverse-generated BIM model with multi-source heterogeneous data in a unified virtual geographic environment to construct a digital twin scene that supports dynamic driving.

[0093] 1. Multi-source data input (left side): This section gathers various data sources required for building the scenario, corresponding to the inputs in the manual S1, S2, and S3.

[0094] (a) Digital elevation model / digital orthophoto: Through processes such as "elevation data cleaning, conversion, and stitching", a realistic terrain base is formed.

[0095] (b) BIM model / 3D point cloud data: i.e. Figure 1 The final output of the process serves as the core architectural entity of the scene.

[0096] (c) Finite element simulation data: Calculation results from tools such as ANSYS, providing information on the mechanical properties of the structure.

[0097] 2. Unified processing (middle part): This is the core preprocessing step for data fusion.

[0098] Cartesian-WGS84 coordinate system transformation: corresponds to instruction manual S3.1. Its function is to unify data from different sources (such as BIM data in local coordinates, monitoring data in engineering coordinates) into the spherical WGS84 geographic coordinate system (or its projected coordinate system), achieving precise spatial alignment of all data. The process involves complex coordinate transformations and datum unification.

[0099] 3. Virtual Geographic Environment and Database (Central Region):

[0100] (a) Virtual geographic environment: Based on the Cesium engine, the processed DEM and DOM are constructed into a three-dimensional spherical scene with a real geographic background.

[0101] (b) 3D Building Model Information Repository: Stores and manages BIM models after coordinate transformation and lightweighting.

[0102] 4. Integration and Lightweight Technology (Right Side): This section describes the key technologies for efficiently integrating BIM models into a web-based 3D engine.

[0103] (a) Structured data parsing technology: corresponding to S3.3 of the instruction manual, used for efficient parsing and processing of time-series or attribute data such as monitoring data and finite element results.

[0104] (b) Cesium Lab 3D Tiles Technology: Refers to section S3.2 of the instruction manual. Its function is to enable smooth visualization of ultra-large-scale 3D models on the web. The process involves using Cesium Lab tools to convert BIM models in formats such as IFC into glTF and 3D Tiles formats suitable for web streaming through geometric simplification, texture compression, and LOD generation.

[0105] (c) Columnar storage format (Parquet): Used for storing and quickly querying massive amounts of structured simulation and monitoring data.

[0106] 5. 3D Spherical Scene Fusion (Final Output): After the above processing, all data is integrated and rendered in the 3D spherical scene built by Cesium to form a dynamic digital twin scene that carries multi-dimensional information such as geometry, physics, and monitoring.

[0107] like Figure 3 The diagram illustrates the system architecture, functional modules, and closed-loop workflow of a virtual simulation platform built on a digital twin scenario.

[0108] 1. Platform Core: Web Browser-based Virtual Simulation Platform: The entire system runs on a B / S architecture, and users access it through a web browser.

[0109] 2. Front-end functional modules: provide user interaction and visualization.

[0110] (a) View Management: Provides functions such as scene manual animation and model perspective switching, and handles user perspective control commands.

[0111] (b) Simulation: This includes three sub-modules: environmental multi-field coupling (simulating wind and temperature fields), structural stress and displacement (integrating lightweight FEA for real-time mechanical response solving), and virtual construction (4D construction progress simulation). The process involves inputting the analysis model and data into the corresponding engine to drive the visual simulation.

[0112] (c) Engineering Information Inquiry: Processes user click query requests, retrieves and displays component attributes, real-time monitoring data, simulation results, etc. from the database.

[0113] 3. Backend Data and Processing Module: Supports data management and computational analysis for frontend functions.

[0114] (a) Data storage: Classified management of structured data (sensor data), geographic information data (DEM / DOM), unstructured data (point cloud / image), and structural model data (BIM / 3D Tiles).

[0115] (b) Modeling and Analysis:

[0116] Algorithm library management: Encapsulates core algorithms such as point cloud processing, finite element calculation, and risk diagnosis.

[0117] Scene modeling: Execution Figure 2 The process of scene fusion and model lightweighting in the process.

[0118] Data analysis involves comparing, statistically analyzing, and trending input data to provide a basis for decision-making.

[0119] Data exchange and export: Provides standard data interfaces to support data exchange with other systems (such as project management software).

[0120] 4. Closed-loop data flow (illustrated loop): The platform implements a complete decision-making closed loop.

[0121] (a) Input: Collect digital elevation data, orthophoto data, point cloud model data, numerical analysis data, construction simulation data, etc. (i.e. all outputs of S1-S3).

[0122] (b) Processing: Calculation and fusion are performed through the “Modeling and Analysis” module.

[0123] (c) Output and Decision: Visualization and deduction are performed in the “Simulation” module. By comparing simulation predictions with “actual data” from the Internet of Things, deviations and risks are identified.

[0124] (d) Feedback and execution: Decision suggestions (such as adjusting the construction plan) are generated and fed back to the physical construction process. Changes in the physical world are captured by sensors and used as a new data input platform, thus forming a continuous closed loop of "data acquisition - model update - simulation prediction - decision optimization".

[0125] Specific case description of virtual simulation platform:

[0126] This case study addresses a critical safety hazard in multi-story frame structures during the construction phase: abnormal structural response caused by dynamic load changes. In actual construction, processes such as concrete pouring and material stacking often lead to sudden increases in local loads. Traditional manual monitoring struggles to capture real-time stress-strain changes within the structure, let alone predict risk development trends. This case simulates a load surge from 3.0 kN / m² to 6.0 kN / m² during the construction of the 5th floor, causing the strain in the critical column C1_F4 to rapidly rise from the safe range (<280 με) to a dangerous level (>320 με), exceeding the warning threshold for concrete structures. The core problem that a digital twin platform needs to solve is: to achieve a shift from "passive response" to "proactive early warning," completing a closed-loop control process in the digital space—risk identification, quantitative assessment, and decision support—before irreversible damage occurs in the physical structure, through real-time data fusion and virtual simulation.

[0127] like Figure 4-7 As shown, the results are analyzed as follows:

[0128] 1. Three-stage risk evolution and verification of closed-loop control effectiveness

[0129] The strain-time history curve clearly shows three stages: the normal construction period (0-100 seconds, strain stabilizes below 180 με); the abnormal load period (100-200 seconds, strain rapidly exceeds the 300 με warning line, reaching a maximum of 335.2 με); and the implementation period (200-300 seconds, strain falls back to 285 με). The slope of the curve changes significantly before and after the decision-making point (t=200 seconds), proving the effectiveness of adjusting the construction load to 4.0 kN / m² and adding temporary supports, reducing strain by 13.6%. Notably, the displacement response lags the strain by approximately 20-30 seconds, verifying the time-cumulative effect of structural deformation and providing a critical time window for early warning.

[0130] 2. Three-dimensional visualization reveals the spatial propagation mechanism of risk.

[0131] The 3D model's dynamic coloring visually illustrates the risk propagation path: initially, only the critical column C1_F4 turns yellow due to abnormal load (t=120 seconds, 285με); as time progresses, adjacent beam members turn orange (t=150 seconds, 305με); finally, the critical column becomes a red alert (t=180 seconds, 325με). The color gradient changes correspond to the stress concentration level, with the dark red area concentrated at the junction of the 4th and 5th floors, which matches the vertical load transfer path in the frame structure. In the sensor layout diagram, green triangles (strain gauges) are concentrated in stress-sensitive areas, and red squares (displacement gauges) are located at deformation monitoring points; their spatial positions match the structural mechanical characteristics.

[0132] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the technical solutions of the present invention. Any technical solution that can be implemented based on the above embodiments without creative effort should be considered to fall within the scope of protection of the patent of the present invention.

Claims

1. A method for intelligent construction of building structures based on point cloud data and digital twins, characterized in that, Includes the following steps: S1. High-precision point cloud data acquisition and preprocessing of building structure: The target building structure is scanned at multiple stations using a 3D laser scanner to obtain complete on-site point cloud data. Then, the on-site point cloud data is processed sequentially with automatic registration, noise reduction filtering and lightweighting. S2. Reverse modeling of building structure digital twin: Based on preprocessed point cloud data, feature extraction and parametric modeling algorithms are used to automatically identify and reconstruct the main structural components of the building to generate a high-fidelity building information model. S3. Construction of digital twin scenarios through multi-source heterogeneous data fusion: The reverse-generated building information model is unified with geographic information system data, finite element analysis model and Internet of Things sensor data in terms of spatiotemporal reference and semantic integration to construct a dynamically updated digital twin scenario. S4. Virtual simulation and intelligent decision-making for building structure construction: Based on the digital twin scenario, the construction progress and structural mechanical response are simulated and visualized in real time. By comparing and analyzing the simulation data and the measured data, the construction plan is dynamically optimized and intelligent decision support is provided.

2. The intelligent construction method for building structures based on point cloud data and digital twins as described in claim 1, characterized in that, In step S1 The automatic registration adopts a strategy that combines target-assisted coarse registration with the iterative nearest point algorithm to unify the multi-site cloud into the same coordinate system. Specifically, it includes: using a coarse registration algorithm based on least squares fitting for initial positioning; and using an iterative nearest point algorithm for fine registration. The iterative nearest point algorithm uses the sum of the squares of the distances between each point in the source point cloud and the nearest point in the target point cloud after rotation and translation as the registration error. By iteratively optimizing the rotation matrix and translation vector, the overall registration error is controlled within ±2mm. The noise reduction uses a statistical outlier removal algorithm to remove identified noise points. Specifically, it includes: first, calculating the average distance from each point in the point cloud to its preset number of nearest neighbors; if the average distance is greater than the sum of the products of the mean of the average distances of all points, the threshold coefficient, and the standard deviation, then the point is identified as a noise point and removed. The lightweighting process employs a voxel meshing method to uniformly downsample the denoised point cloud.

3. The intelligent construction method for building structures based on point cloud data and digital twins as described in claim 1, characterized in that, Step S2 specifically includes: 2.1) Based on spatial geometric features and Euclidean clustering algorithm, the point cloud is automatically segmented into point sets corresponding to different structural components such as columns, beams and slabs; 2.2) The RANSAC algorithm is used to fit the central axis and cross-sectional contour of the segmented component point cloud. The optimal model parameters are found to maximize the number of data points that meet the condition that the distance from the point to the geometric model to be fitted is less than a preset threshold. 2.3) Optimize the spatial relationship of component connection nodes based on the nearest neighbor fitting algorithm; the optimization objective is to minimize the sum of squared distances between the endpoints of adjacent component axes after rigid body transformation. 2.4) After extracting the geometric parameters of the component, including length, orientation, and cross-sectional dimensions, a parametric three-dimensional solid model is generated using the cross-section sweep method.

4. The intelligent construction method for building structures based on point cloud data and digital twins as described in claim 1, characterized in that, Step S3 specifically includes: 3.1) Construct a unified spatial reference transformation model based on a spherical coordinate system, integrate digital elevation models and orthophotos through the Cesium engine, realize the positional alignment of data from different sources, and build a three-dimensional virtual geographic environment with a real geographic background; 3.2) The Building Information Model is parsed and lightweighted using the IFC standard, converted into glTF / 3D Tiles format, and optimized using the level of detail method to achieve efficient loading, rendering, and visualization in the Web environment; 3.3) Establish a unified data standard and interface, and realize node-level semantic association and dynamic mapping between finite element analysis results and geometric models based on the structured data parsing engine of Parquet columnar storage.

5. The intelligent construction method for building structures based on point cloud data and digital twins as described in claim 1, characterized in that, Step S4 specifically includes: 1) A virtual simulation platform based on the Cesium engine and a B / S architecture was developed, integrating a construction progress simulation module and a stress-displacement simulation module; The construction progress simulation module associates the construction plan with the model components to realize 4D dynamic simulation of the construction process and compare and analyze it with the actual progress; the stress-displacement simulation module integrates a lightweight finite element analysis engine and completes mechanical performance simulation through the finite element control equations composed of structural stiffness matrix, nodal displacement vector, and nodal load vector. 2) WebGL technology is used to achieve multi-dimensional data fusion rendering and immersive interaction of point clouds with hundreds of millions of points, large-scale BIM models and mechanical cloud maps, and to map mechanical simulation results to digital twin models in real time; 3) Establish a closed-loop control process: data acquisition - model correction - simulation prediction - command feedback. The twin model and simulation results are dynamically updated based on IoT measured data; by comparing the simulation predictions with the measured values, potential risks are diagnosed, and data-driven decision-making suggestions are provided for adjusting the construction plan.

Images