Intelligent building method and system based on BIM

Abstract

The invention relates to the technical field of intelligent construction, and discloses a BIM-based intelligent construction method and system, and the method comprises the following steps: S1, deploying a laser scanner, an RFID tag and a displacement sensor; s2, coding sequential logic in the construction plan into a JSON format rule, and importing the JSON format rule into a BIM logic coding unit; s3, establishing an end-to-end data pipeline through 5G SA networking; s4, the cloud server runs the pre-trained three-layer residual convolutional neural network; s5, dynamically adjusting the inspection path of the unmanned aerial vehicle according to the updated BIM coordinates; and S6, monitoring the execution attitude data of the construction machinery in real time through the sensor database. By constructing the BIM and the Internet of Things, construction logic rule codes are embedded into the BIM, dynamic interaction of design-construction data is achieved, two-way association of a physical entity and a virtual model is achieved through the data mapping technology, the mechanical arm is driven to conduct automatic deviation correction, and the construction precision and the automatic construction effect are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent construction technology, and in particular to a BIM-based intelligent construction method and system. Background Art

[0002] Currently, the construction industry is gradually transforming towards digitalization and intelligence. As a core tool, BIM technology has been widely used in the design, construction, and operation and maintenance stages of buildings. Through three-dimensional modeling and data integration, BIM has significantly improved design accuracy and construction efficiency. However, traditional BIM applications still have problems such as data silos and lack of real-time performance, making it difficult to meet the dynamic needs of complex construction scenarios. In recent years, the combination of technologies such as the Internet of Things (IoT) and artificial intelligence (AI) with BIM has become a research hotspot, aiming to achieve automated control and intelligent decision-making in the construction process. However, the construction industry has long relied on manual experience and two-dimensional drawings as the basis for construction decisions.

[0003] However, with current technology, traditional manual experience and two-dimensional drawings are used as the basis for construction, resulting in low process coordination efficiency, difficult to control error transmission, and delayed response to safety risks. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a BIM-based intelligent construction method and system to solve the problems of low process collaboration efficiency, difficult error transmission control and delayed safety risk response.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a BIM-based intelligent construction method and system, comprising the following steps: S1. Deploy laser scanners, RFID tags, and displacement sensors, and calibrate the coordinate systems of the laser scanners and the BIM model. S2. Encode the temporal logic in the construction plan into JSON format rules, import it into the BIM logic coding unit, and establish a one-to-one binding relationship between the sensor ID and the BIM model component GUID; S3: Establish an end-to-end data pipeline through 5G SA networking, deploy a rule engine on the edge computing gateway, and perform point cloud data denoising, coordinate transformation, and feature extraction; S4, the cloud server runs a pre-trained 3-layer residual convolutional neural network to identify components with construction deviations and generate PID control parameters for the construction machinery; S5. Dynamically adjust the drone inspection path based on the updated BIM coordinates; S6. Real-time monitoring of the execution posture data of construction machinery is carried out through the sensor database. When the deviation exceeds the limit for three consecutive sampling cycles, the system switches to the manual intervention mode and triggers the sound and light alarm.

[0006] Preferably, the laser scanner in S1 is deployed on the top truss of the construction area in a 15m×15m grid, and the timing logic includes the component lifting sequence and the concrete pouring time.

[0007] Preferably, the RFID tag in S1 uses the UHF frequency band, has a reading and writing distance of ≥10m, and is bound to the material properties of the components in the BIM model.

[0008] Preferably, the step S2 specifically includes the following steps: S201. Convert the process sequence logic in the construction plan into a JSON-LD structured data format, including the component hoisting sequence, concrete pouring time window, and quality acceptance criteria, and import it into the BIM logic coding unit through the Revit API interface; S202. Add an extended attribute set to each component in the BIM model and define the construction rule coding field, including: Safe operating radius threshold; Ambient temperature and humidity control range; Multi-disciplinary collaborative conflict detection rules; S203, using a bidirectional hash mapping algorithm to bind the sensor ID to the BIM component, including: Generate a unique encrypted identifier for each RFID tag; Create a sensor registry in the BIM model properties and associate the component GUID with the sensor MAC address; The integrity of the binding relationship is verified through digital signature.

[0009] Preferably, the step S3 specifically includes the following steps: S301, based on the 5G SA independent networking architecture, configures end-to-end network slicing to transmit laser scanner point cloud data and sensor readings in real time; S302. Deploy a rule engine on the edge computing gateway to perform multi-level data processing: Denoising: An improved RANSAC algorithm is used to remove outliers. The improved RANSAC algorithm achieves efficient outlier removal through the following steps: Normal vector consistency pre-screening: By analyzing the normal vector angle variance of the local area of ​​the point cloud, normal vector mutation points are pre-eliminated, reducing the candidate point set to 30% to 40% of the original data; Dynamic threshold adaptive iteration adjusts the distance threshold in real time based on point cloud density, balancing noise suppression and feature retention requirements. Curvature-constrained region growing merges continuous regions based on octree index, curvature difference, and normal vector angle, ultimately removing isolated points that cannot be included in any region. Coordinate transformation: Convert WGS84 coordinate system point cloud data into BIM local coordinate system and write it into the time series database synchronously; Feature extraction: Extract component geometric features based on PCA principal component analysis and generate feature vector dimensions. S3 specifically includes the following steps: S301, based on the 5G SA independent networking architecture, configures end-to-end network slicing to transmit laser scanner point cloud data and sensor readings in real time; S302. Deploy a rule engine on the edge computing gateway to perform multi-level data processing: Denoising: An improved RANSAC algorithm is used to remove outliers. The improved RANSAC algorithm achieves efficient outlier removal through the following steps: Normal vector consistency pre-screening, calculate the normal vector angle variance of the local area of ​​the point cloud, pre-eliminate normal vector mutation points, and reduce the candidate point set to 30% to 40% of the original data; Dynamic threshold adaptive iteration adjusts the distance threshold in real time based on point cloud density, balancing noise suppression and feature retention requirements. Curvature-constrained region growing merges continuous regions based on octree index, curvature difference, and normal vector angle, ultimately removing isolated points that cannot be included in any region. Coordinate transformation: Convert WGS84 coordinate system point cloud data into BIM local coordinate system and write it into time series database simultaneously; Feature extraction: Extract component geometric features based on PCA principal component analysis and generate feature vector dimensions; S303: After the difference matrix is ​​normalized, it is input into the pre-trained 3-layer residual convolutional neural network.

[0010] Preferably, the S4 specifically includes the following steps: S401, normalize the difference matrix, perform Z-score normalization on the difference matrix between the BIM model and the construction point cloud, eliminate the dimension difference, and generate normalized input data with a mean of 0 and a standard deviation of 1. The Z-score normalization formula is: Where x represents the difference between the theoretical coordinates and the actual point cloud coordinates of a component in the BIM model; μ is the mean of all data in the difference matrix, reflecting the overall average level of construction deviation; σ is the standard deviation of the difference matrix, representing the discrete fluctuation degree of construction deviation; x ′ is the normalized value after standardization; S402, forward inference of the residual convolutional neural network, inputs the normalized data into a pre-trained 3-layer residual network, each layer includes a convolution kernel, batch normalization and ReLU activation function, extracts deep spatial features through residual skip connections, and outputs the component position deviation classification result and deviation amount; S403, PID parameters are dynamically generated. Based on the deviation and historical control data, PID control parameters are generated through full-connection layer mapping. The parameter combination is optimized by combining the gradient descent algorithm to ensure that the joint angle adjustment of the robot arm meets the construction correction accuracy, and the control instructions are issued in real time.

[0011] A BIM-based intelligent construction system, comprising: The system comprises: Data acquisition and calibration module, used to deploy laser scanners, RFID tags and displacement sensors, and calibrate the equipment coordinate system and BIM model benchmark; The logic coding and binding module is used to encode the construction timing logic into JSON format rules and establish the binding relationship between the sensor ID and the BIM component GUID; Edge computing processing module, used to receive sensor data through 5G SA networking and perform point cloud denoising, coordinate transformation, and feature extraction; A cloud-based intelligent analysis module runs a residual convolutional neural network to identify construction deviations and generate PID control parameters for construction machinery; Dynamic path planning module, used to adjust the drone inspection path according to real-time BIM coordinates and avoid dynamic obstacles; The safety monitoring and intervention module is used to monitor the posture data of construction machinery, trigger over-limit alarms and manual intervention.

[0012] Preferably, in the data acquisition and calibration module, displacement sensors are arranged at key nodes of the steel structure, and data is transmitted through the TSN network.

[0013] Preferably, the edge computing processing module includes: Improved RANSAC algorithm unit, through normal vector consistency pre-screening, dynamic threshold iteration and curvature constrained region growing; coordinate conversion unit, converts WGS84 coordinate system point cloud into BIM local coordinate system; The feature extraction unit generates a 128-dimensional feature vector based on PCA principal component analysis and writes it into the time series data.

[0014] Preferably, the cloud-based intelligent analysis module includes: The difference matrix normalization unit processes the input data through Z-score standardization to eliminate dimensional differences; Residual convolutional network unit, a 3-layer residual structure, each layer contains 3×3×64 convolution kernels and batch normalization layers, outputs bias classification results; PID parameter generation unit maps the deviation through the fully connected layer.

[0015] The present invention provides a BIM-based intelligent construction method and system, which has the following beneficial effects: 1. The present invention builds BIM and the Internet of Things, embeds the construction logic rule code into the BIM model, realizes the dynamic interaction of design and construction data, and uses data mapping technology to establish a bidirectional association between physical entities and virtual models, drives the robotic arm to correct errors autonomously, and significantly improves construction accuracy and the effect of automated construction.

[0016] 2. The present invention builds an edge-cloud collaborative architecture based on 5G networks and time-sensitive networks. The edge side completes real-time preprocessing such as point cloud denoising and coordinate conversion. The cloud side analyzes construction deviations and generates control instructions through deep learning, achieving efficient transmission of massive data and millisecond-level response, breaking through the bottleneck of traditional single-point computing, reducing cloud load, and improving dynamic decision-making efficiency in complex scenarios.

[0017] 3. This invention optimizes algorithm performance to address challenges such as construction noise and micro-deviation identification through high-precision point cloud registration, dynamic obstacle avoidance path planning, and anti-interference neural networks. It achieves precise data analysis and equipment control in complex environments, forming a complete technology chain from perception to execution, significantly improving construction efficiency and engineering safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flowchart of a BIM-based intelligent construction method of the present invention; Figure 2 This is an architectural diagram of a BIM-based intelligent construction system of the present invention. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solution of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0020] Please see the attached Figure 1 , an embodiment of the present invention provides a BIM-based intelligent construction method, comprising the following steps: S1. Deploy laser scanners, RFID tags, and displacement sensors, and calibrate the coordinate systems of the laser scanners and the BIM model. S2. Encode the temporal logic in the construction plan into JSON format rules, import it into the BIM logic coding unit, and establish a one-to-one binding relationship between the sensor ID and the BIM model component GUID; S3: Establish an end-to-end data pipeline through 5G SA networking, deploy a rule engine on the edge computing gateway, and perform point cloud data denoising, coordinate transformation, and feature extraction; S4, the cloud server runs a pre-trained 3-layer residual convolutional neural network to identify components with construction deviations and generate PID control parameters for the construction machinery; S5. Dynamically adjust the drone inspection path based on the updated BIM coordinates; S6. Real-time monitoring of the execution posture data of construction machinery is carried out through the sensor database. When the deviation exceeds the limit for three consecutive sampling cycles, the system switches to the manual intervention mode and triggers the sound and light alarm.

[0021] Specifically, S1 deploys laser scanners, RFID tags, and displacement sensors to build a multi-dimensional perception network. It also calibrates the device coordinate system and BIM model benchmark based on the ICP algorithm, achieving millimeter-level digital mapping of construction scenes. This eliminates the temporal and spatial differences between BIM models and on-site data in traditional construction, resolves data silos, ensures dynamic synchronization between design models and actual working conditions, and forms a full-domain real-time monitoring capability. The time required to identify construction errors is shortened from hours to minutes, supporting subsequent AI decision-making cycles and providing a precise data foundation for intelligent construction. S2 encodes the construction timing logic into JSON format rules and imports them into the BIM logic coding unit, converting the process flow (such as the component lifting sequence and the concrete pouring window) into machine-executable standardized instructions, realizing the digital expression and automated scheduling of construction logic. At the same time, by establishing a one-to-one binding relationship between the sensor ID (such as the RFID tag MAC address) and the BIM component GUID, it opens up a precise mapping channel between the physical entity and the virtual model. The effect is to eliminate the response delay and information mismatch caused by manual coordination in traditional construction, and support the real-time linkage update of multi-source data (design parameters, construction progress, monitoring data), providing a structured rule engine for dynamic decision-making; S3 uses 5G SA independent networking to build a highly reliable data channel. Combined with a rule engine deployed on an edge computing gateway, it enables real-time preprocessing of laser point cloud data. It uses an improved RANSAC algorithm to remove outlier noise points, achieve millimeter-level conversion from WGS84 to BIM coordinates (accuracy ±1mm), and extract 128-dimensional feature vectors based on PCA. This ensures real-time construction deviation analysis and reduces the dynamic adjustment response speed of construction machinery to 5 seconds, providing standardized data input with a high signal-to-noise ratio for subsequent AI decision-making. S4 uses a three-layer residual convolutional neural network deployed in the cloud to analyze the difference matrix between the BIM model and the construction point cloud in real time, accurately identifying abnormal working conditions of component positions. It also generates PID control parameters for construction machinery based on historical data training, transforming traditional error correction decisions that rely on manual experience into data-driven automated control, achieving a millisecond-level response closed loop for construction errors. Based on real-time updated BIM coordinates, the S5 achieves adaptive 3D spatial scanning coverage of construction area changes, improving drone blind spot coverage and synchronously updating construction progress image data in the BIM model. This reduces quality acceptance response time from 12 hours for manual inspections to real-time feedback, while also avoiding collision risks with dynamic obstacles on site, ensuring fully autonomous and safe drone operations in complex construction environments. S6 monitors the posture data of construction machinery in real time through the sensor database. When the posture angle deviation exceeds ±0.5° or the displacement deviation exceeds ±10mm for three consecutive sampling cycles, it automatically switches to manual intervention mode and triggers an audible and visual alarm, reducing the accident rate of mechanical operation, ensuring construction continuity while optimizing manpower input, and forming a fully automated safety management and control closed loop of "real-time monitoring-threshold judgment-graded response".

[0022] In S1, laser scanners were deployed on the top trusses of the construction area in a 15m×15m grid, and the timing logic included the component hoisting sequence and concrete pouring time; Specifically, the laser scanner in S1 is deployed in a 15m×15m grid on the top truss of the construction site. Combined with timing logic (component lifting sequence, concrete pouring time window), it builds a full-domain millimeter-level real-time perception network, accurately captures construction dynamics through dense point cloud coverage, and synchronously calibrates the time and space benchmarks of the BIM model and the actual working conditions, eliminating the data gap between design and implementation in traditional construction. After implementation, it can reduce the vertical deviation of steel structure lifting and improve the quality qualification rate of concrete pouring. It also digitally previews the construction process through timing logic, providing a high-precision, full-factor digital chassis for intelligent construction.

[0023] The RFID tag in S1 uses the UHF frequency band, with a reading and writing distance of ≥10m, and is bound to the material properties of the components in the BIM model.

[0024] Specifically, S1 uses UHF band RFID tags and binds them to the material properties of components in the BIM model (such as concrete grade and steel specifications). Through wireless radio frequency identification technology, it realizes the precise association between building material identity information and digital models, thereby opening up the data link between physical components and virtual models, supporting the automatic tracking of material transportation trajectories during construction, real-time updating of inventory status, and reverse positioning of quality traceability, solving the problems of progress management lag caused by errors and omissions in traditional manual records and information islands, and providing a visual decision-making basis based on the BIM model for construction resource scheduling, forming a closed loop of material life cycle management.

[0025] S2 specifically includes the following steps: S201. Convert the process sequence logic in the construction plan into a JSON-LD structured data format, including the component hoisting sequence, concrete pouring time window, and quality acceptance criteria, and import it into the BIM logic coding unit through the Revit API interface; S202. Add an extended attribute set to each component in the BIM model and define the construction rule coding field, including: Safe operating radius threshold; Ambient temperature and humidity control range; Multi-disciplinary collaborative conflict detection rules; S203, using a bidirectional hash mapping algorithm to bind the sensor ID to the BIM component, including: Generate a unique encrypted identifier for each RFID tag; Create a sensor registry in the BIM model properties and associate the component GUID with the sensor MAC address; The integrity of the binding relationship is verified through digital signature.

[0026] Specifically, S201 converts the process sequence logic into the JSON-LD structured data format and imports it into BIM logic units through the Revit API, achieving machine-readable docking between the construction plan and the digital model. This eliminates the semantic ambiguity of manual interpretation of drawings in traditional construction, drives construction process automation through standardized data interfaces, ensures that the process logic is parseable and executable in the BIM environment, and provides a structured rule library for subsequent dynamic scheduling. S202 defines an extended attribute set (safety radius, environmental threshold, conflict rules) in the BIM model, encodes construction specifications as built-in constraints in the model, embeds scattered construction rules (such as safety specifications and professional collaboration requirements) into BIM component attributes, and builds a model-based integrated design-construction verification mechanism. This triggers real-time violation warnings (such as construction machinery crossing the boundary and pipeline collision) to prevent construction safety risks and quality defects at the source. S203 binds sensors and BIM components through bidirectional hash mapping and digital signature technology to establish a bidirectional traceability link between physical entities and virtual models. It is associated with the registry through an encrypted identifier (RFID tag) to ensure the dynamic synchronization of data such as building materials transportation trajectory and installation status with model components. At the same time, the anti-tampering feature of digital signatures is used to ensure data credibility, providing an irrefutable data evidence chain for quality traceability and responsibility definition.

[0027] S3 specifically includes the following steps: S301, based on the 5G SA independent networking architecture, configures end-to-end network slicing to transmit laser scanner point cloud data and sensor readings in real time; S302. Deploy a rule engine on the edge computing gateway to perform multi-level data processing: Denoising: The improved RANSAC algorithm is used to remove outliers. The improved RANSAC algorithm can achieve efficient outlier removal through the following steps: Normal vector consistency pre-screening: By analyzing the normal vector angle variance of the local area of ​​the point cloud, normal vector mutation points are pre-eliminated, reducing the candidate point set to 30% to 40% of the original data; Dynamic threshold adaptive iteration adjusts the distance threshold in real time based on point cloud density, balancing noise suppression and feature retention requirements. Curvature-constrained region growing merges continuous regions based on octree index, curvature difference, and normal vector angle, ultimately removing isolated points that cannot be included in any region. Coordinate transformation: Convert WGS84 coordinate system point cloud data into the BIM local coordinate system and write it into the time series database simultaneously; Feature extraction: Extract component geometric features based on PCA principal component analysis and generate feature vector dimensions.

[0028] Specifically, the 5G SA independent networking architecture in S301 is configured with end-to-end network slicing (eMBB slicing, bandwidth ≥ 100Mbps), providing a highly reliable, low-latency transmission channel for laser scanner point cloud data (single device rate ≥ 50Mbps) and sensor readings. This ensures real-time synchronous transmission of massive data across the entire construction area, solves the problem of model update lag caused by data packet loss rate > 5%, and lays the underlying communication foundation for edge-cloud collaborative computing. S302 deploys a rule engine on the edge computing gateway, implements point cloud denoising and compression through an improved RANSAC algorithm, converts WGS84 coordinates to the BIM local coordinate system, and extracts 128-dimensional geometric feature vectors based on PCA. It pre-processes the original data and generates standardized and lightweight feature inputs to avoid the waste of computing resources of directly processing the original point cloud on the cloud. At the same time, it records spatiotemporal correlation data through a time series database to provide a structured data set for construction progress backtracking and quality analysis.

[0029] S4 specifically includes the following steps: S401, normalize the difference matrix, perform Z-score normalization on the difference matrix between the BIM model and the construction point cloud, eliminate the dimension difference, and generate normalized input data with a mean of 0 and a standard deviation of 1. The Z-score normalization formula is: Where x represents the difference between the theoretical coordinates and the actual point cloud coordinates of a component in the BIM model; μ is the mean of all data in the difference matrix, reflecting the overall average level of construction deviation; σ is the standard deviation of the difference matrix, representing the discrete fluctuation degree of construction deviation; x ′ is the normalized value after standardization; S402, forward inference of the residual convolutional neural network, inputs the normalized data into a pre-trained 3-layer residual network, each layer includes a convolution kernel, batch normalization and ReLU activation function, extracts deep spatial features through residual skip connections, and outputs the component position deviation classification result and deviation amount; S403, PID parameters are dynamically generated. Based on the deviation and historical control data, PID control parameters are generated through full-connection layer mapping. The parameter combination is optimized by combining the gradient descent algorithm to ensure that the joint angle adjustment of the robot arm meets the construction correction accuracy, and the control instructions are issued in real time.

[0030] Specifically, S401 inputs the normalized (Z-score standardization) difference matrix into a pre-trained three-layer residual convolutional neural network, alleviates the gradient vanishing problem through residual connections, extracts deep features to map construction deviations, and upgrades traditional manual visual inspection to AI-driven millimeter-level precision automatic recognition. Through end-to-end training, it establishes a nonlinear mapping relationship between the difference matrix and mechanical control parameters, realizes the intelligent conversion from data features to execution instructions, and supports a second-level response closed loop for construction deviation correction. S402 uses a pre-trained three-layer residual convolutional neural network to extract deep features from normalized difference data, and retains shallow geometric features and integrates deep abstract features through residual jump connections. It combines batch normalization and ReLU activation function to enhance model robustness, and finally outputs the component position deviation classification results and precise deviation amount, realizing an intelligent closed-loop from data to decision-making; S403 realizes intelligent closed-loop control of construction machinery by dynamically generating PID control parameters. Based on the deviation amount and historical control data output by the neural network, the fully connected layer directly maps the generated parameters, and combines the gradient descent algorithm to online optimize the parameter combination, so that the joint angle adjustment amount of the robot arm accurately matches the construction correction requirements, and overcomes nonlinear interference under different working conditions through parameter adaptive adjustment, finally forming a millisecond-level intelligent control closed loop of perception-decision-execution.

[0031] Please see the attached Figure 2 , a BIM-based intelligent construction system, the system includes: Data acquisition and calibration module, used to deploy laser scanners, RFID tags and displacement sensors, and calibrate the equipment coordinate system and BIM model benchmark; The logic coding and binding module is used to encode the construction timing logic into JSON format rules and establish the binding relationship between the sensor ID and the BIM component GUID; Edge computing processing module, used to receive sensor data through 5G SA networking and perform point cloud denoising, coordinate transformation, and feature extraction; A cloud-based intelligent analysis module runs a residual convolutional neural network to identify construction deviations and generate PID control parameters for construction machinery; Dynamic path planning module, used to adjust the drone inspection path according to real-time BIM coordinates and avoid dynamic obstacles; The safety monitoring and intervention module is used to monitor the posture data of construction machinery, trigger over-limit alarms and manual intervention.

[0032] Specifically, a complete intelligent construction management system was constructed through the collaboration of multiple modules. Its core value lies in the realization of comprehensive digitalization and intelligent control of the construction process. First, real-time mapping of the physical construction site and the digital model is established through high-precision sensing equipment. Then, with the help of edge computing and cloud-based intelligent analysis, real-time processing and decision-making of construction data are realized. Finally, closed-loop management is formed through automated equipment control and safety monitoring, which effectively solves the problems of data silos, manual dependence and response lag in traditional construction, and transforms construction management from experience-driven to data-driven, which significantly improves the accuracy, safety and efficiency of engineering construction, and provides a complete solution for modern intelligent construction.

[0033] In the data acquisition and calibration module, displacement sensors are deployed at key nodes of the steel structure, and data is transmitted via the TSN network; Specifically, displacement sensors are deployed at key nodes of the steel structure and transmit data through the TSN network, realizing high-precision dynamic monitoring and real-time feedback control of the entire steel structure construction process. That is, micron-level displacement sensing technology is used to capture the deformation and displacement changes in key processes such as steel structure lifting and welding in real time. Combined with the time-sensitive transmission characteristics of the TSN network, it ensures millisecond-level synchronization of monitoring data and the BIM model coordinate system, thereby effectively preventing the accumulation of steel structure installation deviations.

[0034] The edge computing processing module includes: Improved RANSAC algorithm unit, through normal vector consistency pre-screening, dynamic threshold iteration and curvature constrained region growing; coordinate conversion unit, converts WGS84 coordinate system point cloud into BIM local coordinate system; The feature extraction unit generates a 128-dimensional feature vector based on PCA principal component analysis and writes it into the time series data.

[0035] Specifically, the improved RANSAC algorithm unit uses normal vector consistency pre-screening combined with dynamic threshold iteration to effectively eliminate environmental noise points while retaining the main structure feature points; the coordinate conversion unit uses matrix transformation to achieve millimeter-level precise mapping from the WGS84 global coordinate system to the BIM local coordinate system, ensuring the consistency of virtual and real space data; the feature extraction unit generates a 128-dimensional feature vector through PCA dimensionality reduction, providing standardized input for cloud-based intelligent analysis.

[0036] The cloud-based intelligent analysis module includes: The difference matrix normalization unit processes the input data through Z-score standardization to eliminate dimensional differences; Residual convolutional network unit, a 3-layer residual structure, each layer contains 3×3×64 convolution kernels and batch normalization layers, outputs bias classification results; PID parameter generation unit maps the deviation through the fully connected layer.

[0037] Specifically, the difference matrix normalization unit uses Z-score standardization to eliminate the dimensional differences of multi-source data and improve the consistency of input feature distribution; the residual convolutional network unit extracts deep spatial features through a three-layer residual structure to achieve micro-deviation identification; the PID parameter generation unit maps the deviation to the optimal control parameter in real time based on the fully connected layer, and overcomes interference from different working conditions through adaptive parameter adjustment.

[0038] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A BIM-based intelligent construction method, characterized in that: The following steps are involved: S1. Deploy laser scanners, RFID tags, and displacement sensors, and calibrate the coordinate systems of the laser scanners and the BIM model. S2. Encode the temporal logic in the construction plan into JSON format rules, import it into the BIM logic coding unit, and establish a one-to-one binding relationship between the sensor ID and the BIM model component GUID; S3: Establish an end-to-end data pipeline through 5G SA networking, deploy a rule engine on the edge computing gateway, and perform point cloud data denoising, coordinate transformation, and feature extraction; S4, the cloud server runs a pre-trained 3-layer residual convolutional neural network to identify components with construction deviations and generate PID control parameters for the construction machinery; S5. Dynamically adjust the drone inspection path based on the updated BIM coordinates; S6. Real-time monitoring of the execution posture data of construction machinery is carried out through the sensor database. When the deviation exceeds the limit for three consecutive sampling cycles, the system switches to the manual intervention mode and triggers the sound and light alarm.

2. The BIM-based intelligent construction method according to claim 1, characterized in that: In S1, the laser scanner is deployed on the top truss of the construction area in a 15m×15m grid. The timing logic includes the component lifting sequence and the concrete pouring time.

3. The BIM-based intelligent construction method according to claim 1, characterized in that: The RFID tag in S1 uses the UHF frequency band, has a reading and writing distance of ≥10m, and is bound to the material properties of the components in the BIM model.

4. The BIM-based intelligent construction method according to claim 1, characterized in that: The S2 specifically includes the following steps: S201. Convert the process sequence logic in the construction plan into a JSON-LD structured data format, including the component hoisting sequence, concrete pouring time window, and quality acceptance criteria, and import it into the BIM logic coding unit through the Revit API interface; S202. Add an extended attribute set to each component in the BIM model and define the construction rule coding field, including: Safe operating radius threshold; Ambient temperature and humidity control range; Multi-disciplinary collaborative conflict detection rules; S203, using a bidirectional hash mapping algorithm to bind the sensor ID to the BIM component, including: Generate a unique encrypted identifier for each RFID tag; Create a sensor registry in the BIM model properties and associate the component GUID with the sensor MAC address; The integrity of the binding relationship is verified through digital signature.

5. The BIM-based intelligent construction method according to claim 1, characterized in that: The S3 specifically includes the following steps: S301, based on the 5G SA independent networking architecture, configures end-to-end network slicing to transmit laser scanner point cloud data and sensor readings in real time; S302. Deploy a rule engine on the edge computing gateway to perform multi-level data processing: Denoising: An improved RANSAC algorithm is used to remove outliers. The improved RANSAC algorithm achieves efficient outlier removal through the following steps: Normal vector consistency pre-screening: By analyzing the normal vector angle variance of the local area of ​​the point cloud, normal vector mutation points are pre-eliminated, reducing the candidate point set to 30% to 40% of the original data; Dynamic threshold adaptive iteration adjusts the distance threshold in real time according to the point cloud density, balancing the requirements of noise suppression and feature retention; Curvature-constrained region growing merges continuous regions based on octree index, curvature difference, and normal vector angle, ultimately removing isolated points that cannot be included in any region; Coordinate transformation: Convert WGS84 coordinate system point cloud data into BIM local coordinate system and write it into the time series database simultaneously; Feature extraction: Extract component geometric features based on PCA principal component analysis and generate feature vector dimensions.

6. The BIM-based intelligent construction method according to claim 1, characterized in that: The S4 specifically includes the following steps: S401, normalize the difference matrix, perform Z-score normalization on the difference matrix between the BIM model and the construction point cloud, eliminate the dimension difference, and generate normalized input data with a mean of 0 and a standard deviation of 1. The Z-score normalization formula is: Where x represents the difference between the theoretical coordinates and the actual point cloud coordinates of a component in the BIM model; μ is the mean of all data in the difference matrix, reflecting the overall average level of construction deviation; σ is the standard deviation of the difference matrix, representing the discrete fluctuation degree of construction deviation; x ′ is the normalized value after standardization; S402, forward inference of the residual convolutional neural network, inputs the normalized data into a pre-trained 3-layer residual network, each layer includes a convolution kernel, batch normalization and ReLU activation function, extracts deep spatial features through residual skip connections, and outputs the component position deviation classification result and deviation amount; S403, PID parameters are dynamically generated. Based on the deviation and historical control data, PID control parameters are generated through full-connection layer mapping. The parameter combination is optimized by combining the gradient descent algorithm to ensure that the joint angle adjustment of the robot arm meets the construction correction accuracy, and the control instructions are issued in real time.

7. A BIM-based intelligent construction system, characterized in that: A BIM-based intelligent construction method according to claims 1-6, wherein the system comprises: Data acquisition and calibration module, used to deploy laser scanners, RFID tags and displacement sensors, and calibrate the equipment coordinate system and BIM model benchmark; The logic coding and binding module is used to encode the construction timing logic into JSON format rules and establish the binding relationship between the sensor ID and the BIM component GUID; Edge computing processing module, used to receive sensor data through 5G SA networking and perform point cloud denoising, coordinate transformation, and feature extraction; A cloud-based intelligent analysis module runs a residual convolutional neural network to identify construction deviations and generate PID control parameters for construction machinery; Dynamic path planning module, used to adjust the drone inspection path according to real-time BIM coordinates and avoid dynamic obstacles; The safety monitoring and intervention module is used to monitor the posture data of construction machinery, trigger over-limit alarms and manual intervention.

8. The BIM-based intelligent construction system according to claim 7, characterized in that: In the data acquisition and calibration module, displacement sensors are deployed at key nodes of the steel structure, and data is transmitted through the TSN network.

9. The BIM-based intelligent construction system according to claim 7, characterized in that: The edge computing processing module includes: Improved RANSAC algorithm unit, through normal vector consistency pre-screening, dynamic threshold iteration and curvature constrained region growing; coordinate conversion unit, converts WGS84 coordinate system point cloud into BIM local coordinate system; The feature extraction unit generates a 128-dimensional feature vector based on PCA principal component analysis and writes it into the time series data.

10. The BIM-based intelligent construction system according to claim 7, characterized in that: The cloud-based intelligent analysis module includes: The difference matrix normalization unit processes the input data through Z-score standardization to eliminate dimensional differences; Residual convolutional network unit, a 3-layer residual structure, each layer contains 3×3×64 convolution kernels and batch normalization layers, outputs bias classification results; PID parameter generation unit maps the deviation through the fully connected layer.

Images