Intelligent construction system and method based on digital twinning

Abstract

The invention relates to the technical field of building intellectualization, in particular to an intelligent building system and method based on digital twinning, and the system comprises the following modules: a data collection module; a data processing module; a template comparison module; according to the scheme, by adopting a comparison algorithm based on feature matching, the collected target construction data is compared with the construction templates in the template library after being processed and subjected to feature extraction, the difference between the actual construction process and the ideal model can be accurately judged, and meanwhile, the construction efficiency is improved. According to the method, target construction data is deeply fused by using data fusion technologies such as Kalman filtering, a new digital twinborn model is constructed by using a reverse engineering method based on the fused data, and the new model is enabled to better fit an actual construction condition through strict verification and optimization.

Description

Technical Field

[0001] The present invention relates to the field of building intelligent technology, and in particular to an intelligent construction system and method based on digital twins. Background Art

[0002] Against the backdrop of the rapid development of the current construction industry, traditional construction methods have gradually exposed numerous problems. As building scale continues to expand, structures become increasingly complex, and functional requirements become increasingly diverse, traditional construction models that rely on manual experience and extensive management are unable to meet the efficient, precise, and green construction requirements of modern construction projects. At the same time, the application of digital technology in the construction field is becoming increasingly widespread. Digital twin technology, with its advantages such as precise mapping of physical entities and real-time monitoring, has brought new opportunities for the intelligent transformation of the construction industry. An increasing number of construction companies are beginning to explore the use of digital technology to improve the management and control of the construction process, in order to achieve goals such as optimizing resource allocation, improving construction quality, and ensuring construction safety.

[0003] However, existing construction systems still suffer from numerous drawbacks when applying digital twin technology. On the one hand, data collection methods are limited and lack high precision. Traditional structural monitoring methods cannot accurately measure displacement and stress changes at structural nodes in real time, resulting in a delay in identifying potential structural safety hazards. On the other hand, in terms of data processing and model building, there is a lack of effective data cleaning, feature extraction, and model updating mechanisms. Existing systems struggle to efficiently process and analyze large amounts of complex construction data, unable to accurately determine the differences between the actual construction process and the ideal model, and consequently unable to promptly adjust construction strategies and optimize digital twin models. Therefore, we propose an intelligent construction system and method based on digital twins. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides an intelligent construction system and method based on digital twins, thereby solving the technical problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0006] An intelligent construction system based on digital twins, including the following modules:

[0007] Data acquisition module: used to collect various data during the construction process, including sensor data acquisition submodule and image data acquisition submodule. The sensor data acquisition submodule is used to monitor changes in the mechanical properties of the structure in real time; the image data acquisition submodule uses the autofocus, night vision and high-resolution features of a fixed camera to capture personnel and equipment activities and construction progress;

[0008] Data processing module: cleans, preprocesses and extracts features from the collected data, including a data cleaning and preprocessing submodule and a data feature extraction submodule. The data cleaning and preprocessing submodule uses a sliding average filter algorithm, a statistical method and an interpolation method to process sensor data, a median filter algorithm and morphological operations to process image data, and standardizes and normalizes the cleaned data. The data feature extraction submodule uses a convolutional neural network to extract low-level and high-level features from image data based on derived features.

[0009] Template comparison module: This module includes a template library management submodule and a comparison submodule. The template library management submodule establishes a template library and stores the digital twin models of different construction projects in IFC format according to building type, structural form, and construction stage. The comparison submodule uses a feature matching-based comparison algorithm and a deep learning metric learning algorithm to calculate the similarity of image data features. The two are combined to determine the degree of match between the target building and the template, and thresholds for various data features are set at the same time.

[0010] Model update module: It consists of a new template generation submodule and a real-time update submodule. The new template generation submodule uses data fusion technology to integrate data, adopts reverse engineering methods to build a new digital twin model and stores it in the template library in IFC format. During the construction and operation and maintenance process, the real-time update submodule uses an incremental update algorithm to update the affected parts of the model according to the new data to ensure the consistency and integrity of the model.

[0011] In a possible implementation, in the data cleaning and preprocessing submodule, a sliding average filtering algorithm is used to remove noise. Assume that the data sequence collected by the sensor is x1, x2, ..., x n , the result y after sliding average filtering i for Where k is the size of the sliding window.

[0012] In a possible implementation, in the comparison submodule, the target data feature vector is The template data feature vector is The Euclidean distance d is: The cosine similarity sim is:

[0013] In a possible implementation, the threshold is calculated based on the total duration T of the project and the planned duration t of each stage. i To determine, suppose the total construction period of a project is T, and the planned construction period is t i days, when the actual construction days t exceeds 10% of the planned construction period, that is, t>1.1t i, it is determined that the progress of this construction phase is abnormal, and the system issues an early warning to remind the construction personnel to adjust the construction plan and resource allocation in time.

[0014] In one possible implementation, a smart construction method based on digital twins includes the following steps:

[0015] Step 1: Preliminary preparation: Based on the project design plan, determine the type, quantity, and installation layout of sensors and image acquisition equipment, build data acquisition, transmission, and storage infrastructure, select transmission modules, use a distributed storage system to classify and store data, and select similar digital twin template models from the template library;

[0016] Step 2: Data collection and preliminary processing. After construction begins, the data collection module is started. Various sensors collect data, and fixed cameras and drones collect image data. The collected raw data is initially sorted and cached before being transmitted to the data storage server.

[0017] Step 3: Data in-depth processing and feature extraction. After the data is transferred to the data processing module, data cleaning, preprocessing, standardization, and normalization are performed, and then various data features are extracted;

[0018] Step 4: Template comparison and threshold judgment: Compare the processed feature data with the selected template model, calculate the comprehensive matching degree, and set the threshold according to the construction industry standards, historical project experience, and actual project needs. When the comprehensive matching degree exceeds the preset threshold, the model update process is triggered;

[0019] Step 5: Model update and construction decision optimization. If the model update conditions are triggered, the model update module is started to build a new model and update it in real time to ensure model consistency and integrity. The updated model is used to support construction decisions. The above steps are repeated throughout the construction process to achieve dynamic monitoring and optimization.

[0020] Beneficial effects compared with existing technologies:

[0021] 1. In this solution, by using a feature matching-based comparison algorithm, the collected target construction data is processed and feature extracted, and then compared with the construction templates in the template library, accurately determining the difference between the actual construction process and the ideal model. At the same time, data fusion technologies such as Kalman filtering are used to deeply fuse the target construction data. Based on the fused data, a new digital twin model is constructed using reverse engineering methods. Through rigorous verification and optimization, the new model is made more consistent with the actual construction situation. This precise judgment of the model and generation of the new model effectively improves the accuracy and practicality of the digital twin model.

[0022] 2. This solution achieves comprehensive, real-time collection of multi-source data from the construction process by installing structural sensors at key points within the building structure, deploying environmental sensors at the construction site and within the building, and using fixed cameras and drones to collect image data. This rich data provides a sufficient information foundation for model construction and analysis.

[0023] 3. This solution uses an incremental update algorithm in the model update module. As new data is continuously collected during construction and maintenance, only the affected parts of the model are updated, while ensuring model consistency and integrity. This not only improves the efficiency of model updates and reduces the waste of computing resources, but also reflects changes in building structure and construction status in real time based on the latest data. The updated model provides more timely and effective support for construction decision-making, optimizes resource allocation, and ensures the smooth progress of the project. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention with reference to the accompanying drawings.

[0025] Figure 1 Schematic diagram of the system framework of the digital twin-based intelligent construction system of the present invention;

[0026] Figure 2 This is a flow chart of the method of the digital twin-based intelligent construction method of the present invention. DETAILED DESCRIPTION

[0027] Preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention can be implemented in various forms, and therefore the present invention is not limited to the embodiments described below. In addition, in order to more clearly describe the present invention, components that are not related to the present invention will be omitted from the drawings.

[0028] The technical solution in the embodiments of the present application is to solve the problems of the above-mentioned background technology, and the overall idea is as follows:

[0029] Example 1:

[0030] This embodiment introduces an intelligent construction system based on digital twins, including a data acquisition module, a data processing module, a template comparison module, and a model update module. The modules work together to form an organic and efficient whole. Through the efficient collection and in-depth processing of various types of construction data, the construction model is dynamically optimized according to preset rules to meet diverse construction needs.

[0031] 1. Data Acquisition Module

[0032] 1. Sensor data acquisition submodule

[0033] 1.1 Sensor Types and Deployment

[0034] 1.1.1 Structure Sensor

[0035] Strain gauges, displacement gauges, and other sensors are installed at key nodes in building structures, specifically at beam-column connections and foundations. Strain gauges measure the strain changes in a structure under load, indirectly determining the stress experienced by the structure based on Hooke's law (σ = Eε) (where σ is stress, E is the elastic modulus, and ε is strain). Displacement gauges precisely measure the displacement of structural nodes with submillimeter accuracy. Specifically, in a frame-structured building, strain gauges and displacement gauges are installed at the beam-column joints on each floor to monitor changes in the structure's mechanical properties during construction and operation in real time, providing critical structural data for model construction. During concrete pouring, strain gauges can monitor stress changes during the concrete hardening process, while displacement gauges can provide real-time information on the displacement of structural nodes under the weight of the concrete and construction loads.

[0036] 1.1.2 Environmental Sensors

[0037] Temperature sensors, humidity sensors, and wind speed sensors are deployed at construction sites and within buildings. Temperature sensors use high-precision thermistor sensors with a measurement range of -40°C to 125°C and an accuracy of ±0.5°C. Humidity sensors utilize a capacitive principle to accurately measure relative humidity, with a measurement range of 0% to 100% RH and an accuracy of ±3% RH. Wind speed sensors use three-cup or propeller-type sensors with a measurement range of 0 to 60 m / s and an accuracy of ±0.5 m / s. Specifically, during the exterior wall insulation construction phase, temperature and humidity sensors monitor ambient temperature and humidity, as these can significantly affect the adhesion and curing time of insulation materials. When overhead work is performed on the construction site, wind speed sensors monitor wind speed in real time. When wind speed exceeds a safety threshold (set at 8 m / s), an alarm is issued to ensure construction safety. These environmental sensors collect real-time environmental parameters to assess the impact of environmental factors on the construction process and building performance.

[0038] 1.2 Data Transmission and Storage

[0039] The data collected by the sensors is sent to the on-site data acquisition terminal via a wireless transmission module (such as Bluetooth, Wi-Fi, or LoRa). After preliminary data organization and caching, the data acquisition terminal transmits the data to a data storage server via a wired network (such as Ethernet) or a wireless network (such as 4G / 5G). The data storage server uses a distributed storage architecture, such as the Ceph distributed storage system, to categorize and store sensor data according to multiple dimensions such as time series t, sensor type s, and project name p. The storage path can be represented as S(t, s, p) to facilitate subsequent data query and analysis. Specifically, to query all temperature sensor data for a project within a specific time period, the storage path can be used to quickly locate and retrieve the relevant data.

[0040] 2. Image data acquisition submodule

[0041] 2.1 Image acquisition equipment and layout

[0042] 2.1.1 Fixed Camera

[0043] High-definition network cameras are installed at various locations on the construction site, such as the entrances and exits, the tops of tower cranes, and major construction areas. These cameras have autofocus and night vision capabilities, with a resolution of up to 4K (3840×2160 pixels) and a frame rate of 25 frames per second. They can clearly capture personnel activities, equipment operation, construction progress, and other conditions at the construction site. Specifically, cameras at the entrances and exits of the construction site can identify information about vehicles and personnel entering and leaving the site, and use image recognition algorithms to count the number of vehicles entering and leaving the site and the distribution of employee types each day. The camera on the top of the tower crane can overlook the entire construction site, monitor the overall progress of the construction area, and calculate the completion rate of the construction area by comparing images from different dates.

[0044] 2.1.2 Drones

[0045] Drones are regularly used to capture aerial images of construction sites. Drones are equipped with high-resolution optical and thermal cameras. The optical camera captures panoramic images of the construction site, with a resolution exceeding 20 megapixels, providing macroscopic information about the site. Thermal cameras detect temperature anomalies in building structures or equipment, and by analyzing the temperature distribution in thermal images, identify potential safety hazards or construction quality issues. The drone's flight altitude h, route r, and recording time t can be flexibly set based on the construction schedule and monitoring needs. Specifically, during the main structure construction phase, the drone conducts aerial photography weekly at an altitude of 100 meters, capturing a comprehensive view of the construction site along a pre-planned route to obtain high-definition panoramic images. During the equipment commissioning phase, thermal cameras can be used to inspect electrical equipment, using temperature distribution to determine if the equipment is experiencing abnormalities such as overheating.

[0046] 2.2 Image data processing and analysis

[0047] The collected image data is first compressed and encoded using the device's built-in image processing chip before being transmitted to the image analysis server. On the server, deep learning algorithms are used to analyze the image data. Object detection algorithms (specifically, Faster R-CNN and YOLO) are used to identify target objects such as people, equipment, and building components in the image, and their number and location information are counted. Image segmentation algorithms (specifically, U-Net and Mask R-CNN) are used to monitor the construction progress of building structures and automatically identify completed and unfinished construction sections. Specifically, the Faster R-CNN algorithm is used to identify the number and location of excavators at the construction site, and the U-Net algorithm is used to segment and identify the construction progress of building walls and calculate the area percentage of completed walls. Furthermore, by analyzing image sequences, the movement trajectories of people and equipment can be calculated to evaluate construction efficiency. The image analysis results are correlated and integrated with sensor data to provide richer visual information for model construction. Assuming the position coordinates of the target object in the image are (x, y), the movement speed of the same target object can be calculated by analyzing the changes in its position coordinates in consecutive images. (where Δt is the image acquisition time interval), thereby evaluating the work efficiency of construction equipment or personnel.

[0048] 2. Data Processing Module

[0049] 1. Data cleaning and preprocessing submodule

[0050] 1.1 Data Cleaning Algorithm

[0051] A variety of data cleaning algorithms are used to process the noise, outliers and missing values ​​that may exist in sensor data and image data. The sliding average filter algorithm is used to remove the noise in the sensor data. Assume that the data sequence collected by the sensor is x1, x2, ..., x n , the result y after sliding average filtering i for Here, k is the size of the sliding window, which is usually selected based on the data fluctuations and sampling frequency. Specifically, for temperature sensor data, if the data fluctuations are small and the sampling frequency is high, a smaller k value (set k = 3) can be selected to quickly remove noise. For outlier detection, a statistical method is used, specifically the 3σ criterion. Assuming that the data follows a normal distribution, if a data point x deviates from the mean μ by more than three times the standard deviation σ, that is, |x - μ| > 3σ, it is considered an outlier and corrected or deleted. For handling missing values, interpolation methods (specifically linear interpolation and Lagrange interpolation) are used to fill them according to the characteristics and distribution of the data.

[0052] A median filter algorithm is used to address noise in image data. Median filtering replaces the grayscale value of each pixel in an image with the median of the grayscale values ​​of its neighboring pixels, effectively removing impulsive interference such as salt and pepper noise. During image segmentation and object detection, morphological operations (such as erosion and dilation) are performed on small areas of detected noise to improve the accuracy of image analysis. Let N be the neighborhood of a pixel in an image. After median filtering, the grayscale value G′ of that pixel is the median of the grayscale values ​​of its neighboring N pixels, i.e., G′ = median(G(N)).

[0053] 1.2 Data Standardization and Normalization

[0054] In order to make different types of data comparable, the cleaned data is standardized and normalized. For sensor data, the Z-score standardization method is used to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. Let the original data be x, and the standardized result x′ be: Where μ is the mean of the data and σ is the standard deviation of the data. For image data, before training the deep learning model, the pixel values ​​of the image are normalized to the interval [0, 1] or [-1, 1] to accelerate the convergence of the model. Let the image pixel value be p, and the result p′ after normalization to the interval [0, 1] is (where p min and p max are the minimum and maximum values ​​of the image pixel values, respectively).

[0055] 2. Data feature extraction submodule

[0056] 2.1 Sensor Data Feature Extraction

[0057] For different types of sensor data, representative characteristic parameters are extracted. For structural sensor data, in addition to directly obtained data such as strain and displacement, derived characteristics such as the stress change rate and displacement acceleration of the structure are also calculated. It can be obtained by performing differential calculation on the stress values ​​at different times: Where Δt is the time interval. For environmental sensor data, extracting temperature trends can be done by fitting a temperature-over-time curve through linear regression analysis. The fitting equation is T = a + bt (where T is temperature, t is time, and a and b are fitting coefficients). These characteristic parameters can more comprehensively reflect the building structure and environmental status, providing key information for subsequent model construction and analysis.

[0058] 2.2 Image Data Feature Extraction

[0059] In terms of image data processing, convolutional neural networks (CNN) in deep learning are used for feature extraction. Based on the classic CNN models of VGG16 and ResNet, low-level features (such as edges, textures, etc.) and high-level features (such as object shape, semantic information, etc.) of the image are extracted in the convolution layer and pooling layer of the model. By analyzing the feature maps at different levels, information such as the position, size, and category of the target object in the image is obtained. Specifically, in the convolution layer, convolution operations are performed on the image through convolution kernels of different sizes and strides to extract local features of the image. Assume that the convolution kernel size is m×n ， The step size is s, and the feature map size after convolution operation is (Where H and W are the height and width of the original image.) Furthermore, image features can be fused with sensor data features to further enhance the model's understanding and analysis of the construction process. By combining equipment location information identified in images with operational status data collected by equipment sensors, a more comprehensive understanding of equipment performance can be achieved.

[0060] 3. Template comparison module

[0061] 1. Template library management submodule

[0062] 1.1 Template classification and storage

[0063] A template library is established, and the digital twin models of construction projects of different types, scales, and construction processes are stored as templates in the template library. The template library is classified and managed according to the building type (such as residential, commercial, industrial, etc.), structural form (such as frame structure, brick-concrete structure, steel structure, etc.), and construction stage (such as foundation construction, main structure construction, decoration and renovation construction, etc.). Each template model contains the building's geometric information, structural information, equipment information, construction progress information, and other aspects, and is stored in a standardized format (specifically IFC format) to facilitate subsequent query and comparison. Suppose the classification label of the template in the template library is C (including information such as building type, structural form, construction stage, etc.), and the template model is M ， The storage path is P(C,M) ， The corresponding template model can be quickly searched and obtained through the classification label.

[0064] 1.2 Template update and maintenance

[0065] As new construction projects are completed and experience is accumulated, the template library is continuously updated and maintained. When defects are found in existing template models or they are inconsistent with the actual construction process, the templates are promptly corrected and optimized. At the same time, digital twin models of representative new construction projects are added to the template library to enrich the content of the template library and improve the template library's coverage of different construction scenarios. Specifically, for the emerging prefabricated building construction process, after completing a project, the digital twin model of the project is added to the corresponding category according to the classification rules of the template library, and the construction process, component connection method and other information in the model are detailed and updated.

[0066] 2. Alignment submodule

[0067] 2.1 Alignment Algorithm

[0068] Using a feature matching-based comparison algorithm, the collected target construction data is processed and feature extracted, and then compared with the construction template in the template library. For sensor data features, the Euclidean distance or cosine similarity between the target data features and the template data features is calculated. Let the target data feature vector be The template data feature vector is The Euclidean distance d is: The cosine similarity sim is:

[0069] For image data features, deep learning metric learning algorithms (such as Triplet Loss and Contrastive Loss) are used to calculate the similarity between target image features and template image features. By comprehensively comparing different types of data features, the degree of match between the target image and each template is determined. Let the overall matching degree be S, which is the weighted sum of the sensor data feature matching degree S1 and the image data feature matching degree S2, that is, S = w1S1 + w2S2 (where w1 and w2 are weights, set according to the importance of the data type).

[0070] 2.2 Setting Thresholds

[0071] Reasonable thresholds are set for different types of construction data and features based on standard specifications of the construction industry, historical project experience, and actual engineering needs. The threshold for the structural stress change rate is set based on the stress change range allowed in the building structure design specifications. In common reinforced concrete frame structures, the threshold for the structural stress change rate under normal use is set to no more than 0.05 MPa per hour. The threshold for the construction progress is set based on the total project duration T and the planned duration t of each stage. i To determine. Suppose the total construction period of a project is T, and the planned construction period is t idays, when the actual construction days t exceeds 10% of the planned construction period, that is, t>1.1t i , or when the proportion of the actual completed project volume to the planned project volume is less than 80%, the progress of the construction phase is judged to be abnormal, and the system issues an early warning to remind construction personnel to adjust the construction plan and resource allocation in time.

[0072] Regarding environmental data, temperature and humidity significantly impact building construction and material properties. During concrete pouring, ambient temperatures below 5°C or above 35°C can affect the concrete's hydration reaction and ultimate strength. Therefore, the ambient temperature threshold for concrete pouring is set between 5°C and 35°C. Ambient humidity significantly impacts the processing and use of wood. When relative humidity exceeds 70% RH, wood is susceptible to moisture, deformation, and mold. Therefore, in wood processing areas, the humidity threshold is set at 70% RH.

[0073] 4. Model Update Module

[0074] 1. New template generates submodule

[0075] The collected target construction data is deeply integrated, and data fusion technologies such as Kalman filtering are used to integrate sensor data, image data, and other relevant project information. Based on the fused data, a new digital twin model is constructed using reverse engineering methods. Taking building structures as an example, with the help of finite element analysis software, the design parameters and geometric shape of the building structure are reversed based on the measured data and image information of the structure to construct an accurate structural model. The newly constructed model needs to be strictly verified and optimized. Using historical data and actual monitoring data, the model prediction results are compared with the actual observation results. If there are deviations, optimization algorithms such as genetic algorithms are used to adjust the model parameters and structure until the model performance meets the requirements. The new model is then stored in the template library in a standardized IFC format.

[0076] 2. Real-time update of submodules

[0077] During the construction and operation and maintenance process, as new data is continuously collected, the model needs to be updated in real time. Using an incremental update algorithm, when new data arrives, only the affected parts of the model are updated. For example, for a building structure model, when new structural displacement data or stress data arrives, the status of the corresponding nodes and units in the structural model are updated according to the principles of structural mechanics. Suppose the displacement vector of a node in the structural model is When the new displacement data When arriving, the updated displacement vector is At the same time, during the model update process, the consistency and integrity of the model must be ensured. Since the various parts of the building system are interrelated, when the model of a certain part is updated, the models of other related parts must be updated synchronously.

[0078] Example 2:

[0079] This embodiment introduces an intelligent construction method based on digital twins, which includes the following steps:

[0080] Step 1: Preliminary Preparation: Based on the project design plan, determine the type, quantity, and installation layout of the required sensors and image acquisition equipment. For different building types, specifically residential, commercial, and industrial plants, install structural sensors such as strain gauges and displacement gauges at key structural locations (beam-column joints, foundations, etc.). Environmental sensors for temperature, humidity, and wind speed are deployed throughout the construction site and within the building. Furthermore, install fixed cameras at key locations on the site, and plan the routes and timings for drone aerial photography. Build the data collection, transmission, and storage infrastructure. Select appropriate wireless transmission modules (such as Bluetooth, Wi-Fi, or LoRa) for data transmission between sensors and data acquisition terminals. Connect the data acquisition terminals to the data storage server using a wired network (Ethernet) or wireless network (4G / 5G). Use a distributed storage system (such as Ceph) to categorize and store data according to dimensions such as time series t, sensor type s, and project name p. The storage path is designated S(t,s,p). A digital twin template model similar to the project type, scale, and construction process is selected from the template library. The template model is stored in the standardized IFC format and contains various information such as building geometry, structure, equipment, and construction progress.

[0081] Step 2: Data Collection and Preliminary Processing. After construction begins, the data collection module is activated. Various sensors collect data at a set frequency. Structural sensors monitor the strain and displacement of the building structure under construction loads in real time, calculating stress data using Hooke's law σ = Eε (where σ is stress, E is the elastic modulus, and ε is strain). Environmental sensors collect environmental parameters such as temperature, humidity, and wind speed at the construction site, providing data for construction environment assessment. Fixed cameras and drones collect image data according to a schedule. Object detection algorithms (such as Faster R-CNN and YOLO) are used to identify target objects such as people, equipment, and building components in the images. Image segmentation algorithms (such as U-Net and Mask R-CNN) are used to analyze the construction progress of the building structure. The collected raw data is sent to the data collection terminal via a wireless transmission module. The data collection terminal performs preliminary data sorting and buffering before transmitting it to a data storage server via a wired or wireless network.

[0082] Step 3: Data deep processing and feature extraction. After the data is transferred to the data processing module, data cleaning and preprocessing are performed. For sensor data, a sliding average filter algorithm is used to remove noise. Suppose the data sequence collected by the sensor is x1, x2, ..., x n , the result after sliding average filtering Where k is the sliding window size, which is determined according to data fluctuations and sampling frequency. Outliers are detected and processed according to the 3σ criterion. If the deviation of a data point x from the mean μ exceeds 3 times the standard deviation σ, that is, |x-μ|>3σ, it is corrected or deleted. Interpolation methods (such as linear interpolation and Lagrange interpolation) are used to fill missing values. For image data, the median filter algorithm is used to remove noise. During image segmentation and target detection, morphological operations (such as corrosion and expansion) are performed on small noise areas to optimize image quality. Then, the cleaned data is standardized and normalized. For sensor data, the Z-score standardization method is used to convert the original data x into The mean is 0 and the standard deviation is 1. For image data, the pixel values ​​are normalized to the range [0, 1] or [-1, 1] before deep learning model training. Then, the data features are extracted. For structural sensor data, the stress change rate is calculated. Derived features such as displacement and acceleration; for environmental sensor data, features such as temperature change trends, humidity extremes, and wind speed peaks are extracted; for image data, convolutional neural networks (such as VGG16 and ResNet) are used to extract low-level features (such as edges and textures) and high-level features (such as object shape and semantic information) in the convolutional layer and pooling layer.

[0083] Step 4: Template comparison and threshold judgment, compare the processed feature data with the selected template model. For sensor data features, calculate the target data feature vector With the template data feature vector Euclidean distance between or cosine similarity For image data features, the metric learning algorithm in deep learning (such as Triplet Loss, Contrastive Loss) is used to calculate the similarity. The comprehensive matching degree of sensor and image data features is obtained to obtain the comprehensive matching degree S=w1S1+w2S2, where w1 and w2 are weights, which are set according to the importance of the data type. The threshold is set according to the construction industry standards and specifications, historical project experience and actual project needs. In the construction of common building structures, the hourly threshold of the structural stress change rate under normal use is set to not exceed a certain proportion of the design stress (specifically 5%); in terms of construction progress, if the actual construction days t exceed 10% of the planned construction period, that is, t>1.1t 计划 , or the ratio of the actual completed work volume to the planned work volume is less than 80%, the construction progress is considered abnormal. Regarding environmental data, abnormal environmental parameters are considered abnormal if the ambient temperature during concrete pouring is below 5°C or above 35°C, or if the humidity in the wood processing area exceeds 70% RH. When the comprehensive matching degree exceeds the preset threshold, the model update process is triggered.

[0084] Step 5: Model update and construction decision optimization. If the comparison result triggers the model update condition, the model update module is started. The new template generation submodule uses data fusion technologies such as Kalman filtering to integrate newly collected sensor data, image data and other relevant project information. A new digital twin model is constructed using the reverse engineering method. Taking the building structure as an example, with the help of finite element analysis software, the design parameters and geometric shape of the building structure are reversed according to the measured data and image information of the structure to construct an accurate structural model. The newly constructed model needs to be strictly verified and optimized. The model prediction results are compared with the actual observation results using historical data and actual monitoring data. If there is a deviation, the optimization algorithm such as genetic algorithm is used to adjust the parameters and structure of the model until the model performance meets the requirements. The new model is then stored in the template library in a standardized IFC format. The model real-time update submodule incrementally updates the existing model based on the new data. For example, for a building structure model, when new structural displacement data or stress data arrives, the status of the corresponding nodes and units in the structural model is updated according to the principles of structural mechanics. Suppose the displacement vector of a node in the structural model is When the new displacement data When arriving, the updated displacement vector is At the same time, to ensure the consistency and integrity of the model, when updating a specific model component, other related models are updated simultaneously. For example, when updating the equipment operating status model, the energy consumption model and the building environment model are updated accordingly. The updated model provides support for construction decision-making. If the model indicates that construction progress in a certain area is lagging, analysis can be used to identify factors affecting progress, such as improper equipment allocation or insufficient manpower. Construction plans can then be adjusted and resource allocation optimized to ensure smooth project progress. Throughout the construction process, steps two through five are repeated continuously to achieve dynamic monitoring and optimization of the construction process.

[0085] Finally, it should be noted that the above embodiments are merely examples for the purpose of illustrating the present invention and are not intended to limit the embodiments. Those skilled in the art will readily appreciate that other variations or modifications based on the above description are possible. It is not necessary and impossible to provide an exhaustive list of all embodiments. However, obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. An intelligent construction system based on digital twins, characterized by: Includes the following modules: Data acquisition module: used to collect various data during the construction process, including sensor data acquisition submodule and image data acquisition submodule. The sensor data acquisition submodule is used to monitor changes in the mechanical properties of the structure in real time; the image data acquisition submodule uses the autofocus, night vision and high-resolution features of a fixed camera to capture personnel and equipment activities and construction progress; Data processing module: cleans, preprocesses and extracts features from the collected data, including a data cleaning and preprocessing submodule and a data feature extraction submodule. The data cleaning and preprocessing submodule uses a sliding average filter algorithm, a statistical method and an interpolation method to process sensor data, a median filter algorithm and morphological operations to process image data, and standardizes and normalizes the cleaned data. The data feature extraction submodule uses a convolutional neural network to extract low-level and high-level features from image data based on derived features. Template comparison module: This module includes a template library management submodule and a comparison submodule. The template library management submodule establishes a template library and stores the digital twin models of different construction projects in IFC format according to building type, structural form, and construction stage. The comparison submodule uses a feature matching-based comparison algorithm and a deep learning metric learning algorithm to calculate the similarity of image data features. The two are combined to determine the degree of match between the target building and the template, and thresholds for various data features are set at the same time. Model update module: It consists of a new template generation submodule and a real-time update submodule. The new template generation submodule uses data fusion technology to integrate data, adopts reverse engineering methods to build a new digital twin model and stores it in the template library in IFC format. During the construction and operation and maintenance process, the real-time update submodule uses an incremental update algorithm to update the affected parts of the model according to the new data to ensure the consistency and integrity of the model.

2. The intelligent construction system based on digital twin according to claim 1, characterized in that: In the data cleaning and preprocessing submodule, a sliding average filtering algorithm is used to remove noise. Assume that the data sequence collected by the sensor is x1, x2, ..., x n , the result y after sliding average filtering i for Where k is the size of the sliding window.

3. The intelligent construction system based on digital twin according to claim 1, characterized in that: In the comparison submodule, the target data feature vector is set to The template data feature vector is The Euclidean distance d is: The cosine similarity sim is:

4. The intelligent construction system based on digital twin according to claim 1, characterized in that: The threshold is based on the total duration T of the project and the planned duration t of each stage. i To determine, suppose the total construction period of a project is T, and the planned construction period is t i days, when the actual construction days t exceeds 10% of the planned construction period, that is, t>1.1t i , it is determined that the progress of this construction phase is abnormal, and the system issues an early warning to remind the construction personnel to adjust the construction plan and resource allocation in time.

5. An intelligent construction method based on digital twins, characterized in that: The following steps are involved: Step 1: Preliminary preparation: Based on the project design plan, determine the type, quantity, and installation layout of sensors and image acquisition equipment, build data acquisition, transmission, and storage infrastructure, select transmission modules, use a distributed storage system to classify and store data, and select similar digital twin template models from the template library; Step 2: Data collection and preliminary processing. After construction begins, the data collection module is started. Various sensors collect data, and fixed cameras and drones collect image data. The collected raw data is initially sorted and cached before being transmitted to the data storage server. Step 3: Data in-depth processing and feature extraction. After the data is transferred to the data processing module, data cleaning, preprocessing, standardization, and normalization are performed, and then various data features are extracted; Step 4: Template comparison and threshold judgment: Compare the processed feature data with the selected template model, calculate the comprehensive matching degree, and set the threshold according to the construction industry standards, historical project experience, and actual project needs. When the comprehensive matching degree exceeds the preset threshold, the model update process is triggered; Step 5: Model update and construction decision optimization. If the model update conditions are triggered, the model update module is started to build a new model and update it in real time to ensure model consistency and integrity. The updated model is used to support construction decisions. The above steps are repeated throughout the construction process to achieve dynamic monitoring and optimization.

Images