Engineering life cycle-oriented digital twin model construction and updating method and system

Abstract

The application relates to the field of twin model construction, in particular to a digital twin model construction and updating method and system for the whole life cycle of engineering. The method comprises the following steps: periodically inspecting a construction site according to a preset route by a UAV, eliminating motion blur, and extracting a construction stage image sequence; performing geometric state difference modeling based on the construction stage image sequence to construct a digital twin model; tracking the behavior of construction personnel on the construction stage image sequence, dynamically updating the digital twin model, and constructing a dynamically updated twin model; importing a pre-constructed completed BIM model; performing spatial registration and progress calculation on the dynamically updated twin model according to the completed BIM model to obtain a progress completion rate; adjusting engineering resource allocation according to the progress completion rate, and outputting an engineering adjustment scheme. The application realizes real-time visualization of a construction site and accurate progress monitoring, and improves the efficiency of building engineering management.

Description

Technical Field

[0001] This invention relates to the field of digital twin model construction, and in particular to a method and system for constructing and updating digital twin models for the entire lifecycle of engineering projects. Background Technology

[0002] Throughout the entire lifecycle of a project, it involves multiple phases including design, construction, and operation and maintenance. Each phase generates a large amount of data, including construction progress, component geometry information, construction worker behavior, equipment operating status, and material consumption. This information is characterized by its strong time sequence, large volume, and diverse sources. If it cannot be effectively integrated, updated in real time, and dynamically analyzed, it will lead to inaccurate monitoring of the construction process, difficulty in predicting progress deviations, lack of basis for operation and maintenance decisions, and even affect project quality and safety. Existing digital modeling and management methods mainly rely on phased BIM modeling or construction records. While these methods can statically manage engineering components and spatial layouts, they lack the ability to perceive and update the dynamic construction process in real time throughout its entire lifecycle. The limitations of this approach are: the inability to continuously monitor construction progress and behavior, making it difficult to detect construction deviations in a timely manner; and the lack of intelligent analysis and prediction methods for the large amount of time-series and behavioral data generated during the construction phase, failing to support dynamic scheduling and optimized management. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention proposes a method and system for constructing and updating digital twin models throughout the entire lifecycle of engineering projects, thereby resolving at least one of the aforementioned technical issues.

[0004] To achieve the above objectives, this invention provides a method for constructing and updating a digital twin model for the entire lifecycle of an engineering project, comprising the following steps: Step S1: Based on the UAV, periodically inspect the construction site according to the preset route, perform motion blur removal, and extract the image sequence of the construction stage; Step S2: Perform geometric state differential modeling based on the image sequence of the construction phase to construct a digital twin model; Step S3: Track the behavior of construction personnel in the video sequence during the construction phase, and dynamically update the digital twin model to build a dynamically updated twin model; Step S4: Import the pre-built as-built BIM model; perform spatial registration and progress estimation on the dynamically updated twin model based on the as-built BIM model to obtain the progress completion rate; Step S5: Adjust the allocation of engineering resources according to the progress completion rate and output the engineering adjustment plan; perform intelligent engineering cycle collaborative control processing based on the engineering adjustment plan.

[0005] The specific steps of step S1 are as follows: The drones conduct periodic inspections of the construction site according to a preset route and extract the first construction images. Collect flight attitude information and timestamps during the drone inspection process; The flight attitude information and timestamp are used to calibrate the construction image to obtain a second construction image. Perform time-series alignment and motion blur filtering on the second construction image to output the third construction image; Keyframe identification was performed on the third construction image, and multiple keyframes were extracted. Construction phase identification and labeling are performed based on multiple keyframes, and a construction phase image sequence is output.

[0006] The specific steps of step S2 are as follows: Semantic segmentation is performed on image sequences during the construction phase to identify multiple construction components; Calculate the outline, floor height, span, and cross-sectional dimensions of the construction component to obtain its geometric parameters; Based on the image sequence of the construction phase, the geometric parameters are tracked for temporal changes to obtain the geometric parameter sequence at different time points; Geometric state difference analysis is performed based on the geometric parameter sequence to obtain the time-varying characteristics of the component. Dynamic digital 3D modeling is performed based on the time-varying characteristics of the components to construct a digital twin model.

[0007] The specific steps of step S3 are as follows: On-site material flow detection is performed based on the image sequence of the construction phase, and the digital twin model is annotated in three dimensions to construct a material annotation model; Target identification is performed on the construction phase image sequence to mark personnel at the construction site; Continuous behavioral tracking of personnel at the construction site is conducted to extract their behavioral trajectories; the key behavioral trajectories include personnel movement trajectories, equipment operating paths, and process operation actions. The construction behavior trajectory is analyzed to obtain construction behavior characteristics; the construction behavior characteristics include operation path, dwell time, operation frequency and process execution sequence. Identify the spatial coordinates and timestamps of construction activities to obtain construction tags; The characteristics of construction behavior are digitally mapped based on construction tags, and the material labeling model is dynamically updated to build a dynamically updated twin model.

[0008] The specific steps for constructing a material annotation model are as follows: Based on the image sequence from the construction phase, on-site material flow monitoring is performed, and a 3D annotation is applied to the digital twin model. On-site material flow detection is performed on the image sequence during the construction phase to identify material flow information; Material attribute identification is performed based on image sequences during the construction phase to extract multi-dimensional material attribute information; Material consumption rate is calculated based on multi-dimensional material attribute information to obtain the consumption rate of different materials; The consumption rate and material flow information are used to create a three-dimensional annotation of the digital twin model, thus constructing a material annotation model.

[0009] The specific steps of step S4 are as follows: Import a pre-built as-built BIM model; Based on the as-built BIM model, the dynamic updated twin model is spatially registered, and the morphological deviation of each component is calculated to obtain the morphological deviation parameters of different components. Based on the aforementioned morphological deviation parameters, the construction progress is estimated to obtain the progress completion rate.

[0010] The specific steps of step S5 are as follows: Based on the progress completion rate, the completion time is predicted to obtain the predicted completion time point; Extract the construction planning log; calculate the completion time of key milestones and the overall completion time based on the construction planning log; Based on the completion time of the key nodes and the overall completion time, the project progress deviation is calculated for the predicted completion time point to obtain the expected progress deviation value. Based on the expected schedule deviation, adjust the allocation of engineering resources and output the engineering adjustment plan; Intelligent engineering cycle collaborative control processing is carried out based on engineering adjustment schemes.

[0011] This specification provides a system for constructing and updating digital twin models for the entire engineering lifecycle, used to execute the method for constructing and updating digital twin models for the entire engineering lifecycle as described above, including: The image acquisition module is used to conduct periodic inspections of the construction site based on the UAV according to the preset route, and to perform motion blur removal and extract the image sequence of the construction stage. The modeling module is used to perform geometric state differential modeling based on the image sequence of the construction phase and build a digital twin model; The update module is used to track the behavior of construction personnel in the image sequence during the construction phase and dynamically update the digital twin model to build a dynamically updated twin model. The progress estimation module is used to import the pre-built as-built BIM model; based on the as-built BIM model, it performs spatial registration and progress estimation on the dynamically updated twin model to obtain the progress completion rate. The control module is used to adjust the allocation of engineering resources according to the progress completion rate and output the engineering adjustment plan; and to perform intelligent engineering cycle collaborative control processing based on the engineering adjustment plan.

[0012] The specific benefits of this invention are as follows: Periodic drone inspections enable full coverage and high-precision image acquisition of the construction site, forming a continuous and standardized image sequence for each construction stage. This eliminates blind spots in manual inspections, ensures the integrity of construction data, and improves data traceability and construction safety. Through component identification and 3D modeling, 2D image information can be transformed into a structured digital model, achieving precise visualization at the component level. The model supports construction progress analysis, deviation monitoring, and virtual simulation, providing reliable data support for construction optimization and lifecycle management. Tracking and dynamically updating the behavior of construction personnel allows the digital twin model to reflect the real-time status of the construction site. This method can monitor personnel and equipment activities and operational behaviors, improving safety management levels, while providing real-time data support for construction progress analysis, resource scheduling, and anomaly early warning. By comparing the BIM model with the dynamically updated twin model, component deviations and construction progress completion rates can be quantified. This method can accurately assess the progress of each process and area, identify construction deviations and quality problems, achieve construction status visualization, and support lifecycle management. Adjusting and coordinating resources based on progress completion rates can optimize the allocation of construction personnel, equipment, and materials, shortening the delay time of key nodes. Dynamic digital twin models can intuitively display the effects of adjustments, enabling intelligent management of the construction cycle and improving efficiency, safety, and scientific decision-making. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating the steps of a method for constructing and updating a digital twin model for the entire lifecycle of an engineering project, as described in this invention. Figure 2 This is a detailed flowchart illustrating the implementation steps of step S1. Figure 3 This is a flowchart illustrating the detailed implementation steps of step S2. Detailed Implementation

[0014] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0015] This application provides a method and system for constructing and updating digital twin models throughout the entire lifecycle of an engineering project. The implementing entities of this method and system include, but are not limited to, mechanical equipment, data processing platforms, cloud server nodes, and network upload devices that can be considered general computing nodes in this application. The data processing platform includes, but is not limited to, at least one of an audio / image management system, an information management system, and a cloud-based data management system.

[0016] Please see Figures 1 to 3 This invention provides a method for constructing and updating digital twin models for the entire lifecycle of engineering projects, including the following steps: Step S1: Based on the UAV, periodically inspect the construction site according to the preset route, perform motion blur removal, and extract the image sequence of the construction stage; Step S2: Perform geometric state differential modeling based on the image sequence of the construction phase to construct a digital twin model; Step S3: Track the behavior of construction personnel in the video sequence during the construction phase, and dynamically update the digital twin model to build a dynamically updated twin model; Step S4: Import the pre-built as-built BIM model; perform spatial registration and progress estimation on the dynamically updated twin model based on the as-built BIM model to obtain the progress completion rate; Step S5: Adjust the allocation of engineering resources according to the progress completion rate and output the engineering adjustment plan; perform intelligent engineering cycle collaborative control processing based on the engineering adjustment plan.

[0017] In the embodiments of the present invention, see Figure 1 This is a flowchart illustrating the steps of a method for constructing and updating a digital twin model for the entire lifecycle of an engineering project, as described in this invention. In this example, the steps of the method include: Step S1: Based on the UAV, periodically inspect the construction site according to the preset route, perform motion blur removal, and extract the image sequence of the construction stage; In this embodiment, the UAV inspection route is planned according to the scale and characteristics of the construction area, employing a strategy combining grid coverage and densification in key areas. The route altitude is typically set between 40 and 80 meters to ensure a ground resolution of 2–5 centimeters per pixel, while the forward overlap is controlled at 75%–85% and the lateral overlap at 65%–75%, ensuring the accuracy of subsequent 3D reconstruction and image stitching. The UAV is equipped with a high-resolution RGB camera with a resolution ≥3840×2160 and a frame rate set to 30 fps, acquiring continuous images through time-triggered or distance-triggered methods. During the inspection, the onboard flight control system records GNSS / RTK positioning information and IMU attitude data in real time, with sampling frequencies of 5–10 Hz and 100–200 Hz, respectively, and generates a timestamp synchronously with each frame of image, achieving a one-to-one correspondence between images and pose information. After image acquisition, the original image sequence is initially screened, including removing blurry frames and frames with abnormal lighting. Frames with sharpness below the threshold are removed by Laplacian operator sharpness evaluation or deep learning blur detection method, with SSIM < 0.85.

[0018] Step S2: Perform geometric state differential modeling based on the image sequence of the construction phase to construct a digital twin model; In this embodiment, after obtaining the image sequence of the construction phase, semantic segmentation and recognition of components are performed. Deep learning semantic segmentation models, such as U-Net or DeepLabv3+, are used to perform pixel-level classification of the images, identifying components such as beams, columns, floor slabs, walls, and scaffolding. The input image size is uniformly 512×512 or 1024×1024, and the training data includes different construction phases, lighting conditions, and occlusion conditions, with at least 8 categories. After segmentation, edge detection or contour extraction algorithms are used to generate component contours, and combined with image spatial calibration information, pixel coordinates are converted into three-dimensional spatial coordinates to restore the geometric parameters of the components, including floor height, span, cross-sectional dimensions, and three-dimensional position. Multi-view images, combined with SfM, Structure from Motion, and MVS (Multi-View Stereo) algorithms, are used to perform three-dimensional modeling of the components, ultimately forming a digital twin model, where each component has three-dimensional geometric attributes, spatial coordinates, and a unique identifier. Based on this, the component model is quality checked, and the model accuracy is verified by reprojection error or mesh matching, generally controlled within the range of 0.5–1 pixel or ±2–5 cm.

[0019] Step S3: Track the behavior of construction personnel in the video sequence during the construction phase, and dynamically update the digital twin model to build a dynamically updated twin model; In this embodiment, construction personnel detection is performed using target detection models such as YOLOv5 / YOLOv8 or Faster R-CNN, identifying categories including personnel and safety equipment. The detection outputs bounding boxes and confidence scores, along with timestamps. Cross-frame tracking is achieved through a multi-target tracking algorithm, assigning a unique ID to each person, and the trajectory points include 3D spatial coordinates and time information. Trajectory smoothing employs Kalman filtering to reduce jitter caused by occlusion or vibration. Combining pose estimation and action recognition models such as OpenPose, HRNet, and LSTM, construction operation actions, such as carrying, installation, and pouring, are extracted. The dynamically updated twin model maps personnel positions, trajectories, and operational states to the 3D scene, dynamically updating and displaying personnel movement paths, equipment operating trajectories, and operational actions to achieve real-time simulation and visualization of the construction site. The update frequency can be consistent with the inspection cycle or achieved at the minute level during critical stages, ensuring the model is synchronized with the on-site construction status.

[0020] Step S4: Import the pre-built as-built BIM model; perform spatial registration and progress estimation on the dynamically updated twin model based on the as-built BIM model to obtain the progress completion rate; In this embodiment, the as-built BIM model is imported into a digital twin platform, and the geometric information, nodes, and topological relationships of components are analyzed, and the coordinate system and units are unified. Spatial registration is performed on the dynamically updated twin model using rigid body transformation and the ICP algorithm to achieve the correspondence between real-time components and BIM components. After registration, the dimensional and positional deviations of each component are calculated. The dimensional deviations ΔL, ΔW, and ΔH, and the positional deviation ΔP are obtained by comparing the real-time model with the BIM design data, with accuracy controlled within ±2–5 cm. The completion criterion for a component is set as follows: if both dimensional and positional deviations are within the threshold range, the component is considered complete. The ratio of the number of completed components to the total planned number of components is statistically analyzed, and a weighted progress completion rate is calculated based on the component weight coefficients. For example, the weight of the main structural components is 0.6, auxiliary components 0.3, and electromechanical installation 0.1, forming an overall construction progress completion rate curve. The progress completion rate can be updated daily or at key nodes to quantify the current project progress and provide a basis for resource adjustments.

[0021] Step S5: Adjust the allocation of engineering resources according to the progress completion rate and output the engineering adjustment plan; perform intelligent engineering cycle collaborative control processing based on the engineering adjustment plan.

[0022] In this embodiment, after obtaining the progress completion rate, lagging nodes and resource-deficient links are identified. Optimization algorithms such as linear programming or genetic algorithms are used to adjust the scheduling of construction personnel, machinery, and materials, with objectives including shortening critical node delays, improving resource utilization, and reducing construction costs. The adjustment plan includes the number and scheduling of construction teams, equipment operation arrangements, critical material supply plans, and necessary overtime or night shift arrangements. The adjustment plan is mapped to a digital twin model, and intelligent collaborative control of the engineering cycle is achieved by dynamically updating the construction status, personnel locations, and equipment operating trajectories. Simulations are used to simulate the adjustment effects, verify the feasibility of critical nodes, and optimize resource allocation schemes. The dynamic digital twin model reflects the adjusted construction status in real time and can be used for progress monitoring, construction scheduling, and anomaly early warning, realizing full-process management from resource optimization to construction cycle control, ensuring construction progresses according to plan, and providing decision support.

[0023] In this embodiment, see Figure 2 The diagram below illustrates the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: The drones conduct periodic inspections of the construction site according to a preset route and extract the first construction images. Collect flight attitude information and timestamps during the drone inspection process; The flight attitude information and timestamp are used to calibrate the construction image to obtain a second construction image. Perform time-series alignment and motion blur filtering on the second construction image to output the third construction image; Keyframe identification was performed on the third construction image, and multiple keyframes were extracted. Construction phase identification and labeling are performed based on multiple keyframes, and a construction phase image sequence is output.

[0024] In this embodiment, a unified spatial reference frame is established in the construction area, typically using WGS84 or a local ENU coordinate system, and flight path planning is performed in conjunction with existing BIM models or terrain data. The flight path design employs a grid-based coverage method, while increasing waypoint density for key areas such as tower cranes, foundation pits, and material storage yards. Flight altitude is generally set between 40–80 m to ensure a ground resolution of 2–5 cm / pixel; forward overlap is controlled at 75%–85%, and lateral overlap at 65%–75% to meet the requirements of subsequent image stitching and 3D reconstruction. The UAV is equipped with an RGB camera with a resolution of at least 3840×2160, set to a frame rate of 30 fps, acquiring continuous images by capturing one frame approximately every 5 m via distance triggering or one frame every 2 seconds via time triggering. Inspection frequency is adjusted according to the construction progress, for example, once daily during the regular phase, increasing to 2–3 times daily during critical phases. Shooting parameters are typically set to ISO 100–400 and shutter speed ≥1 / 800 s to reduce blurring caused by camera shake. Simultaneously with image acquisition, spatial position and attitude information during flight is recorded, including three-dimensional coordinates (x, y, z), attitude angles (roll, pitch, yaw), and velocity vectors. Position data is sourced from GNSS or RTK modules, with a sampling frequency typically of 5–10 Hz; attitude data is sourced from IMUs, with a sampling frequency of 100–200 Hz. Both are processed uniformly using fusion algorithms such as extended Kalman filtering to improve accuracy and stability. Each image frame is accompanied by a high-precision timestamp at the millisecond level to ensure a strict correspondence between the image and pose data. Due to differences in sampling frequencies between different sensors, pose data needs to be compensated using time interpolation methods to align it with the image frames on the time axis. When the image frame rate is 30 fps, the attitude parameters for each frame are calculated using linear interpolation.

[0025] Camera intrinsic parameter calibration is completed, obtaining focal length, principal point position, and distortion parameters including radial distortion k1, k2, k3 and tangential distortion p1, p2, often achieved using a checkerboard calibration method. Camera extrinsic parameters are determined by combining flight attitude information and optimized using ground control points. Typically, 5–10 ground control points are deployed, evenly distributed along the edges and center of the construction area; their coordinates are obtained through RTK measurements with centimeter-level accuracy. Structured bundle adjustment is used to jointly optimize all images, improving overall accuracy by minimizing reprojection error to within 0.5–1 pixel. Each image frame is mapped to a unified spatial coordinate system, achieving geometric distortion correction and spatial registration. Timestamps are used to filter out abnormal data frames and ensure consistency across multiple data sources in the temporal dimension. The resulting second construction image possesses accurate spatial location information and unified coordinate attributes, and can be directly used for spatial analysis and model construction.

[0026] The image sequence is sorted according to timestamps, and different batches of images are aligned using time interpolation or dynamic time warping methods to ensure comparability of the same spatial area at different time points. During temporal alignment, time deviations are controlled within ±50 ms to ensure data consistency. Image quality is evaluated and screened. A sharpness evaluation function based on the Laplacian operator is used to calculate the sharpness index of each frame. Frames with an index below a set threshold (e.g., less than 100) are considered blurry. Slightly blurry images can be restored using blind deconvolution algorithms; severely blurry images are directly discarded. Further analysis using frequency domain analysis or deep learning methods can improve the accuracy of the screening. The processed images are temporally continuous, spatially stable, and have consistent sharpness, forming the third construction image sequence.

[0027] Calculate the structural similarity index (SSIM) between adjacent frames or the matching results based on feature points such as SIFT or ORB. When the similarity is below 0.85 or the feature matching changes significantly, the frame is marked as a candidate keyframe. Combine pose variation constraints, such as spatial displacement exceeding 2 m between adjacent frames or heading angle changes exceeding 10°, to further enhance the spatial distribution uniformity of keyframes. To improve semantic expressiveness, deep feature extraction methods can be used. The image is input into a convolutional neural network to generate high-dimensional feature vectors, and clustering methods such as K-means are used to select cluster centers as the final keyframes. The number of keyframes is generally controlled at 5%–15% of the total number of frames, which preserves the main information while reducing the computational load of subsequent processing.

[0028] Using keyframes as input, a construction stage recognition model is constructed, mapping image content to specific construction stage categories, such as earthwork excavation, foundation construction, main structure construction, electromechanical installation, and decoration. The model typically employs a convolutional neural network structure such as ResNet50 or EfficientNet. The input image size is uniformly 224×224 or 256×256, with a training dataset of no less than 5000 images, and the training and validation data are split in an 8:2 ratio. Optimization is achieved using the cross-entropy loss function, resulting in a stable classification accuracy of over 90%. For each keyframe, a stage label and corresponding confidence score are output, and sequence optimization is performed based on temporal order. For example, a Hidden Markov Model or temporal convolutional network is used to smooth the classification results, avoiding abrupt changes in stage recognition. Keyframes of the same stage are integrated chronologically to form a continuous construction stage image sequence, with added time labels and spatial location information, providing a clear representation of the construction progress.

[0029] In this embodiment, see Figure 3 The diagram below illustrates the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Semantic segmentation is performed on image sequences during the construction phase to identify multiple construction components; Calculate the outline, floor height, span, and cross-sectional dimensions of the construction component to obtain its geometric parameters; Based on the image sequence of the construction phase, the geometric parameters are tracked for temporal changes to obtain the geometric parameter sequence at different time points; Geometric state difference analysis is performed based on the geometric parameter sequence to obtain the time-varying characteristics of the component. Dynamic digital 3D modeling is performed based on the time-varying characteristics of the components to construct a digital twin model.

[0030] In this embodiment, based on the image sequence of the construction phase, the images undergo unified preprocessing, including size normalization (e.g., adjusting to 512×512 or 1024×1024), brightness equalization, and noise suppression to ensure the consistency of the input data. A deep learning semantic segmentation model is used for pixel-level classification of the images; common methods include U-Net, DeepLabv3+, or the SegFormer model based on the Transformer structure. Training data needs to cover different construction phases and lighting conditions. Labeled categories typically include at least eight types of components such as beams, columns, floor slabs, walls, scaffolding, and mechanical equipment. A sample size of 8000–12000 images is recommended, with a balanced category distribution. During training, the initial learning rate is set to 1e-4, the batch size to 8–16, and the number of training epochs to 80–120. A combination of cross-entropy loss and Dice loss is used to improve boundary recognition capabilities. The segmentation result output is a multi-class mask image, with each pixel corresponding to a component label. To further improve segmentation accuracy, a Conditional Random Field (CRF) can be introduced for post-processing to optimize boundary details and make the component outlines clearer.

[0031] The 2D boundaries of the components are obtained using edge detection algorithms such as the Canny operator or contour extraction methods such as the contour tracking algorithm in OpenCV. The contours are then simplified and smoothed using polygon approximation methods. Combining spatial calibration information from the image, derived from prior pose and camera parameters, pixel coordinates are converted to actual spatial coordinates, achieving scale restoration from 2D to 3D. Floor height calculation is typically performed by identifying the vertical distance between the floor slab and the ground. Height information can be obtained using multi-view reconstruction or depth estimation methods such as SfM or MVS algorithms based on multi-view geometry, with accuracy controllable within ±3 cm. Span parameters are obtained by measuring the maximum horizontal extension distance of the beam or slab, while cross-sectional dimensions are obtained by analyzing the width and height of the component's cross-sectional area; for example, column cross-sections can be fitted using the minimum bounding rectangle. To ensure stability, the measurement results of the same component in consecutive frames are processed using mean filtering or median filtering.

[0032] Each component is assigned a unique identifier, and a correspondence can be established between adjacent frames using IoU-based intersection-over-union matching or feature matching. When the IoU of the same component in adjacent frames is greater than 0.7, they are considered the same target. For cases of occlusion or viewpoint changes, appearance features such as color histograms or depth feature vectors can be used for auxiliary matching. After component matching is completed, their geometric parameters at different time points are recorded chronologically to form time-series data. For example, for a beam component, the sequence of its span changes over time can be recorded, with the time interval set to an inspection cycle such as once a day. To reduce errors caused by measurement fluctuations, Kalman filtering or sliding window averaging methods can be used to smooth the time series, making parameter changes more continuous.

[0033] To calculate parameter differences between adjacent time points, such as changes in floor height (Δh), span (Δl), and cross-sectional dimensions (Δs), and further calculate the rate of change (Δh / Δt), a threshold filtering mechanism can be set to avoid outliers from affecting the results. For example, when the change exceeds a reasonable range, such as a floor height change exceeding 0.5 m, the result can be discarded or recalibrated. First-order and second-order difference analyses can be introduced to describe the trend and acceleration of change, respectively, thus more comprehensively depicting the dynamic characteristics of the construction process. For example, the concrete pouring stage shows rapid height growth, while the change tends to slow down in the structural stabilization stage. Statistical analysis of data from multiple time points can also extract periodic or abrupt change characteristics. Based on initial geometric parameters, parametric modeling methods such as BIM-based component modeling are used to generate a basic three-dimensional structural model, where each component is uniquely determined by its outline, dimensions, and spatial location. Geometric change information in the time series is mapped into the model to achieve dynamic updates of component morphology. For example, when the height of a certain floor increases, the corresponding floor slab position in the model moves upward over time. To achieve continuous change effects, interpolation methods such as linear interpolation or spline interpolation can be used to generate intermediate states between different time points. The model update frequency can be consistent with the inspection cycle, such as updating daily, and can be increased to once every 6 hours during critical phases. To enhance realism, the original image texture can be mapped onto the surface of the 3D model, achieving the fusion of geometric and visual information. The resulting digital twin model can reflect the spatial structure and progress status of the construction site in real time, and supports progress comparison, deviation analysis, and prediction functions. For example, by comparing the planned model and the actual model, the progress deviation error range can be controlled within 5%.

[0034] In this embodiment, step S3 includes the following steps: On-site material flow detection is performed based on the image sequence of the construction phase, and the digital twin model is annotated in three dimensions to construct a material annotation model; Target identification is performed on the construction phase image sequence to mark personnel at the construction site; Continuous behavioral tracking of personnel at the construction site is conducted to extract their behavioral trajectories; the key behavioral trajectories include personnel movement trajectories, equipment operating paths, and process operation actions. The construction behavior trajectory is analyzed to obtain construction behavior characteristics; the construction behavior characteristics include operation path, dwell time, operation frequency and process execution sequence. Identify the spatial coordinates and timestamps of construction activities to obtain construction tags; The characteristics of construction behavior are digitally mapped based on construction tags, and the material labeling model is dynamically updated to build a dynamically updated twin model.

[0035] In this embodiment, material category identification and flow detection are performed on the images. Material categories typically include reinforcing bars, formwork, concrete components, precast slabs, aggregates, and construction equipment, with the number of categories set to 6–10. Object detection models such as YOLOv5 or YOLOv8 are used to detect the images, with an input resolution of 640×640 or 1280×1280, a confidence threshold of 0.5, and a non-maximum suppression threshold of 0.45 to balance detection accuracy and speed. Training data should include different stacking states, occlusion conditions, and lighting conditions, with a recommended sample size of at least 10,000 images. After detection, multi-frame association methods, such as IoU-based matching with appearance features, are used to track the same material across frames, identifying its spatial position changes and thus determining the material's flow path. Combining the spatial calibration results of previous images, the material's position in the image is mapped to a three-dimensional spatial coordinate system, achieving three-dimensional positioning of the material. In the existing 3D model, the corresponding locations are labeled. The labeling information includes material type, quantity (which can be estimated by counting with a detection frame or by volume estimation), and time information. To ensure stability, the positions of the same material in consecutive frames are weighted and averaged within a window of 3–5 frames.

[0036] For personnel identification at construction sites, a specialized human detection model is used for target recognition, such as a YOLO series or Faster R-CNN model. The input image size is generally uniformly 640×640, and the detection categories mainly include personnel and optional safety equipment categories such as safety helmets and safety vests. Training data needs to cover different postures (standing, bending, walking), different densities of single individuals, groups of people, and different occlusion conditions; the sample size is recommended to be between 8000 and 15000 images. During model training, a learning rate of 1e-3, a batch size of 16, and approximately 100 training epochs are set, achieving a target detection accuracy (mAP) of over 85%. The detection results are output as personnel bounding boxes and their confidence scores, along with timestamp information. To reduce false positives and false negatives, background modeling or semantic segmentation results can be used for auxiliary filtering, such as excluding misidentification of personnel outside construction areas. For personnel in continuous image sequences, an ID-based assignment method is used for labeling, giving each person a unique identifier in the time series. Based on the personnel detection results, cross-frame tracking is performed to form continuous behavioral trajectories. Multi-target tracking algorithms such as DeepSORT or ByteTrack are employed, combining target location bounding boxes, motion information, and appearance features for matching. An IoU threshold of 0.3–0.5 is set during matching, and cosine similarity is introduced to measure the consistency of appearance features, improving tracking stability under occlusion conditions. Each person forms a time-series trajectory in consecutive frames, with trajectory points composed of spatial locations obtained through projection transformation and timestamps. For equipment operation paths such as tower cranes and small machinery, the same method can be used for detection and tracking, or identification can be based on specific equipment models. For process operation actions, human keypoints can be extracted using pose estimation methods such as OpenPose or HRNet, and combined with action recognition models such as LSTM or 3D CNN to identify specific operational behaviors, such as handling, installation, or pouring. The trajectory sampling interval is typically once per frame, approximately 0.03–0.1 s, and trajectory smoothing algorithms such as Kalman filtering are used to reduce jitter.

[0037] Multi-dimensional feature analysis is performed on the acquired behavioral trajectories. Based on trajectory points, the length and direction of the movement path of personnel or equipment are calculated to form operational path features. Secondly, by analyzing time periods in the trajectory where the speed is close to zero (e.g., speed <0.1 m / s lasting more than 5 seconds), dwell areas are identified and dwell time is calculated. Work frequency can be obtained by statistically analyzing the number of occurrences of a certain type of operation per unit time, such as the number of times a material is moved per hour or the number of times equipment is run. The sequence of work processes is identified by sorting the action sequences by time and matching them with a predefined construction process template to identify the order of moving, installing, and fixing operations. To improve feature stability, the trajectory data is divided into time windows, such as 5–10 minutes per window, and statistical analysis is performed within each window. Abnormal behavior can be detected by setting thresholds, such as abnormally prolonged dwell time or a path deviation from the designated construction area exceeding 2 meters.

[0038] Based on the behavioral trajectory, the two-dimensional image coordinates are converted into three-dimensional spatial coordinates, relying on the results of previous image calibration and 3D reconstruction to achieve coordinate mapping. Each trajectory point corresponds to spatial coordinates x, y, z and timestamp information with millisecond-level accuracy. By projecting the trajectory points onto the 3D model, the construction area, such as floor, zone, or component location, can be determined. Further, combined with construction task division rules, the trajectory is associated with specific construction tasks; for example, a trajectory can be labeled as "second-floor beam installation operation." The label content includes spatial location, time interval, behavior type, and associated component information. To ensure label accuracy, a spatial buffer method can be used, such as setting a range of 1–2 m belonging to the same component for matching, while constraining temporal continuity, such as a label duration of no less than 10 seconds.

[0039] The spatial locations of personnel, equipment, and materials are mapped in real time to corresponding locations in the 3D model, and visualized through color, markers, or animation. Different types of operations can be represented by different colors, and personnel movement paths can be displayed through trajectory lines. Material status is updated based on behavioral characteristics; for example, when a handling activity is detected, the corresponding material location is moved from its original area to the target area, and quantity information is updated simultaneously. The model update frequency can be set to match the image acquisition frequency, such as once per minute or every 5 minutes, and during critical construction phases, the real-time update latency can be controlled within 1–2 seconds. To ensure data consistency, a time synchronization mechanism is used for the update process, ensuring that all objects are refreshed on the same time reference. Through continuous updates, the digital model can reflect the real-time personnel activities, equipment operation, and material flow at the construction site, realizing the transformation from a static model to a dynamic digital twin model, and supporting functions such as progress monitoring, resource scheduling, and anomaly early warning.

[0040] In this embodiment, the specific steps for detecting on-site material flow based on construction phase image sequences and constructing a material annotation model by performing three-dimensional annotation on the digital twin model are as follows: On-site material flow detection is performed on the image sequence during the construction phase to identify material flow information; Material attribute identification is performed based on image sequences during the construction phase to extract multi-dimensional material attribute information; Material consumption rate is calculated based on multi-dimensional material attribute information to obtain the consumption rate of different materials; The consumption rate and material flow information are used to create a three-dimensional annotation of the digital twin model, thus constructing a material annotation model.

[0041] The material flow information includes all stages of material entry, processing, and transportation; the multi-dimensional material attribute information includes material type, quantity, status, location, and flow time.

[0042] In this embodiment, the images are preprocessed, including color correction, brightness equalization, and noise reduction, to ensure consistency across different time points. Material detection employs a deep learning object detection model, such as YOLOv5 or YOLOv8. The input image resolution is set to 1280×1280 to ensure accuracy in recognizing small-sized materials. The confidence threshold is set to 0.5, and the non-maximum suppression threshold is set to 0.45 to balance the probability of missed detections and false detections. After detection, the same material is continuously tracked using a cross-frame association algorithm. This method combines IoU matching with appearance feature similarity. The matching threshold IoU can be set to 0.3–0.5 to ensure unique identification even under occlusion or changing viewing angles. The material's state changes at each stage are recorded using time series data, including arrival at the warehouse, inspection passed or awaiting processing, processing completed, and transportation to the construction site. Each status node records a timestamp and spatial coordinates. Three-dimensional coordinate mapping is achieved through image spatial calibration, forming a complete material flow trajectory and time sequence. Material attributes include type (e.g., steel bars, formwork, concrete components), quantity, status (awaiting processing, processing, installed), location, and flow time. Type recognition relies on the classification output of the target detection model, supplemented by color features, shape features, and texture information for verification, improving the accuracy of distinguishing different material categories. Quantity calculation is achieved through detection frame counting and stacking estimation methods, such as combining stacking height and surface area with density coefficients for volume conversion, yielding accurate quantities. Status recognition combines action analysis and time sequence judgment; materials being processed are accompanied by processing actions or equipment operation trajectories; material transportation is shown as continuous positional movement. Location attributes are obtained through image calibration and 3D reconstruction methods, with accuracy controllable within the 2–5 cm range. Flow time is obtained by recording the timestamps of material entering and leaving each stage, providing a time basis for calculating consumption rates. On-site workers use RFID tags to record material processing status information. During subsequent inspections, RFID scanning can identify processing information and the specific flow of materials.

[0043] By utilizing material attribute information and flow trajectories, the consumption rate of different materials is calculated. The consumption rate is defined as the change in material quantity or usage per unit time, calculated as ΔN / Δt, where ΔN is the quantity consumed per unit time and Δt is the time interval, which can be set daily or hourly, adjusted according to the inspection frequency. Material state sequences are filtered, with only changes from the ready-to-use state to the construction and installation state being statistically analyzed to exclude quantities temporarily stored or transferred. For volumetric materials such as concrete and aggregates, consumption can be calculated through changes in stockpile volume, and the pixel dimensions are converted to spatial dimensions using voxel estimation methods from 3D image reconstruction. To reduce measurement noise, a moving average is applied to the consumption at consecutive time points, with a window length of 3–5 frames or 1–3 days. Through statistical analysis, dynamic consumption rate curves for various materials are obtained, reflecting the rhythm and peak periods of different material usage during construction. Combining the material's 3D location, flow trajectory, and consumption rate, this information is mapped into a digital twin model, achieving 3D annotation. The spatial location of materials is mapped to corresponding components or areas in the digital model. Material types are represented by colors or icons, such as red for steel bars and blue for formwork. Consumption rates can be indicated by color gradients or dynamic animations to show high-usage areas, displaying real-time material consumption trends. Information on the flow of materials is displayed in the 3D model as time markers or path lines; for example, entry paths are marked with dashed lines, processing paths with arrows, and transportation trajectories with continuous curves. The marker update frequency is synchronized with image acquisition, and can be set to daily updates or real-time updates for critical stages with a 1-2 minute delay. Continuity and stability are ensured through 3D buffers and time windows. Changes in material quantity are presented through dynamic counting tags, and the change process for volumetric materials can be displayed through a semi-transparent 3D display.

[0044] In this embodiment, step S4 includes the following steps: Import a pre-built as-built BIM model; Based on the as-built BIM model, the dynamic updated twin model is spatially registered, and the morphological deviation of each component is calculated to obtain the morphological deviation parameters of different components; the morphological deviation parameters include dimensional deviation and positional deviation.

[0045] Based on the aforementioned morphological deviation parameters, the construction progress is estimated to obtain the progress completion rate.

[0046] In this embodiment, the BIM model from the as-built stage is imported into the digital twin platform. The BIM model includes the 3D geometric information, component attributes, and topological relationships of building components. The model format is typically IFC, Revit, or OBJ, and can be unified to a format supported by the platform using conversion tools. During the import process, the component geometric information is parsed, including node coordinates, dimensional parameters (length, width, height), component type, and material attributes. Unique identifiers for components, such as component IDs or numbers, are also extracted for subsequent matching with the digital twin model. For complex building components, such as steel frame structures or precast slabs, the model contains detailed cross-sectional parameters, hole locations, and connection node information. Spatial discretization can be performed using meshing or voxelization methods to improve the accuracy of comparison with the measured digital twin model. After the model is imported, the overall model needs to undergo coordinate system unification and unit conversion. The 3D position, size, and orientation of each component in the twin model are dynamically updated using UAV imagery and sensor information. The goal of spatial registration is to align the real-time components with their corresponding components in the as-built BIM model in a unified coordinate system. Registration methods typically employ rigid body transformation matrices, including translation, rotation, and scaling adjustments. Initial coarse registration can be achieved using the ICPiterative ClosestPoint algorithm, followed by fine registration using nonlinear optimization methods based on feature points. For example, least squares optimization can be used to reduce reprojection error, keeping the average registration error within 5–10 cm. After registration, morphological deviation parameters are calculated by comparing the geometric information of the real-time component and the BIM component. Dimensional deviations are obtained by comparing the differences between the component's length, width, and height and the BIM design dimensions. Positional deviations are obtained by calculating the differences in the three-dimensional coordinates of the component's geometric center or key nodes. The beam length deviation ΔL can be defined as the measured length minus the BIM length, and the center point position deviation ΔP can be calculated using Euclidean distance, with an accuracy controlled within ±2–5 cm. For complex components, deviations can be calculated separately using segmentation or meshing to obtain more detailed morphological deviation information.

[0047] After obtaining the dimensional and positional deviations of each component, deviation thresholds can be used to determine the completion status of component construction. Construction completion criteria are set; for example, if both dimensional and positional deviations are within ±5 cm, the component is considered complete; exceeding the threshold indicates incomplete construction or quality deviations. The construction progress completion rate is calculated by statistically analyzing the ratio of completed components to the total planned number of components. To enhance accuracy, weighting coefficients can be assigned to different types of components; for example, structural main components have a higher weight (e.g., 0.6), while auxiliary facilities have a lower weight (e.g., 0.2–0.3). The overall progress completion rate is calculated using a weighted average. Progress calculations can be dynamically updated using time-series data, such as updating the completion rate daily and plotting construction progress curves to display the completion status of different areas, floors, or component categories. Furthermore, components with large deviations can be marked to generate an early warning list, providing a visual basis for construction management decisions.

[0048] In this embodiment, step S5 includes the following steps: Based on the progress completion rate, the completion time is predicted to obtain the predicted completion time point; Extract the construction planning log; calculate the completion time of key milestones and the overall completion time based on the construction planning log; Based on the completion time of the key nodes and the overall completion time, the project progress deviation is calculated for the predicted completion time point to obtain the expected progress deviation value. Based on the expected schedule deviation, adjust the allocation of engineering resources and output the engineering adjustment plan; Intelligent engineering cycle collaborative control processing is carried out based on engineering adjustment schemes.

[0049] In this embodiment, the remaining workload is estimated. The remaining workload can be inferred from the percentage of completed components to determine the number of uncompleted components, and the estimated construction time is calculated by combining the construction cycle and complexity of various components. For structural beams and columns, the construction time can be obtained by multiplying the average construction time per component by the number of uncompleted components; for processes such as decoration or electromechanical installation, the average daily construction capacity is used for estimation. Using the progress completion rate curve over time, regression prediction methods such as linear regression, exponential smoothing, or ARIMA time series models can be used to predict the remaining construction period, thereby deriving the predicted completion time. The prediction accuracy can be verified using historical stage data, for example, by comparing the actual completion time at each key stage and adjusting the regression coefficients.

[0050] The construction planning log records the construction schedule, including daily tasks, key milestones, and the sequence of procedures. The log is structured and parsed to extract key milestone information, such as completion of foundation construction, topping out of the main structure, completion of mechanical and electrical installation, and completion of the finishing phase. The start and finish times of the planned milestones are recorded. Based on the planned duration of each key milestone and the dependencies between milestones, the overall completion time is calculated. Key milestone completion times can be marked using milestones set in the log, or analyzed using Gantt charts or network planning methods such as CPM or PERT to obtain the earliest completion time, latest completion time, and float time for each milestone.

[0051] The projected completion date is compared with the completion dates of key milestones and the overall completion date in the construction planning log to calculate the expected schedule deviation. The deviation can be calculated using either absolute deviation ΔT = predicted time – planned time or relative deviation percentage deviation = ΔT / planned time × 100%, used to quantify the degree of lag or advancement of the current construction status relative to the plan. When the predicted completion date is 10 days later than planned, and key milestones are lagging by an average of 2–5 days, the project can be considered delayed. The overall expected schedule deviation is obtained by weighting the deviations of each key milestone according to their importance (e.g., structural completion weight 0.4, mechanical and electrical installation weight 0.3).

[0052] After obtaining the expected schedule deviation, resource allocation is adjusted using resource optimization algorithms. The types of resources required for lagging critical nodes are identified, including construction personnel, machinery and equipment, and material supplies, and the current available resource quantity is assessed. Optimization methods such as linear programming, constraint satisfaction, or genetic algorithms are used to adjust resource allocation to shorten the construction period of lagging nodes or improve construction efficiency. When structural construction is delayed by 5 days, the number of construction teams can be increased, construction time extended, or the construction sequence optimized, while equipment scheduling is adjusted to meet the increased work demands. The adjustment plan includes construction team configuration, machinery and equipment scheduling, critical material supply time and quantity, and temporary overtime or night shift arrangements.

[0053] Based on the adjusted engineering plan, the digital twin model is updated in real time to achieve intelligent collaborative control of the engineering cycle. The adjusted construction plan and resource allocation are mapped into the digital twin model, including the location of construction personnel, equipment operation trajectory, material allocation, and key node process arrangement. Through dynamic simulation of the construction process, the feasibility and effectiveness of the adjustment plan can be verified in real time, such as observing changes in construction team density, adjustments to material transportation routes, and improvements in equipment utilization. Optimization objectives can be set during the control process, such as minimizing overall completion delays, maximizing resource utilization, and controlling cost consumption. Based on the simulation, the construction progress status in the digital twin model is updated to keep the model synchronized with on-site construction. Collaborative control not only reflects construction progress optimization but also displays resource distribution, key node status, and future construction predictions through a visual interface, forming an intelligent management and control system that achieves full-process management from plan adjustment to real-time dynamic control.

[0054] In this embodiment, a system for constructing and updating a digital twin model for the entire lifecycle of an engineering project is provided. An image acquisition module provides construction images, a modeling module constructs a digital twin model based on the images, an updating module dynamically updates the model in real time and tracks its behavior, a progress estimation module analyzes the construction progress based on the as-built BIM, and a control module adjusts resources and implements intelligent collaborative control based on the completion rate. This system is used to execute the aforementioned method for constructing and updating a digital twin model for the entire lifecycle of an engineering project, including: The image acquisition module is used to conduct periodic inspections of the construction site based on the UAV according to the preset route, and to perform motion blur removal and extract the image sequence of the construction stage. The modeling module is used to perform geometric state differential modeling based on the image sequence of the construction phase and build a digital twin model; The update module is used to track the behavior of construction personnel in the image sequence during the construction phase and dynamically update the digital twin model to build a dynamically updated twin model. The progress estimation module is used to import the pre-built as-built BIM model; based on the as-built BIM model, it performs spatial registration and progress estimation on the dynamically updated twin model to obtain the progress completion rate. The control module is used to adjust the allocation of engineering resources according to the progress completion rate and output the engineering adjustment plan; and to perform intelligent engineering cycle collaborative control processing based on the engineering adjustment plan.

[0055] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0056] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein are implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for constructing and updating a digital twin model for the entire lifecycle of an engineering project, characterized in that, Includes the following steps: Step S1: Based on the UAV, periodically inspect the construction site according to the preset route, perform motion blur removal, and extract the image sequence of the construction stage; Step S2: Perform geometric state differential modeling based on the image sequence of the construction phase to construct a digital twin model; Step S3: Track the behavior of construction personnel in the video sequence during the construction phase, and dynamically update the digital twin model to build a dynamically updated twin model; Step S4: Import the pre-built as-built BIM model; perform spatial registration and progress estimation on the dynamically updated twin model based on the as-built BIM model to obtain the progress completion rate; Step S5: Adjust the allocation of engineering resources according to the progress completion rate and output the engineering adjustment plan; perform intelligent engineering cycle collaborative control processing based on the engineering adjustment plan; The specific steps of step S1 are as follows: The drones conduct periodic inspections of the construction site according to a preset route and extract the first construction images. Collect flight attitude information and timestamps during the drone inspection process; The flight attitude information and timestamp are used to calibrate the construction image to obtain a second construction image. Perform time-series alignment and motion blur filtering on the second construction image to output the third construction image; Keyframe identification was performed on the third construction image, and multiple keyframes were extracted. Based on multiple keyframes, construction stages are identified and marked, and a sequence of construction stage images is output. The specific steps of step S2 are as follows: Semantic segmentation is performed on image sequences during the construction phase to identify multiple construction components; Calculate the outline, floor height, span, and cross-sectional dimensions of the construction component to obtain its geometric parameters; Based on the image sequence of the construction phase, the geometric parameters are tracked for temporal changes to obtain the geometric parameter sequence at different time points; Geometric state difference analysis is performed based on the geometric parameter sequence to obtain the time-varying characteristics of the component. Dynamic digital 3D modeling is performed based on the time-varying characteristics of the components to construct a digital twin model; The specific steps of step S3 are as follows: On-site material flow detection is performed based on the image sequence of the construction phase, and the digital twin model is annotated in three dimensions to construct a material annotation model; Target identification is performed on the construction phase image sequence to mark personnel at the construction site; Continuous behavioral tracking of personnel at the construction site to extract their behavioral trajectories; The construction behavior trajectory is analyzed to obtain construction behavior characteristics; the construction behavior characteristics include operation path, dwell time, operation frequency and process execution sequence. Identify the spatial coordinates and timestamps of construction activities to obtain construction tags; The characteristics of construction behavior are digitally mapped based on construction tags, and the material labeling model is dynamically updated to build a dynamically updated twin model.

2. The method for constructing and updating a digital twin model for the entire lifecycle of an engineering project according to claim 1, characterized in that, The specific steps for detecting on-site material flow based on construction phase image sequences and constructing a material annotation model by adding 3D annotations to the digital twin model are as follows: On-site material flow detection is performed on the image sequence during the construction phase to identify material flow information; Material attribute identification is performed based on image sequences during the construction phase to extract multi-dimensional material attribute information; Material consumption rate is calculated based on multi-dimensional material attribute information to obtain the consumption rate of different materials; The consumption rate and material flow information are used to create a three-dimensional annotation of the digital twin model, thus constructing a material annotation model.

3. The method for constructing and updating a digital twin model for the entire lifecycle of an engineering project according to claim 2, characterized in that, The material flow information includes all stages of material entry, processing, and transportation; the multi-dimensional material attribute information includes material type, quantity, status, location, and flow time.

4. The method for constructing and updating a digital twin model for the entire lifecycle of an engineering project according to claim 3, characterized in that, The specific steps of step S4 are as follows: Import a pre-built as-built BIM model; Based on the as-built BIM model, the dynamic updated twin model is spatially registered, and the morphological deviation of each component is calculated to obtain the morphological deviation parameters of different components. Based on the aforementioned morphological deviation parameters, the construction progress is estimated to obtain the progress completion rate.

5. The method for constructing and updating a digital twin model for the entire lifecycle of an engineering project according to claim 4, characterized in that, The morphological deviation parameters include dimensional deviation and positional deviation.

6. The method for constructing and updating a digital twin model for the entire lifecycle of an engineering project according to claim 5, characterized in that, The specific steps of step S5 are as follows: Based on the progress completion rate, the completion time is predicted to obtain the predicted completion time point; Extract the construction planning log; calculate the completion time of key milestones and the overall completion time based on the construction planning log; Based on the completion time of the key nodes and the overall completion time, the project progress deviation is calculated for the predicted completion time point to obtain the expected progress deviation value. Based on the expected schedule deviation, adjust the allocation of engineering resources and output the engineering adjustment plan; Intelligent engineering cycle collaborative control processing is carried out based on engineering adjustment schemes.

7. A system for constructing and updating digital twin models for the entire lifecycle of engineering projects, characterized in that, The method for constructing and updating a digital twin model for the entire engineering lifecycle as described in claim 1 includes: The image acquisition module is used to conduct periodic inspections of the construction site based on the UAV according to the preset route, and to perform motion blur removal and extract the image sequence of the construction stage. The modeling module is used to perform geometric state differential modeling based on the image sequence of the construction phase and build a digital twin model; The update module is used to track the behavior of construction personnel in the image sequence during the construction phase and dynamically update the digital twin model to build a dynamically updated twin model. The progress estimation module is used to import the pre-built as-built BIM model; based on the as-built BIM model, it performs spatial registration and progress estimation on the dynamically updated twin model to obtain the progress completion rate. The control module is used to adjust the allocation of engineering resources according to the progress completion rate and output the engineering adjustment plan; and to perform intelligent engineering cycle collaborative control processing based on the engineering adjustment plan.

Images