Building implementation time twin modeling method and system based on digital twinning

Abstract

The application discloses a building implementation time twin modeling method and system based on digital twinning, relates to the technical field of building engineering twin modeling, and comprises the following steps: collecting multi-source physical field real-time monitoring data of structure strain time sequence, environment temperature and humidity time sequence and construction machinery pose sequence in a building implementation process through a field internet of things sensing network; constructing a basic digital model grid which is geometrically isomorphic with a physical field; and directly assigning physical field data to grid vertices to form a basic twin body. The basic twin body is subjected to multi-physical field coupling simulation to generate a twin simulation field, evolution feature points exceeding stress gradient, temperature gradient or displacement gradient are extracted, a pre-trained physical guide neural network is input, and a virtual regulation and control field is generated according to the physical properties and gradient information of the feature points. The application realizes dynamic balance correction of the twin simulation field, and improves the physical field coupling precision and dynamic adaptability of the building implementation digital twin modeling.

Description

Technical Field

[0001] This invention belongs to the field of building engineering twin modeling technology, specifically a building implementation twin modeling method and system based on digital twins. Background Technology

[0002] Existing digital twin modeling technologies for building construction often employ on-site IoT sensor networks to collect multi-source physical field monitoring data on structural strain, environmental temperature and humidity, and construction machinery posture. This only completes the discrete storage of physical data and the independent construction of geometric models. Multi-physics coupling simulations only output static or dynamic distribution results of stress, temperature, and displacement fields. They rely on traditional numerical simulation methods to complete field distribution calculations without combining gradient characteristics for targeted processing or introducing neural networks to achieve dynamic control of the simulation field.

[0003] In existing technologies, geometric model construction and physical field attribute assignment are separated. The basic digital model mesh cannot be directly bound to real-time physical field data. There is a disconnect between the mapping of physical attributes between the twin and the physical building site. Multiphysics coupling simulation only presents the field distribution state and cannot identify the spatial locations where stress gradient, temperature gradient, and displacement gradient exceed the threshold. The imbalance state of the twin simulation field cannot be generated into an appropriate virtual control field through intelligent means, making it difficult to achieve dynamic balance correction of the simulation field.

[0004] The present invention aims to enable real-time monitoring data of multi-source physical fields to be directly assigned to the vertices of the geometric isomorphic model mesh to form a basic twin that carries physical properties. The evolutionary feature points of gradient over-limit are extracted from the twin simulation field and a virtual control field that tends to be in equilibrium is generated through a physical-guided neural network. Summary of the Invention

[0005] This invention aims to solve at least one of the technical problems existing in the prior art; To this end, the present invention proposes a digital twin-based twin modeling method for building implementation, comprising: Real-time monitoring data of multi-source physical fields during the construction process is obtained through on-site IoT sensor networks. The real-time monitoring data of multi-source physical fields includes structural strain time series, environmental temperature and humidity time series, and construction machinery posture series. Based on the real-time monitoring data of the multi-source physical field, a basic digital model mesh that is geometrically isomorphic to the physical building implementation site is constructed in the virtual space. The vertices of the basic digital model mesh are endowed with physical field properties derived from the real-time monitoring data of the multi-source physical field, forming a basic twin that carries physical properties. Multiphysics coupling simulation is performed on the basic twin to solve the dynamic distribution of stress field, temperature field and displacement field inside the basic twin caused by the difference of physical field properties, and generate a twin simulation field describing the physical field coupling state. Evolution feature points that satisfy preset evolution characteristics are extracted from the twin simulation field. The evolution feature points are composed of the spatial positions in the twin simulation field where the stress gradient, temperature gradient or displacement gradient exceeds the corresponding evolution threshold. Evolutionary feature points and their neighborhood data are input into a pre-trained physical guidance neural network. The generator of the physical guidance neural network generates a virtual control field that makes the twin simulation field tend to be in equilibrium based on the physical field properties and gradient information of the evolutionary feature points.

[0006] Furthermore, the acquisition of real-time monitoring data of multi-source physical fields during the building construction process through on-site IoT sensor networks includes: Strain sensor arrays deployed at key parts of the building structure periodically collect micro-strain data of the structure to form a time series sequence of structural strain. Environmental sensor nodes deployed in the construction area collect air temperature and relative humidity data to form a time series of environmental temperature and humidity. Positioning and attitude sensors attached to large construction machinery collect the spatial coordinates and attitude angle data of the machinery to form a sequence of the construction machinery's pose. The time series of structural strain, environmental temperature and humidity, and construction machinery pose are processed by timestamp alignment, outlier removal, and missing data interpolation to form a multi-source physical field real-time monitoring data set with a unified time reference.

[0007] Furthermore, the step of constructing a basic digital model mesh in virtual space that is geometrically isomorphic to the physical building implementation site based on the real-time monitoring data of the multi-source physical field includes: Import the design geometry from the building information model, discretize the design geometry into a mesh, and generate a basic digital model mesh composed of triangular facets or tetrahedral units. The current value of the structural strain time series in the real-time monitoring data of the multi-source physical field is assigned to the strain attribute of the corresponding grid vertex in the basic digital model mesh. The current value of the environmental temperature and humidity time series in the real-time monitoring data of the multi-source physical field is assigned to the temperature and humidity attributes of the corresponding spatial region grid vertex in the basic digital model grid. The current value of the construction machinery pose sequence is converted into the position and attitude of the virtual construction machinery model and placed in the corresponding spatial coordinates of the basic digital model grid, thus completing the construction of the basic twin that carries physical attributes.

[0008] Furthermore, the multiphysics coupled simulation of the basic twin to solve the dynamic distribution of stress, temperature, and displacement fields within the basic twin caused by differences in physical field properties includes: Based on the strain and temperature / humidity properties of the vertices of the basic twin mesh, the stress constitutive relation and thermal expansion coefficient inside the mesh element are calculated. Establish finite element control equations that take into account the temperature-stress coupling effect; Apply mechanical load boundary conditions calculated from the posture sequence of construction machinery and heat flow boundary conditions determined from the time sequence of ambient temperature and humidity; Solving the finite element control equations yields the stress tensor, temperature scalar, and displacement vector at each mesh vertex of the basic twin, which together constitute the twin simulation field describing the coupling state of the physical fields.

[0009] Further, the extraction of evolutionary feature points satisfying preset evolutionary characteristics from the twin simulation field includes: Traverse the stress field data of all grid vertices in the twin simulation field and calculate the stress gradient magnitude at each grid vertex; The stress gradient magnitude is compared with a preset stress evolution threshold, and all mesh vertices whose stress gradient magnitude exceeds the stress evolution threshold are marked as potential evolution feature points. Spatial clustering is performed on the marked potential evolutionary feature points, and the grid vertex with the largest stress gradient magnitude is selected from each cluster as the representative point. The representative point is the evolutionary feature point that satisfies the preset evolutionary characteristics. Meanwhile, based on temperature field and displacement field data, the same gradient and threshold comparison method is used to extract temperature evolution feature points and displacement evolution feature points.

[0010] Further, the step of inputting the evolutionary feature points and their neighborhood data into a pre-trained physical guidance neural network, wherein the generator of the physical guidance neural network generates a virtual control field that makes the twin simulation field tend to be in equilibrium based on the physical field properties and gradient information of the evolutionary feature points, including: The spatial coordinates of the evolution feature points, the stress tensor, temperature scalar, displacement vector of the evolution feature points, and the corresponding stress gradient, temperature gradient, and displacement gradient are used as the input feature vectors of the physical-guided neural network. The generator of the physical-guided neural network encodes and performs nonlinear transformation on the input feature vector and outputs a set of control parameters, including virtual stress compensation and virtual heat flux compensation applied to the vertices of the neighborhood grid of the evolution feature point. The control parameters are Gaussian diffused in the neighborhood of the evolutionary feature point to generate a spatially continuous virtual control field that covers part of the basic twin. The virtual control field is designed to counteract the excessively high physical field gradient at the evolutionary feature point.

[0011] Furthermore, it also includes: The virtual control field is mapped inversely to the basic digital model mesh, and the vertex physical field properties of the basic digital model mesh are iteratively corrected. The corrected basic digital model mesh constitutes an iteratively updated digital twin. In the virtual space, the digital twin that drives the iterative updates performs multi-physics coupled simulation for the next time step, and the newly generated twin simulation field is compared with the physical field state calculated from real-time monitoring data by residual comparison. When the residual is less than the preset tolerance, it is determined that the iteratively updated digital twin is consistent with the state of the physical building implementation site, and the iteratively updated digital twin at this time is marked as a converged real-time digital twin model. The converged real-time digital twin model is subjected to model lightweighting, and mesh vertices and physical attributes whose influence on the evolution of the physical field is lower than the contribution threshold are removed to generate a lightweight real-time digital twin model. The lightweight real-time digital twin model is pushed to an interactive visualization platform and combined with construction schedule data to generate a twin status report for construction management. The process of back-mapping the virtual control field to the basic digital model mesh and iteratively correcting the vertex physical field properties of the basic digital model mesh includes: The virtual stress compensation and virtual heat flux compensation corresponding to each spatial location in the virtual control field are superimposed on the current stress and temperature attributes of the corresponding grid vertices of the basic digital model grid. For the basic digital model mesh superimposed with the virtual control field, the displacement properties of its mesh vertices are recalculated so that the physical field distribution satisfies the new equilibrium relationship. The basic digital model mesh, after all attribute updates are completed, constitutes the digital twin after one iteration update.

[0012] Furthermore, the process of driving the iteratively updated digital twin in the virtual space to perform multiphysics coupled simulation for the next time step, and comparing the residuals of the newly generated twin simulation field with the physical field state calculated from real-time monitoring data, includes: Acquire real-time monitoring data of multi-source physical fields at the next sampling time; Based on the monitoring data at the next sampling time, the theoretical stress state and theoretical temperature state at the key locations of the building implementation site at the next sampling time are calculated by interpolation or mechanism model. The digital twin that drives the iterative update performs multiphysics coupling simulation for the next time step, and obtains the twin simulation field for the next time step; Calculate the difference between the state values ​​of the twin simulation field at key locations and the theoretical state values, and take the root mean square of the differences at all key locations as the residual.

[0013] Furthermore, the process of lightweighting the converged real-time digital twin model includes: Calculate the contribution of each grid vertex in the converged real-time digital twin model to the overall stress field and temperature field distribution. The contribution is defined by the change in field distribution caused by removing the grid vertex. Set a contribution threshold, identify and mark all grid vertices whose contribution is below the contribution threshold; The marked mesh vertices and their associated mesh cells are merged and removed. At the same time, the physical properties on the retained mesh vertices are resampled and smoothed to generate a lightweight real-time digital twin model that balances geometric complexity and physical property accuracy. The process involves pushing a lightweight, real-time digital twin model to an interactive visualization platform and combining it with construction schedule data to generate a twin status report for construction management, including: The geometric and physical property data of the lightweight real-time digital twin model are streamed to an interactive visualization platform for 3D rendering; Synchronize the current construction schedule from the construction management database and identify the current construction tasks and target components in the plan; In the interactive visualization platform, the lightweight real-time digital twin model area related to the current construction task is highlighted, and the key physical field data of the lightweight real-time digital twin model area is overlaid. The system automatically extracts key physical field data statistics of the highlighted area, deviations from the design allowable values, and distribution of evolution feature points, and generates a formatted twin status report document containing text, data tables, and key indicator illustrations.

[0014] Furthermore, the present invention also includes a digital twin-based building implementation-time twin modeling system, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the digital twin-based building implementation-time twin modeling method described above.

[0015] Compared with the prior art, the beneficial effects of the present invention are: Multi-source physical field real-time monitoring data, including structural strain time series, environmental temperature and humidity time series, and construction machinery pose series, are directly assigned to the vertices of the basic digital model mesh that is geometrically isomorphic to the physical building implementation site, forming a basic twin that carries physical attributes. The geometric model and physical field attributes are integrated and bound together. The physical field attributes are synchronously mapped to the mesh vertices along with the real-time monitoring data. The basic twin and the physical field state of the physical building implementation site maintain real-time correspondence, eliminating the disconnect between geometric modeling and discrete binding of physical attributes. The distribution of physical field attributes in the virtual space is synchronized with the physical site in real time.

[0016] Spatial locations where stress gradients, temperature gradients, or displacement gradients exceed corresponding evolution thresholds are extracted from the twin simulation field as evolution feature points. The evolution feature points and their neighborhood data are input into a pre-trained physical guidance neural network. The generator generates a virtual control field that makes the twin simulation field tend to be in equilibrium based on the physical field properties and gradient information of the evolution feature points. The gradient over-limit positions of the twin simulation field are accurately located. The physical guidance neural network completes the generation of the virtual control field based on the gradient information. The dynamic distribution of the twin simulation field converges to the equilibrium state. The field distribution imbalance state after multi-physics coupling is dynamically corrected. The virtual control field and the twin simulation field form a suitable coupling state. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of the digital twin-based building implementation twin modeling method described in this invention. Figure 2 A flowchart of the construction of a basic twin; Figure 3 The Gaussian diffusion spatial distribution of the virtual control field; Figure 4 The iterative convergence residual curve of the digital twin; Figure 5 Box plot of evolutionary feature point density distribution. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See Figure 1 This invention provides a digital twin-based method for building implementation twin modeling, the specific method including: Through an IoT sensor network deployed at the construction site, real-time monitoring data of multi-source physical fields reflecting the construction process are continuously collected. This data specifically includes sequences of structural strain changes over time, sequences of ambient temperature and humidity changes over time, and sequences of construction machinery position and attitude changes over time. In virtual space, a basic digital model mesh with the same geometry as the physical construction site is constructed based on the aforementioned real-time monitoring data. Each vertex of this mesh is assigned a physical field attribute value derived from the corresponding monitoring data, thus forming a basic twin that carries real physical properties. Numerical simulation calculations involving multi-physics coupling are performed on this basic twin to solve for the dynamic changes in stress, temperature, and displacement fields caused by differences in the distribution of physical field attributes within it, generating a twin simulation field that describes the coupling state of these physical fields. Spatial location points that meet preset evolution conditions, i.e., evolution feature points, are automatically identified and extracted from this twin simulation field. These feature points consist of mesh vertices in the simulation field where the stress gradient, temperature gradient, or displacement gradient values ​​exceed their respective preset thresholds. These evolutionary feature points and the physical field data within a certain neighborhood are input into a pre-trained physical guidance neural network. The generator part of the network will calculate and output a virtual control field that can act on the twin based on the physical properties and gradient information of the input feature points. The goal of the control field is to make the original twin simulation field evolve towards a more balanced state.

[0020] In one embodiment of the invention, at the construction site, strain sensor arrays deployed at key stress-bearing parts of the building structure collect micro-strain data of the structure at fixed intervals. These time-sequentially arranged collection results form a structural strain time series. Environmental sensor nodes distributed throughout the construction area continuously collect air temperature and relative humidity data, forming an environmental temperature and humidity time series. Positioning and attitude sensors installed on the bodies of large construction machinery such as tower cranes and pump trucks collect the three-dimensional spatial coordinates and attitude angle data such as pitch and yaw in real time, forming a construction machinery posture sequence. The above three sequences are timestamped to ensure that all data have a unified time reference. Outliers in the collected data are removed, and missing data points are filled in using interpolation methods, ultimately forming a standardized and consistent multi-source physical field real-time monitoring data set. When constructing the basic digital model mesh in virtual space, the design geometry of the building information model is first imported, and the geometry is discretized into a mesh to generate a basic digital model mesh composed of triangular facets or tetrahedral elements. The latest values ​​of the structural strain time series from the processed multi-source physical field real-time monitoring data set are assigned to the grid vertices in the basic digital model mesh corresponding to the actual sensor placement locations, serving as their strain attributes. The current values ​​of the environmental temperature and humidity time series are assigned to the grid vertices in the basic digital model mesh corresponding to the actual spatial monitoring areas, serving as their temperature and humidity attributes. The current values ​​of the construction machinery pose series are converted into the specific position coordinates and rotational attitude of the construction machinery's 3D model in virtual space, and this virtual model is placed at the corresponding spatial coordinates in the basic digital model mesh. At this point, a basic twin carrying physical attributes is constructed.

[0021] In practical implementation, strain sensor arrays periodically collect structural micro-strain data at a specific frequency. The structural strain time series is composed of these collection points arranged in chronological order. For example, fiber optic strain sensors are deployed at key stress-bearing parts of the shear wall of a high-rise building's core tube. The sensors collect micro-strain values ​​once per second, and continuous collection for 24 hours forms a structural strain time series containing 86,400 data points. In practical implementation, environmental sensing nodes collect air temperature and relative humidity data at a specific frequency. The environmental temperature and humidity time series is composed of these collection points arranged in chronological order. For example, one temperature and humidity sensing node is deployed in each of the five areas (east, west, south, north, and center) of a construction site. Each node records temperature and humidity values ​​every 30 seconds, and continuous operation for ten days forms an environmental temperature and humidity time series containing 28,800 temperature-humidity data pairs for each node. In practice, positioning and attitude sensors attached to large construction machinery collect spatial coordinates and attitude angle data of the machinery at a specific frequency. The posture sequence of the construction machinery is composed of these time-sequentially arranged collection points. For example, a global navigation satellite system receiver and inertial measurement unit are installed at the base of the boom of a tower crane. The equipment records the three-dimensional spatial coordinates of the crane hook and the pitch and rotation angles of the boom at a frequency of ten times per second. A dense posture sequence of the construction machinery can be formed within one lifting operation cycle. In practice, the above sequence is timestamped to align the data records from different sensing devices, each with its own internal clock, to the Coordinated Universal Time (UTC) time base. The network time protocol is used to synchronize the clocks of each sensing node, so that all data points are accompanied by a unified timestamp accurate to the millisecond level. In practice, outlier removal is performed on the timestamped data. An abnormal collection point that deviates significantly from the overall distribution of the data sequence is identified and removed based on the principle of three standard deviations. In practice, missing data points that may exist after outlier removal are imputed. For structural strain time series and construction machinery pose series, cubic spline interpolation is used to fit curves based on valid data points before and after to fill in missing values; for environmental temperature and humidity time series, linear interpolation is used for completion. In practice, the structural strain time series, environmental temperature and humidity time series, and construction machinery pose series, after timestamp alignment, outlier removal, and missing data imputation, together constitute a multi-source physical field real-time monitoring data set with a unified time reference.

[0022] In practical implementation, a basic digital model mesh is constructed based on a set of real-time monitoring data of multi-source physical fields. (See [reference]). Figure 2The design geometry of building components is imported from the Building Information Model (BIM). This design geometry describes the geometry of walls, beams, and slabs using boundary representation. In practice, the imported design geometry is discretized into a mesh. A leading-edge propagation method is used to generate a basic digital model mesh composed of triangular facets. For complex 3D solid components, the Delaunay triangulation method is used to generate tetrahedral element meshes. In practice, the current value of the structural strain time series in the multi-source physical field real-time monitoring dataset is assigned to the strain attribute of the corresponding mesh vertex in the basic digital model mesh. The current value refers to the data at the latest valid acquisition time. For example, if the latest micro-strain value collected by the strain sensor at the column base located at the intersection of axis 3 and axis B is -125 micro-strain, the mesh vertex corresponding to the spatial coordinates of this column base is found in the basic digital model mesh, and the strain attribute value of this mesh vertex is set to -125. In practical implementation, the current values ​​of the environmental temperature and humidity time series in the multi-source physical field real-time monitoring data set are assigned to the temperature and humidity attributes of the corresponding spatial region grid vertices in the basic digital model grid. For example, if the latest temperature collected by the environmental sensor node located in the southeast corner of the construction area is 28.5 degrees Celsius and the relative humidity is 65%, then in the basic digital model grid, all grid vertices located within a spherical area with a radius of ten meters centered on this node will have their temperature attribute set to 28.5 and their humidity attribute set to 65. In practical implementation, the current values ​​of the construction machinery pose sequence are converted into the position and attitude of the virtual construction machinery model. For example, the latest data of the tower crane's construction machinery pose sequence includes east coordinates, north coordinates, elevation, pitch angle, and rotation angle. Using this data, a three-dimensional model of the tower crane is instantiated in virtual space, with the model's base center placed at the corresponding east coordinates, north coordinates, and elevation. The model's boom and counterweight boom components are then rotated according to the pitch and rotation angles. In practical implementation, the virtual construction machinery model, having completed position and attitude transformation, is placed at the corresponding spatial coordinates of the basic digital model mesh. Together with the building structure mesh, which has been assigned physical properties, it forms a basic twin that carries these physical properties. Within this basic twin, each mesh vertex carries a physical field property value derived from real-time monitoring, and the virtual construction machinery model is located in its actual spatial position. In some embodiments, the quality of the basic digital model mesh generated by mesh discretization needs to be evaluated. The evaluation formula can be: , in: Indicates the overall quality coefficient of the mesh. This indicates the total number of cells in the grid. Indicates the first Volume of each unit Indicates the first The surface area of ​​each cell is used as a coefficient to quantify the regularity of the mesh cells. In some embodiments, converting the current value of the construction machinery pose sequence into the position and orientation of the virtual construction machinery model involves coordinate transformation, from sensor readings in the local coordinate system of the construction machinery to the model pose in the global engineering coordinate system. It can be understood that the construction of the basic digital model mesh depends entirely on the input of a multi-source physics field real-time monitoring dataset.

[0023] Indicates the first The surface area of ​​each cell is used as a coefficient to quantify the regularity of the mesh cells. In some embodiments, converting the current value of the construction machinery pose sequence into the position and orientation of the virtual construction machinery model involves coordinate transformation, from sensor readings in the local coordinate system of the construction machinery to the model pose in the global engineering coordinate system. It can be understood that the construction of the basic digital model mesh depends entirely on the input of a multi-source physics field real-time monitoring dataset.

[0024] In one embodiment of the present invention, a multiphysics coupling simulation is performed on the constructed basic twin. This simulation process calculates the stress-strain constitutive relationship of the material within each mesh element, as well as the material's coefficient of thermal expansion, based on the strain and temperature / humidity properties assigned to the mesh vertices of the basic twin. A set of finite element governing equations is established, comprehensively considering the coupling effects of temperature changes and stress fields. In the governing equations, mechanical loads derived from the pose sequence of the construction machinery are applied as mechanical boundary conditions, while heat flow or temperature conditions determined by the environmental temperature and humidity time series are applied as thermal boundary conditions. The finite element governing equations are solved numerically to calculate the stress tensor, temperature scalar, and displacement vector at each mesh vertex in the basic twin. These data, distributed throughout the entire twin model, collectively constitute a twin simulation field describing the current physical field coupling state.

[0025] In practical implementation, multiphysics coupling simulation is performed on the foundation twin bearing physical properties. The simulation is based on the strain and temperature / humidity properties assigned to the vertices of the foundation twin mesh. Specifically, the stress constitutive relation within the mesh element is calculated based on the strain properties of the foundation twin mesh vertices. For reinforced concrete, Hooke's law is used to calculate the vertex stress from the vertex strain value, and then shape function interpolation is used to obtain the stress at the Gaussian integration point of the element. The thermal expansion coefficient of the material is calculated based on the temperature / humidity properties of the foundation twin mesh vertices. The thermal expansion coefficient is a function of temperature and moisture content, and the value of the thermal expansion coefficient corresponding to each mesh element is determined by looking up tables or empirical formulas. In practical implementation, finite element governing equations considering the temperature-stress coupling effect are established. These governing equations describe how thermal strain caused by temperature field changes affects the stress field distribution, and how the heat generated by the work done by the stress field affects the temperature field. The equation set is established based on the principles of energy conservation and momentum conservation. In practical implementation, mechanical load boundary conditions calculated from the posture sequence of construction machinery are applied. For example, the mass of the suspended load, the angle between the sling and the vertical direction, and the angular velocity of the boom rotation are extracted from the posture sequence of the tower crane. These are then converted into concentrated or distributed forces acting on the corresponding nodes of the digital model mesh of the building structure foundation through mechanical calculations. In practical implementation, heat flux boundary conditions determined by the environmental temperature and humidity time series are applied. For example, using temperature and wind speed data from the environmental temperature and humidity time series, the heat flux density of the convective heat transfer between the outer surface of the digital model mesh of the building structure foundation and the air is calculated using Newton's law of cooling, and this is applied as a thermal boundary condition to the corresponding surface elements of the foundation digital model mesh. In practical implementation, the established finite element governing equations are solved. Numerical methods such as direct methods or iterative methods are used to solve large-scale sparse linear equation systems, calculating the displacement vector of each mesh vertex in the Cartesian coordinate system, the temperature scalar of each mesh vertex, and the stress tensor at each mesh vertex in the foundation twin. In practical implementation, the stress tensor, temperature scalar, and displacement vector of all mesh vertices together constitute a twin simulation field describing the physical field coupling state. The twin simulation field is a dataset covering the entire geometric space of the basic twin, containing the distribution of multiple physical quantities. In some embodiments, establishing the specific form of the finite element governing equations can be expressed as solving a set of coupled equations concerning the displacement field U and the temperature field T. The matrix expression of the governing equations is as follows: , in: Represents the structural stiffness matrix. Represents the thermal-stress coupling matrix. Represents the heat conduction matrix. This represents the displacement vector of all nodes. Represents the temperature vector of all nodes. Represents the mechanical load vector. This represents the heat flow vector.

[0026] In one embodiment of the present invention, evolutionary feature points are extracted from a twin simulation field. The stress field data of all grid vertices in the twin simulation field are traversed, and the magnitude of the stress gradient vector at each grid vertex is calculated. This stress gradient magnitude is compared with a preset stress evolution threshold, and all grid vertices with stress gradient magnitudes exceeding the threshold are marked as potential stress evolution feature points. Spatial clustering analysis is performed on all marked potential stress evolution feature points. From each obtained spatial cluster, the grid vertex with the largest stress gradient magnitude within it is selected as the representative point of that cluster. This representative point is the stress evolution feature point that satisfies the preset evolutionary characteristics. Using the same gradient calculation and threshold comparison method as for the stress field, temperature evolution feature points and displacement evolution feature points can be extracted in parallel based on the temperature field data and displacement field data in the twin simulation field. The extracted evolutionary feature points and their neighborhood data are input into a pre-trained physical-guided neural network. The network's input feature vector consists of the spatial coordinates of the evolutionary feature point, the stress tensor, temperature scalar, and displacement vector data of that point, as well as the corresponding stress gradient, temperature gradient, and displacement gradient data. The generator of the physics-guided neural network encodes and nonlinearly transforms the input feature vector, ultimately outputting a set of control parameters. These parameters include virtual stress compensation and virtual heat flux compensation proposed to be applied to the grid vertices in the neighborhood of the evolutionary feature point. These control parameters are then diffused according to a Gaussian function around the evolutionary feature point, generating a spatially continuous virtual control field covering a portion of the basic twin. The physical significance of this virtual control field is to counteract or reduce excessively high physical field gradients at the evolutionary feature point.

[0027] In the specific implementation, evolutionary feature points that satisfy preset evolutionary characteristics are extracted from the twin simulation field describing the physical field coupling state. The extraction process begins by traversing the stress field data of all grid vertices in the twin simulation field. Specifically, the stress gradient magnitude at each grid vertex is calculated. This magnitude is obtained by calculating the partial derivatives of each component of the stress tensor at the grid vertex in three spatial directions, and its value characterizes the severity of stress change. The calculated stress gradient magnitude at each grid vertex is compared with a preset stress evolution threshold. This threshold is pre-set based on the yield strength and safety factor of the structural material. All grid vertices with stress gradient magnitudes exceeding the stress evolution threshold are marked as potential evolutionary feature points. Spatial clustering analysis is then performed on all marked potential evolutionary feature points. A density clustering algorithm based on Euclidean distance is used to group spatially adjacent potential evolutionary feature points into the same cluster, with each cluster representing a stress concentration region. In practice, from each cluster obtained by spatial clustering, the grid vertex with the largest stress gradient magnitude within that cluster is selected as the representative point of that cluster. This representative point is the stress evolution feature point that satisfies the preset evolution characteristics. See Table 1 for a simplified example of the extraction process.

[0028] Table 1: Example table of stress evolution feature points extracted from twin simulation fields

[0029] In practical implementation, based on temperature and displacement field data, the same gradient calculation and threshold comparison method as for stress field is used to extract temperature evolution feature points and displacement evolution feature points in parallel. Temperature evolution feature point extraction is based on the comparison of temperature gradient and temperature evolution threshold, while displacement evolution feature point extraction is based on the comparison of displacement gradient and displacement evolution threshold. In practical implementation, the extracted evolution feature points and their neighborhood data are input into a pre-trained physical guidance neural network. The neighborhood data of the evolution feature point is defined as the physical field data of all grid vertices within a spherical space centered on the evolution feature point and with radius R, where the value of R is determined by the average grid size. In practical implementation, the input feature vector of the physical guidance neural network is composed of the following concatenated data: the three-dimensional spatial coordinates of the evolution feature point, the six independent components of the stress tensor at the evolution feature point, the temperature scalar at the evolution feature point, the three components of the displacement vector at the evolution feature point, the three components of the stress gradient vector at the evolution feature point, the three components of the temperature gradient at the evolution feature point, and the nine components of the displacement gradient tensor at the evolution feature point. In specific implementations, the generator of the physically guided neural network encodes and performs nonlinear transformations on the input feature vector. Encoding is achieved through fully connected layers, and nonlinear transformations are achieved through multiple hidden layers containing activation functions, ultimately outputting a set of control parameters. In specific implementations, the control parameters include virtual stress compensation and virtual heat flux compensation applied to the vertices of the grid in the neighborhood of the evolutionary feature point. The virtual stress compensation is a six-dimensional vector used to correct the stress tensor, and the virtual heat flux compensation is a three-dimensional vector used to correct the heat flux. In specific implementations, the control parameters are Gaussian diffused within the neighborhood of the evolutionary feature point to generate a spatially continuous virtual control field covering a portion of the basic twin. The Gaussian diffusion function determines the decay mode of the compensation from the evolutionary feature point to the surrounding space. In some embodiments, the average stress gradient of all grid vertices in the neighborhood of the evolutionary feature point is calculated to describe the degree of imbalance in the region. The calculation formula is: , in: This represents the neighborhood average stress gradient magnitude. This represents the total number of grid vertices in the neighborhood. Indicates the first Stress gradient vector at each vertex of the neighborhood grid This represents the operation of vector magnitude.

[0030] See Figure 3In the digital twin-based building implementation twin modeling method, the spatial distribution of the virtual control field is visualized as a double Gaussian diffusion pattern, intuitively reflecting the physical field gradient compensation logic for the evolutionary feature points. Specifically, with the control center (evolutionary feature point) marked by a star as the core, the intensity of the virtual control field exhibits a Gaussian decay distribution along the spatial coordinates, forming a gradient transition zone between two high-value hot zones and the surrounding low-value cold zones. This distribution pattern originates from the Gaussian diffusion process of the control parameters output by the physical guidance neural network in the neighborhood of the evolutionary feature point: the virtual stress compensation and virtual heat flux compensation reach their peak at the control center, corresponding to the red core area with an intensity of approximately 1.05 in the figure; as the spatial distance from the control center increases, the compensation decreases exponentially according to a Gaussian function, gradually transitioning to the blue edge area where the intensity approaches 0, achieving smooth cancellation of the excessively high physical field gradient in the neighborhood of the evolutionary feature point, causing the twin simulation field to converge towards an equilibrium state. The visualization results fully present the spatial range and intensity gradient of the virtual control field, providing an intuitive basis for evaluating the control effect, optimizing the parameters of the physical guidance neural network, and verifying the convergence of the digital twin.

[0031] In one embodiment of the present invention, a virtual control field is back-mapped to a basic digital model mesh, and the physical field properties of the mesh vertices are iteratively corrected. The virtual stress compensation and virtual heat flux compensation corresponding to each spatial location in the virtual control field are superimposed onto the current stress and temperature attribute values ​​of the corresponding mesh vertices in the basic digital model mesh. For the basic digital model mesh after superimposing the virtual control field, the displacement attributes of all mesh vertices are recalculated to ensure that the updated physical field distribution satisfies the new mechanical and thermal equilibrium relationship. The basic digital model mesh with all updated attributes constitutes a digital twin after one iterative update. In virtual space, the iteratively updated digital twin is driven to perform multiphysics coupling simulation for the next time step. At the next sampling moment, new real-time monitoring data of the multi-source physical fields are acquired. Based on the latest monitoring data at the next sampling moment, the theoretical stress state value and theoretical temperature state value that the building implementation site should have at key locations at that moment are calculated using interpolation algorithms or simplified models based on physical mechanisms. The iteratively updated digital twin is then driven to perform simulation to obtain the twin simulation field for the next time step. The difference between the state values ​​of this simulated twin at the key locations and the previously calculated theoretical state values ​​is calculated. The square root of the average of the squared differences at all key locations is then taken to obtain the root mean square residual. When this residual is less than a preset tolerance threshold, it is determined that the state of the currently iteratively updated digital twin is consistent with the state of the physical building implementation site. The digital twin at this moment is marked as a converged real-time digital twin model.

[0032] In the specific implementation, a spatially continuous virtual control field generated by a physics-guided neural network is back-mapped onto the basic digital model mesh. Each spatial coordinate in the virtual control field corresponds to a set of virtual stress compensation and virtual heat flux compensation. In the specific implementation, the physical field properties of the vertices of the basic digital model mesh are iteratively corrected. The virtual stress compensation corresponding to a spatial coordinate in the virtual control field is superimposed on the current stress property of the nearest grid vertex in the basic digital model mesh to that spatial coordinate; this superposition is a vector addition operation. In the specific implementation, the virtual heat flux compensation corresponding to a spatial coordinate in the virtual control field is superimposed on the equivalent heat flux corresponding to the current temperature property of the nearest grid vertex in the basic digital model mesh to that spatial coordinate, completing the indirect correction of the temperature property. In the specific implementation, the displacement properties of the grid vertices of the basic digital model mesh superimposed with the virtual control field are recalculated, and the updated stress and temperature properties are recalculated. By solving the new equilibrium equations, a displacement distribution satisfying the new force and heat balance relationships is obtained. In practice, the basic digital model mesh that has completed all attribute updates constitutes a digital twin after one iteration. Compared with the digital twin before the update, the physical field gradient distribution inside the updated digital twin tends to be flatter.

[0033] In practical implementation, a digital twin driven by virtual space iterative updates performs multi-physics coupling simulation for the next time step, corresponding to the next data sampling period. Real-time monitoring data of the multi-source physical fields is acquired at the next sampling moment. This monitoring data is collected and transmitted by the on-site IoT sensor network at the end of the next fixed time interval. Based on the monitoring data at the next sampling moment, an interpolation algorithm is used to calculate the theoretical stress and temperature states at key locations on the building site at the next sampling moment. The interpolation algorithm uses linear interpolation on the time axis between the monitoring data from the previous and next sampling moments to obtain the estimated values ​​of the key locations at the next sampling moment. Furthermore, based on the monitoring data at the next sampling moment, a mechanistic model is used to calculate the theoretical stress and temperature states at key locations on the building site at the next sampling moment. This mechanistic model consists of simplified physical equations used to infer the state at locations without deployed sensors based on data from monitoring points. In practice, the iteratively updated digital twin is used for multiphysics coupled simulation in the next time step. Using the iteratively updated digital twin as the initial state, boundary conditions corresponding to the next time step are applied, and a finite element solver is run to obtain the twin simulation field for the next time step. In practice, the difference between the state values ​​and theoretical state values ​​at key locations in the twin simulation field is calculated. State values ​​include stress and temperature values, and each key location generates a difference. In practice, the root mean square of the differences at all key locations is calculated as the residual. The residual is used to quantify the overall deviation between the twin simulation field and the theoretical state. Table 2 shows a comparison of the residual calculation data for three key locations at a certain next sampling time.

[0034] Table 2: Example of residual calculation between the twin simulation field and theoretical state value at the next time step

[0035] In specific implementation, when the residual is less than a preset tolerance, it is determined that the iteratively updated digital twin is consistent with the state of the physical building implementation site. The preset tolerance is a positive number pre-set according to engineering accuracy requirements. In some embodiments, the root mean square formula for calculating the residual is: , in: This represents the root mean square residual. This indicates the total number of key positions involved in the comparison. Indicates the first The difference between the simulated field state values ​​and the theoretical state values ​​at key locations. In some embodiments, the theoretical stress state can be calculated using a simplified model based on the plane stress assumption of elasticity.

[0036] See Figure 4In the iterative convergence phase of the twin modeling method during digital twin-based building implementation, the evolution of the root mean square residual with the number of iterations intuitively reflects the approximation process between the twin and the physical site. Specifically, the root mean square residual is used as the core convergence index. Its value is calculated from the root mean square difference between the twin simulation field and the theoretical state of the physical site at key locations, and is used to quantify the degree of deviation between the virtual twin and the physical field of the actual building implementation site. As the number of iterations increased from 1 to 8, the residual curve exhibited a typical monotonically decreasing convergence trend: in the initial iteration (1st iteration), the residual was approximately 2.8, indicating a significant difference in the physical field distribution between the basic digital twin and the physical site; after corrections in the 2nd to 5th iterations, the residual rapidly decreased to around 0.22, with the compensation effect of the virtual control field effectively smoothing the physical field gradient at the evolution feature points and reducing the deviation between the simulation and the measured state; by the 6th iteration, the residual dropped below the preset convergence threshold (0.1) for the first time, signifying that the state of the digital twin and the physical building implementation site had reached consistency; in the subsequent 7th and 8th iterations, the residual further approached 0, verifying the stability and robustness of the convergence results. The convergence characteristics of the curve verified the effectiveness of the method of generating a virtual control field and iteratively correcting the physical field properties of the digital twin using a physical-guided neural network, providing a quantitative basis for determining the convergence state of the real-time digital twin model and promoting subsequent lightweight processing and visualization applications.

[0037] In one embodiment of the present invention, a lightweight model is performed on the converged real-time digital twin model. The contribution of each mesh vertex in the model to the overall stress and temperature field distribution is calculated. The contribution is defined as the change in the overall stress or temperature distribution caused by removing the mesh vertex from the model. A contribution threshold is set, and all mesh vertices with contributions below this threshold are identified and marked. Merging and deletion operations are performed on the marked mesh vertices and their associated mesh cells. Simultaneously, the physical property data carried by the retained mesh vertices are resampled and smoothed to generate a lightweight real-time digital twin model that balances geometric complexity and physical property accuracy. The geometric and physical property data of the lightweight real-time digital twin model are pushed to an interactive visualization platform for 3D rendering and display via streaming. Current construction schedule data is synchronously obtained from the construction management database to identify the current construction tasks and target components set in the plan. In the interactive visualization platform, the model area relevant to the current construction task in the lightweight real-time digital twin model is highlighted, and key physical field data, such as stress contour maps and temperature values, are overlaid on this area. The system automatically extracts statistical values ​​of key physical field data, deviations between these values ​​and design allowable values, and the distribution of evolutionary feature points within the highlighted area. The extracted information is then formatted into a twin status report document containing text descriptions, data tables, and key indicator graphs.

[0038] In the specific implementation, the converged real-time digital twin model undergoes model lightweighting. The contribution of each mesh vertex in the converged real-time digital twin model to the overall stress and temperature field distribution is calculated. The contribution is defined by the change in field distribution caused by removing the mesh vertex. A specific contribution threshold is set, based on a trade-off between visualization accuracy and computational resource consumption. All mesh vertices with contributions below the threshold are identified and marked. The marked mesh vertices and their associated mesh cells are merged and removed. For marked mesh vertices, if they are inside the model, vertex deletion and mesh re-partitioning are performed; if they are on the boundary, they are merged with adjacent vertices. Triangular facets or tetrahedral cells associated with the marked mesh vertices are removed or degenerated. Simultaneously, the physical properties of the retained mesh vertices are resampled and smoothed. Resampling uses interpolation algorithms to redistribute attribute values ​​from the original dense mesh's physical property data to new, sparser mesh vertices. Smoothing uses Gaussian filtering or mean filtering to reduce high-frequency noise in the attribute data. In practice, a lightweight real-time digital twin model is generated that balances geometric complexity and physical property accuracy. The number of mesh vertices in the lightweight real-time digital twin model is significantly less than that in the original converged real-time digital twin model, but the physical field distribution characteristics of key regions are preserved.

[0039] In practical implementation, the geometric and physical property data of the lightweight real-time digital twin model are streamed to an interactive visualization platform for 3D rendering. The geometric data includes the coordinates of mesh vertices and the connection information of elements. The physical property data includes the stress, temperature, and displacement values ​​of each vertex. The streaming transmission employs a segmented loading method to support network transmission and real-time rendering of large-scale models. In practical implementation, the current construction schedule is synchronized from the construction management database. The database is queried via an application programming interface (API) to obtain the list of planned construction tasks, task start and end times, and associated component numbers. In practical implementation, the current construction tasks and target components in the construction schedule are identified. The ongoing current construction task is determined by comparing the current system time with the task start and end times, and the unique identifier of the target component is parsed from the description of the current construction task. In practical implementation, the lightweight real-time digital twin model area related to the current construction task is highlighted on the interactive visualization platform. Highlighting is achieved by changing the rendering color of the model area, adding contour lighting, or using semi-transparent overlays. In practical implementation, key physical field data of the lightweight real-time digital twin model area are overlaid and displayed on the interactive visualization platform. This key physical field data is rendered on the 3D model surface in the form of isosurface cloud maps, color maps, or numerical labels. In practical implementation, statistical values ​​of key physical field data in highlighted areas are automatically extracted. These statistical values ​​include the maximum, minimum, average, and standard deviation of stress values ​​at all grid vertices within the highlighted area. In practical implementation, the deviation between the physical field data of the highlighted area and the design allowable values ​​is automatically extracted. The physical field data of the grid vertices within the highlighted area are compared with the allowable stress and allowable deformation values ​​of the material specified in the structural design code, and the percentage deviation is calculated. In practical implementation, the distribution of evolutionary feature points in the highlighted area is automatically extracted, and the number, spatial density, and cluster center locations of these evolutionary feature points are statistically analyzed. In practical implementation, a formatted twin status report document containing text, data tables, and key indicator graphs is generated. The text section describes the current construction task overview and model status summary. The data tables present the extracted statistical values ​​and deviation values, and the key indicator graphs show the trend of physical field data over time or space using line graphs or bar charts. In some embodiments, the contribution of mesh vertices to the overall stress field and temperature field distribution. It can be calculated using the following formula:

[0040] in: This represents the contribution of vertex v. Weighting coefficients representing stress field variations. This represents the norm of the overall stress field change in the model caused by removing vertex v. Weighting coefficients representing changes in the temperature field. This represents the absolute value of the change in the average temperature of the overall temperature field of the model caused by removing vertex v. In some embodiments, synchronizing the current construction schedule from the construction management database can employ a timed polling or event-triggered mechanism. Optionally, the highlighting can use a blinking effect to enhance visual cues. The twin status report document can be output in PDF or HTML web page format. It is understood that lightweight model processing reduces data storage and transmission load, and improves the smoothness of interactive visualization. The twin status report provides construction managers with a decision support document that integrates real-time physical status and planning information.

[0041] See Figure 5 In the statistical analysis of the density distribution of evolutionary feature points, a funnel-shaped box plot visually presents the overall distribution and dispersion characteristics of the density values ​​of evolutionary feature points during the construction process. From the core statistics in the figure, we can see: Median: The median density of evolutionary feature points is approximately 0.61, representing the central trend of the sample data and reflecting the typical density level of evolutionary feature points at the current stage. Interquartile Range (IQR): The lower quartile (Q1) is approximately 0.53, the upper quartile (Q3) is approximately 0.82, and the interquartile range is 0.29, indicating that about 50% of the density data is concentrated between 0.53 and 0.82, showing a certain degree of clustering around the median. Extreme Value Range: The minimum density value is approximately 0.32, and the maximum value is approximately 0.94. The overall data distribution range is relatively wide, reflecting the spatial heterogeneity and dynamic changes of the evolutionary feature points. Distribution pattern: The funnel-shaped box plot, wider at the top and narrower at the bottom, indicates higher data dispersion in high-density regions (above the upper quartile) and relatively concentrated data in low-density regions (below the lower quartile), suggesting that the distribution of evolutionary feature points in high-gradient regions is more random and volatile. Box plots provide quantitative evidence for evaluating the spatial clustering of evolutionary feature points in digital twin models and identifying anomalous density regions. They can be used to guide the retention of key regions and the removal of non-key regions in subsequent model lightweighting processes, while also providing prior information on density distribution for the generation of control fields in physically guided neural networks.

[0042] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A digital twin-based method for building implementation using twin modeling, characterized in that, include: Real-time monitoring data of multi-source physical fields during the construction process is obtained through on-site IoT sensor networks. The real-time monitoring data of multi-source physical fields includes structural strain time series, environmental temperature and humidity time series, and construction machinery posture series. Based on the real-time monitoring data of the multi-source physical field, a basic digital model mesh that is geometrically isomorphic to the physical building implementation site is constructed in the virtual space. The vertices of the basic digital model mesh are endowed with physical field properties derived from the real-time monitoring data of the multi-source physical field, forming a basic twin that carries physical properties. Multiphysics coupling simulation is performed on the basic twin to solve the dynamic distribution of stress field, temperature field and displacement field inside the basic twin caused by the difference of physical field properties, and generate a twin simulation field describing the physical field coupling state. Evolution feature points that satisfy preset evolution characteristics are extracted from the twin simulation field. The evolution feature points are composed of the spatial positions in the twin simulation field where the stress gradient, temperature gradient or displacement gradient exceeds the corresponding evolution threshold. Evolutionary feature points and their neighborhood data are input into a pre-trained physical guidance neural network. The generator of the physical guidance neural network generates a virtual control field that makes the twin simulation field tend to be in equilibrium based on the physical field properties and gradient information of the evolutionary feature points.

2. The digital twin-based building implementation twin modeling method according to claim 1, characterized in that, The acquisition of multi-source physical field real-time monitoring data during the building construction process through on-site IoT sensor networks includes: An array of strain sensors deployed at key parts of a building structure periodically collects micro-strain data of the structure, forming a time series sequence of structural strain. Environmental sensor nodes deployed in the construction area collect air temperature and relative humidity data to form a time series of environmental temperature and humidity. Positioning and attitude sensors attached to large construction machinery collect the spatial coordinates and attitude angle data of the machinery to form a sequence of the construction machinery's pose. The time series of structural strain, environmental temperature and humidity, and construction machinery pose are processed by timestamp alignment, outlier removal, and missing data interpolation to form a multi-source physical field real-time monitoring data set with a unified time reference.

3. The digital twin-based building implementation twin modeling method according to claim 2, characterized in that, The construction of a basic digital model mesh in virtual space, based on the real-time monitoring data of the multi-source physical fields and geometrically isomorphic to the physical building implementation site, includes: Import the design geometry from the building information model, discretize the design geometry into a mesh, and generate a basic digital model mesh composed of triangular facets or tetrahedral units. The current value of the structural strain time series in the real-time monitoring data of the multi-source physical field is assigned to the strain attribute of the corresponding grid vertex in the basic digital model mesh. The current value of the environmental temperature and humidity time series in the real-time monitoring data of the multi-source physical field is assigned to the temperature and humidity attributes of the corresponding spatial region grid vertex in the basic digital model grid. The current value of the construction machinery pose sequence is converted into the position and attitude of the virtual construction machinery model and placed in the corresponding spatial coordinates of the basic digital model grid, thus completing the construction of the basic twin that carries physical attributes.

4. The digital twin-based building implementation twin modeling method according to claim 3, characterized in that, The multiphysics coupled simulation of the basic twin, which solves for the dynamic distribution of stress, temperature, and displacement fields within the basic twin caused by differences in physical field properties, includes: Based on the strain and temperature / humidity properties of the vertices of the basic twin mesh, the stress constitutive relation and thermal expansion coefficient inside the mesh element are calculated. Establish finite element control equations that take into account the temperature-stress coupling effect; Apply mechanical load boundary conditions calculated from the posture sequence of construction machinery and heat flow boundary conditions determined from the time sequence of ambient temperature and humidity; Solving the finite element control equations yields the stress tensor, temperature scalar, and displacement vector at each mesh vertex of the basic twin, which together constitute the twin simulation field describing the coupling state of the physical fields.

5. The digital twin-based building implementation twin modeling method according to claim 4, characterized in that, The extraction of evolutionary feature points that satisfy preset evolutionary characteristics from the twin simulation field includes: Traverse the stress field data of all grid vertices in the twin simulation field and calculate the stress gradient magnitude at each grid vertex; The stress gradient magnitude is compared with a preset stress evolution threshold, and all mesh vertices whose stress gradient magnitude exceeds the stress evolution threshold are marked as potential evolution feature points. Spatial clustering is performed on the marked potential evolutionary feature points, and the grid vertex with the largest stress gradient magnitude is selected from each cluster as the representative point. The representative point is the evolutionary feature point that satisfies the preset evolutionary characteristics. Meanwhile, based on temperature field and displacement field data, the same gradient and threshold comparison method is used to extract temperature evolution feature points and displacement evolution feature points.

6. The digital twin-based building implementation twin modeling method according to claim 5, characterized in that, The process of inputting evolutionary feature points and their neighborhood data into a pre-trained physical guidance neural network, wherein the generator of the physical guidance neural network generates a virtual control field that makes the twin simulation field tend to be in equilibrium based on the physical field properties and gradient information of the evolutionary feature points, includes: The spatial coordinates of the evolution feature points, the stress tensor, temperature scalar, displacement vector of the evolution feature points, and the corresponding stress gradient, temperature gradient, and displacement gradient are used as the input feature vectors of the physical-guided neural network. The generator of the physical-guided neural network encodes and performs nonlinear transformation on the input feature vector and outputs a set of control parameters, including virtual stress compensation and virtual heat flux compensation applied to the vertices of the neighborhood grid of the evolution feature point. The control parameters are Gaussian diffused in the neighborhood of the evolutionary feature point to generate a spatially continuous virtual control field that covers part of the basic twin. The virtual control field is designed to counteract the excessively high physical field gradient at the evolutionary feature point.

7. The digital twin-based building implementation twin modeling method according to claim 6, characterized in that, Also includes: The virtual control field is mapped inversely to the basic digital model mesh, and the vertex physical field properties of the basic digital model mesh are iteratively corrected. The corrected basic digital model mesh constitutes an iteratively updated digital twin. In the virtual space, the digital twin that drives the iterative updates performs multi-physics coupled simulation for the next time step, and the newly generated twin simulation field is compared with the physical field state calculated from real-time monitoring data by residual comparison. When the residual is less than the preset tolerance, it is determined that the iteratively updated digital twin is consistent with the state of the physical building implementation site, and the iteratively updated digital twin at this time is marked as a converged real-time digital twin model. The converged real-time digital twin model is subjected to model lightweighting, and mesh vertices and physical attributes whose influence on the evolution of the physical field is lower than the contribution threshold are removed to generate a lightweight real-time digital twin model. The lightweight real-time digital twin model is pushed to an interactive visualization platform and combined with construction schedule data to generate a twin status report for construction management. The process of back-mapping the virtual control field to the basic digital model mesh and iteratively correcting the vertex physical field properties of the basic digital model mesh includes: The virtual stress compensation and virtual heat flux compensation corresponding to each spatial location in the virtual control field are superimposed on the current stress and temperature attributes of the corresponding grid vertices of the basic digital model grid. For the basic digital model mesh superimposed with the virtual control field, the displacement properties of its mesh vertices are recalculated so that the physical field distribution satisfies the new equilibrium relationship. The basic digital model mesh, after all attribute updates are completed, constitutes the digital twin after one iteration update.

8. The digital twin-based building implementation twin modeling method according to claim 7, characterized in that, The process of driving the iteratively updated digital twin in virtual space to perform multiphysics coupled simulation for the next time step involves comparing the residuals of the newly generated twin simulation field with the physical field state calculated from real-time monitoring data, including: Acquire real-time monitoring data of multi-source physical fields at the next sampling time; Based on the monitoring data at the next sampling time, the theoretical stress state and theoretical temperature state at the key locations of the building implementation site at the next sampling time are calculated by interpolation or mechanism model. The digital twin that drives the iterative update performs multiphysics coupling simulation for the next time step, and obtains the twin simulation field for the next time step; Calculate the difference between the state values ​​of the twin simulation field at key locations and the theoretical state values, and take the root mean square of the differences at all key locations as the residual.

9. The digital twin-based building implementation twin modeling method according to claim 8, characterized in that, The process of lightweighting the converged real-time digital twin model includes: Calculate the contribution of each grid vertex in the converged real-time digital twin model to the overall stress field and temperature field distribution. The contribution is defined by the change in field distribution caused by removing the grid vertex. Set a contribution threshold, identify and mark all grid vertices whose contribution is below the contribution threshold; The marked mesh vertices and their associated mesh cells are merged and removed. At the same time, the physical properties on the retained mesh vertices are resampled and smoothed to generate a lightweight real-time digital twin model that balances geometric complexity and physical property accuracy. The process involves pushing a lightweight, real-time digital twin model to an interactive visualization platform and combining it with construction schedule data to generate a twin status report for construction management, including: The geometric and physical property data of the lightweight real-time digital twin model are streamed to an interactive visualization platform for 3D rendering; Synchronize the current construction schedule from the construction management database and identify the current construction tasks and target components in the plan; In the interactive visualization platform, the lightweight real-time digital twin model area related to the current construction task is highlighted, and the key physical field data of the lightweight real-time digital twin model area is overlaid. The system automatically extracts key physical field data statistics of the highlighted area, deviations from the design allowable values, and distribution of evolution feature points, and generates a formatted twin status report document containing text, data tables, and key indicator illustrations.

10. A digital twin-based building implementation twin modeling system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the digital twin-based building implementation twin modeling method as described in any one of claims 1 to 9.

Images