A visual digital twin traceability system and method based on single point data

Abstract

The application discloses a kind of visualization digital twin tracing systems and methods based on single point data, belong to digital twin tracing technical field.The application obtains engineering BIM model, split definition minimum tracing unit into tracing single point, complete single point internal grid division, local and global coordinate system mapping, constructs the topological model containing process, space, structure association.Data base is built by collecting single point full life cycle multi-source tracing data, based on topological model, data is hung to BIM virtual component, and digital twin body is constructed.When quality anomaly occurs, abnormal image preprocessing is collected, abnormal data is identified by pre-training Mask R-CNN model, and the abnormal global coordinate interval is calculated by feature registration.Solve, through topological index matching target tracing single point and corresponding grid unit.Based on topological model, search the spatial, structure associated adjacent components of abnormal single point, determine the influence degree by classification, and delimit the overall structure influence range and risk level.

Description

Technical Field

[0001] This invention relates to the field of digital twin traceability technology, specifically a visualized digital twin traceability system and method based on single-point data. Background Technology

[0002] As my country's construction industry undergoes a profound transformation towards industrialization, digitalization, and intelligence, the scale and technological complexity of large and complex construction projects continue to increase. As a core link in engineering quality management, safety risk control, and the implementation of lifelong quality responsibility, engineering quality traceability directly determines the quality and safety level of the entire life cycle of the project through its traceability accuracy, response efficiency, and control capabilities.

[0003] Current engineering traceability models generally use regions, time periods, and batches as core traceability units, floors and construction sections as spatial units, months and weeks as time units, and inspection batches as acceptance units for traceability management. This approach fails to extend traceability granularity to individual engineering components with independent functions and responsibilities, and cannot cover specific work points within those components. Construction, acceptance, personnel, image, and testing data throughout the entire project lifecycle are scattered across multiple independent media such as construction management systems, document management platforms, supervision ledgers, and subcontractor documents. Data is not strongly linked to engineering components, resulting in widespread issues of data gaps, missing data, and inconsistencies. Existing technologies can only perform qualitative analysis based on overall component defects, unable to match specific defect coordinates and severity for refined classification and judgment, easily leading to overestimation of risk or omission of key impact areas. Summary of the Invention

[0004] The purpose of this invention is to provide a visualized digital twin traceability system and method based on single-point data to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] Firstly, this application provides a visual digital twin traceability method based on single-point data, comprising the following steps:

[0007] Obtain the building information model, break down the smallest physical unit within the entire lifecycle of the project, and define it as a traceability point; divide the internal coordinate grid of the physical component corresponding to each traceability point, establish the mapping relationship between the local relative coordinate system of the component and the global coordinate system of the project; and construct the topological structure model of the traceability points of the entire project.

[0008] Collect multi-source traceability data from a single point throughout its entire lifecycle and store it in a traceability database; based on a topology model, mount the traceability database onto the corresponding virtual components to obtain a digital twin;

[0009] When a quality anomaly occurs in a physical project, image data of the abnormal area is collected and preprocessed; a pre-trained engineering quality defect detection model is used to infer and identify the image data and output abnormal data; the abnormal data is registered with the digital twin to calculate the global coordinate interval corresponding to the abnormal area; based on the topology model, the target traceability point and the corresponding target grid cell belonging to the global coordinate interval are matched through the topology index to obtain matching data;

[0010] Based on the topological model and matching data, we retrieve adjacent single points that have spatial and structural associations with the anomalous single point, and analyze the impact range of the anomalous single point on adjacent components and the overall structure.

[0011] In conjunction with the first aspect, in the first embodiment of the first aspect of this application, the step of acquiring the building information model and breaking down the smallest physical unit within the entire lifecycle of the project, defined as tracing a single point, includes:

[0012] Obtain the building information model of the target project and perform pre-compliance verification of the model; set the split boundary rules, specifically: the minimum physical entity with independent structural function, capable of independently completing the entire construction process, capable of independently issuing quality acceptance conclusions, and capable of independently binding the operators and acceptance personnel shall be the split limit; it is prohibited to disassemble the integral load-bearing structural components that cannot be split; it is prohibited to merge multiple physical entities with independent processes into the same unit.

[0013] Based on Building Information Modeling (BIM), a top-down, layer-by-layer decomposition is performed according to professional level, system level, engineering zone level, floor level, and component type level to output an initial component set. For each component in the initial component set, a filtering process is performed based on the decomposition boundary rules to obtain a candidate minimum physical unit set. For each candidate unit in the candidate minimum physical unit set, it is verified whether the unit has a complete, closed, and unambiguous geometric boundary in three-dimensional space, specifically, that it has no spatial overlap, no geometric nesting, and no boundary intersection with other candidate units. It is also verified that the unit corresponds to an independent construction operation procedure, an independent quality acceptance batch, and is independently bound to the operators and inspectors. The minimum physical unit set that passes all verifications is output and defined as a traceability point.

[0014] In conjunction with the first aspect, in the second embodiment of the first aspect of this application, the step of dividing the physical component corresponding to each traceability point into an internal coordinate grid and establishing a mapping relationship between the component's local relative coordinate system and the project's global coordinate system includes:

[0015] The coding fields are defined in a fixed logical order from top to bottom. Each field corresponds to a unique dimension of project management. Each field has a preset fixed length and coding dictionary. Each traceability point is assigned a code.

[0016] For each traceable point, the geometric information of the corresponding physical component is extracted from the Building Information Model (BIM), including the component's closed contour surface, minimum bounding cube, key feature point set, and entity boundary range. This information is then standardized to output a 3D solid model. Based on the 3D solid model, the minimum bounding cube is used as the mesh to divide the reference space. According to a preset basic granularity standard, the reference space is subjected to 3D orthogonal meshing to generate a 3D mesh unit set. Boundary clipping is performed on the 3D mesh unit set, removing mesh units completely outside the component entity boundary range. For mesh units intersecting the component entity boundary, clipping is performed along the entity boundary.

[0017] Using the pre-defined origin of the building information model as the coordinate origin, and the north-south direction of the project as the X-axis, the east-west direction as the Y-axis, and the elevation direction as the Z-axis, a right-handed Cartesian global coordinate system is established. The original spatial positions of all components in the entire project are defined based on the global coordinate system. Using the pre-defined reference feature points of the components as the local coordinate origin, and the length direction of the main body of the component as the local X-axis, the width direction as the local Y-axis, and the height direction as the local Z-axis, a right-handed Cartesian local relative coordinate system is established. The mapping relationship between the local relative coordinate system of the components and the global coordinate system of the project is established.

[0018] In conjunction with the first aspect, in the third embodiment of the first aspect of this application, the construction of the topology model for tracing a single point throughout the entire project includes:

[0019] Each traceability point is defined as a unique topological node, and the node identifier adopts a single-point code; basic attributes, spatial attributes, and control attributes are attached to the node; the process is decomposed into single-point atomic processes, and based on the immediate predecessor and immediate successor constraints, unidirectional upstream and downstream edges of the process are constructed with the immediate predecessor node as the starting point and the immediate successor node as the ending point; the force transmission path is extracted from the 3D solid model, and directed structural force-related edges and undirected structural indirect-related edges are constructed; undirected spatial adjacency edges are constructed based on the intersection of global coordinate intervals; and nodes and edges are integrated to form a topological structure model.

[0020] In conjunction with the first aspect, in the fourth embodiment of the first aspect of this application, the step of mounting the traceability database onto the corresponding virtual component based on the topology model to obtain a digital twin includes:

[0021] Using the traceability single-point code as the association key, a corresponding mapping relationship is established between the virtual components of the building information model, the nodes of the topology model, and the traceability single point, forming a mounting benchmark. The multi-source traceability data of the single point throughout its entire life cycle in the traceability database is structurally split according to the traceability single-point code to generate a node-level traceability dataset corresponding to the topology node, and the internal grid coordinate mapping data of the single point is synchronously associated. The node system, edge system, and attributes of the topology model are bound to the corresponding mapped virtual components. The node-level traceability dataset is mounted to the topology node attribute library bound to the virtual component, establishing a two-way real-time synchronization mechanism between the traceability database and the digital twin data.

[0022] In conjunction with the first aspect, in the fifth embodiment of the first aspect of this application, the pre-trained engineering quality defect detection model performs inference and identification on image data and outputs abnormal data, including:

[0023] The Mask R-CNN instance segmentation algorithm is adopted as the basic algorithm framework for engineering quality defect detection. An image dataset covering all types of components, materials, and categories of quality defects in building engineering is collected. The defect type, instance contour mask, and pixel-level coordinate information of each frame are labeled. The dataset is divided into training, validation, and test sets according to a preset ratio, and standardized preprocessing and multi-scene data augmentation are performed. A Mask R-CNN dual-branch network structure is built using ResNet-FPN as the backbone feature extraction network. The detection branch is responsible for defect classification and bounding box regression, while the mask segmentation branch is responsible for instance-level contour pixel segmentation. Transfer learning is performed based on ImageNet pre-trained weights. The model is trained using a benchmark dataset, and the loss function and model parameters are optimized through cross-validation to eliminate overfitting risks. After the accuracy on the test set reaches the target, the pre-trained main model is solidified. Inference and recognition are performed on the image data, outputting abnormal data, including abnormal type, abnormal region contour, and abnormal region pixel coordinates.

[0024] In conjunction with the first aspect, in the sixth embodiment of the first aspect of this application, the step of performing feature registration between the abnormal data and the digital twin, and calculating the global coordinate interval corresponding to the abnormal region, includes:

[0025] From the image data, stable rigid geometric feature points of the components are extracted to generate a two-dimensional feature point set; from the digital twin, corresponding three-dimensional feature points with the same name as the two-dimensional feature points in the two-dimensional feature point set are extracted to generate a three-dimensional feature point set; based on the mathematical registration logic of rigid body transformation, the two-dimensional feature point set and the three-dimensional feature point set are matched one by one, and point pairs with incorrect matching and excessive error are eliminated through iterative optimization to obtain a set of corresponding feature point pairs;

[0026] Specifically, for mathematical registration, priority is given to selecting globally unique and unambiguous rigid reference feature points from two feature point sets, including component corner endpoints, axis intersections, and pre-embedded fixed marker points. These feature points have unique corresponding topological attributes in the 2D image and 3D model. Based on the global topological location and exclusive neighborhood attributes of the feature points, one-to-one matching of the reference feature points is completed, establishing at least 3 sets of unambiguous initial pairs of corresponding points as the initial anchor points and registration references for full matching.

[0027] Using the initial anchor point pair as a reference, hierarchical initial matching of all feature points is carried out according to the rule of relative position invariance under rigid body transformation. In the two-dimensional feature point set, with the initial anchor point as a reference, the relative position and adjacency topology of each feature point to be matched relative to the anchor point are recorded. In the three-dimensional feature point set, according to the completely consistent relative position and adjacency topology, uniquely corresponding candidate three-dimensional feature points are selected, and a one-to-one initial matching relationship is established to form an initial matching point pair set.

[0028] Based on the rules of rigid body transformation, a full consistency check is performed on the initial set of matching point pairs. In the two-dimensional feature point set, two feature points are arbitrarily selected, and their relative positional and angular relationships are calculated. In the three-dimensional feature point set, two corresponding matching points are selected, and their relative positional and angular relationships are calculated. The consistency of the two sets of relationships is compared. If the deviation exceeds the preset allowable range, the point pair is determined to have an incorrect match and is marked as a point pair to be removed. After completing the pairwise consistency check of all point pairs, all marked incorrect point pairs are removed, forming a preliminary set of valid matching point pairs.

[0029] Based on the initial set of effective matching point pairs, the current rigid body transformation relationship is constructed, and iterative optimization and verification are carried out. Each point in the 3D feature point set is projected onto the 2D image plane according to the current rigid body transformation relationship to obtain the corresponding projected 2D point. The positional deviation between the projected point and the actual 2D feature point is compared. All point pairs are sorted according to the magnitude of the deviation, and point pairs with deviations exceeding the preset accuracy threshold are removed. The rigid body transformation relationship is reconstructed with the remaining point pairs, and the above process of projection verification, deviation sorting, and removal of out-of-tolerance points is repeated. The iteration continues until the deviations of all remaining point pairs meet the accuracy requirements and no new out-of-tolerance point pairs are generated.

[0030] Based on the geometric principles of perspective projection, using a set of corresponding feature point pairs as the solution benchmark and combining the device's intrinsic parameters, the pose parameters of the image acquisition device in the global coordinate system are calculated, including the device's three-dimensional spatial position and the lens's three-dimensional orientation in the global coordinate system. A reprojection error check is performed on the pose parameters by backprojecting all three-dimensional feature points to the image pixel coordinate system using the pose parameters. The deviation between the backprojected coordinates and the actual image feature point coordinates is compared. When the deviation exceeds a preset threshold, the process returns to the registration stage to optimize the matching point pairs until the error meets the accuracy requirements, thus determining the pose parameters.

[0031] Based on the pose parameters involved in the device, the pixel coordinate system is converted into the device imaging plane coordinate system, and then into the device's own coordinate system. Through rigid body transformation of the pose parameters, it is transformed into the engineering global coordinate system to generate a global three-dimensional coordinate point set of the abnormal region contour. Full-dimensional extreme value statistics are performed on the global three-dimensional coordinate point set to obtain the maximum and minimum values ​​of the point set in the X, Y, and Z axes of the engineering global coordinate system, forming the global coordinate interval corresponding to the abnormal region.

[0032] In conjunction with the first aspect, in the seventh embodiment of the first aspect of this application, the step of matching the target tracing point and the corresponding target mesh cell belonging to the global coordinate interval based on the topology model and using the topology index to obtain matching data includes:

[0033] Based on spatial topology indexing, the system filters out unique traceability node topology nodes whose global coordinate boundaries completely cover the coordinate range of the abnormal area through spatial inclusion relationships. These nodes are identified as target traceability nodes, and their codes and internal grid mapping data are output. Based on the global coordinate lookup table of the internal grid cells of the target traceability node, the system filters out grid cells that cover the range of the abnormal area through spatial intersection and inclusion determination logic. These grid cells are identified as target grid cells, and their numbers, coordinate boundaries, and associated data indexes are output. The output data is then integrated as matching data.

[0034] In conjunction with the first aspect, in the eighth embodiment of the first aspect of this application, the step of retrieving adjacent single points that have spatial and structural associations with the anomalous single point based on the topological structure model and matching data, and analyzing the impact range of the anomalous single point on adjacent components and the overall structure, includes:

[0035] Based on the topological model, all spatially adjacent points that have spatial boundary contact or coordinate interval adjacency with the anomalous point are retrieved through spatial adjacency edges. All structurally associated points that have direct and indirect force transmission relationships with the anomalous point are traversed and retrieved through structural force-related edges, forming a list of associated adjacent points. Based on this list, and considering the anomaly type, severity, adjacency level of the associated edges, and force transmission weight, the impact of the anomaly on each adjacent point component is graded. A full force path traversal is completed along the structural associated edges, and the force diffusion path of the anomaly is determined by combining this with the structural design safety threshold, thus defining the affected boundaries and risk levels of the overall structure.

[0036] Secondly, this application provides a visualized digital twin traceability system based on single-point data, including:

[0037] The topology model construction module includes: a single-point tracing acquisition unit to acquire the building information model and break down the smallest physical unit within the entire lifecycle of the project, defining it as a single-point tracing; a mapping relationship establishment unit to divide the internal coordinate grid of the physical component corresponding to each single-point tracing and establish the mapping relationship between the local relative coordinate system of the component and the global coordinate system of the project; and a topology model construction unit to construct the topology model of all single-point tracing in the entire project.

[0038] The digital twin construction module includes: a multi-source traceability data acquisition unit that collects multi-source traceability data from a single point throughout its entire lifecycle and stores it in a traceability database; and a digital twin construction unit that, based on a topology model, mounts the traceability database onto the corresponding virtual components to obtain a digital twin.

[0039] The data matching module includes: an image data acquisition unit that acquires image data of the abnormal area when quality anomalies occur in the physical project and performs preprocessing; an abnormal data output unit that pre-trains an engineering quality defect detection model, performs inference and identification on the image data, and outputs abnormal data; a global coordinate interval calculation unit that performs feature registration between the abnormal data and the digital twin, and calculates the global coordinate interval corresponding to the abnormal area; and a data matching unit that, based on a topology model, matches the target traceability point and the corresponding target grid cell belonging to the global coordinate interval through a topology index to obtain matching data.

[0040] Impact Range Analysis Module: Includes: The impact range analysis unit, based on the topological structure model and matching data, retrieves adjacent single points that have spatial and structural associations with the abnormal single point, and analyzes the impact range of the abnormality on adjacent components and the overall structure.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] 1. This invention defines traceability points with independent structure, process, acceptance, and responsibility attributes, thereby reducing the traceability granularity from the traditional regional batch level to the individual component level. Furthermore, by dividing the components into internal coordinate grids, the traceability accuracy is further improved to the component internal coordinate level, achieving a strong binding between each work point and the work and acceptance personnel.

[0043] 2. This invention constructs a traceability single-point topology model that includes multi-dimensional associations of processes, space, and structure, and establishes a strong constraint mechanism between upstream and downstream processes. The completion of data collection and acceptance of the preceding process are prerequisites for the commencement of the subsequent process, thereby realizing closed-loop collection and full-link connectivity of traceability data. It can achieve bidirectional traceability of the entire life cycle from raw material entry, construction, acceptance to operation and maintenance.

[0044] 3. This invention combines the precise global coordinate range of defects and the target grid cell matching data to match the force weight and adjacency level of the associated edges, thereby achieving a precise determination of the degree of influence and avoiding misjudgment of risks from the root. Attached Figure Description

[0045] Figure 1 This is a schematic diagram illustrating the steps of a visual digital twin traceability method based on single-point data according to the present invention;

[0046] Figure 2 This is a system structure diagram of a visual digital twin traceability system based on single-point data according to the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0048] Example: Figures 1-2 As shown, the present invention provides a technical solution:

[0049] like Figure 1 As shown, this application provides a visual digital twin traceability method based on single-point data, including the following steps:

[0050] Step S100: Obtain the building information model, break down the smallest physical unit within the entire life cycle of the project, and define it as a traceability point; divide the internal coordinate grid of the physical component corresponding to each traceability point, establish the mapping relationship between the local relative coordinate system of the component and the global coordinate system of the project; construct the topological structure model of the traceability points of the entire project.

[0051] Specifically, obtain the building information model of the target project and perform pre-compliance verification of the model; set splitting boundary rules, specifically: the minimum physical entity with independent structural function, capable of independently completing the entire construction process, capable of independently issuing quality acceptance conclusions, and capable of independently binding the operators and acceptance personnel shall be the splitting lower limit; it is prohibited to disassemble the indivisible overall load-bearing structural components, and it is prohibited to merge multiple physical entities with independent processes into the same unit.

[0052] Based on Building Information Modeling (BIM), a top-down, layer-by-layer decomposition is performed according to professional level, system level, engineering zone level, floor level, and component type level to output an initial component set. For each component in the initial component set, a filtering process is performed based on the decomposition boundary rules to obtain a candidate minimum physical unit set. For each candidate unit in the candidate minimum physical unit set, it is verified whether the unit has a complete, closed, and unambiguous geometric boundary in three-dimensional space, specifically, that it has no spatial overlap, no geometric nesting, and no boundary intersection with other candidate units. It is also verified that the unit corresponds to an independent construction operation procedure, an independent quality acceptance batch, and is independently bound to the operators and inspectors. The minimum physical unit set that passes all verifications is output and defined as a traceability point.

[0053] Furthermore, a fixed logical order of coding fields is defined from top to bottom. Each field corresponds to a unique dimension of project management. Each field has a preset fixed length and coding dictionary, and a code is assigned to each traceability point.

[0054] For each traceable point, the geometric information of the corresponding physical component is extracted from the Building Information Model (BIM), including the component's closed contour surface, minimum bounding cube, key feature point set, and entity boundary range. This information is then standardized to output a 3D solid model. Based on the 3D solid model, the minimum bounding cube is used as the mesh to divide the reference space. According to a preset basic granularity standard, the reference space is subjected to 3D orthogonal meshing to generate a 3D mesh unit set. Boundary clipping is performed on the 3D mesh unit set, removing mesh units completely outside the component entity boundary range. For mesh units intersecting the component entity boundary, clipping is performed along the entity boundary.

[0055] Using the pre-defined origin of the building information model as the coordinate origin, and the north-south direction of the project as the X-axis, the east-west direction as the Y-axis, and the elevation direction as the Z-axis, a right-handed Cartesian global coordinate system is established. The original spatial positions of all components in the entire project are defined based on the global coordinate system. Using the pre-defined reference feature points of the components as the local coordinate origin, and the length direction of the main body of the component as the local X-axis, the width direction as the local Y-axis, and the height direction as the local Z-axis, a right-handed Cartesian local relative coordinate system is established. The mapping relationship between the local relative coordinate system of the components and the global coordinate system of the project is established.

[0056] Furthermore, each traceability point is defined as a unique topological node, and the node identifier adopts a single-point code; basic attributes, spatial attributes, and control attributes are attached to the nodes; the process is decomposed into single-point atomic processes, and based on the immediate predecessor and immediate successor constraints, unidirectional upstream and downstream edges of the process are constructed with the immediate predecessor node as the starting point and the immediate successor node as the ending point; the force transmission path is extracted from the 3D solid model to construct directed structural force-related edges and undirected structural indirect-related edges; undirected spatial adjacency edges are constructed based on the intersection of global coordinate intervals; and nodes and edges are integrated to form a topological structure model.

[0057] In one specific embodiment, a 32-story shear wall structure high-rise residential project in a provincial capital city was used as the implementation object. The project has 2 underground floors and 30 floors above ground, with a total construction area of ​​32,000 square meters. The implementation process and corresponding experimental data are as follows:

[0058] First, a Revit building information model with LOD400 accuracy was obtained for all disciplines of the project. Pre-compliance verification of the model was performed, eliminating 12 redundant components and correcting 8 spatially conflicting components, resulting in a standardized and compliant model. After setting the splitting boundary rules, the model was decomposed layer by layer from top to bottom according to discipline, system, project zone, floor, and component type, outputting 12,860 initial components. After filtering by the splitting rules, 11,245 candidate minimum physical units were obtained. Then, dual verification of geometric boundaries and process independence was performed, eliminating 32 boundary overlapping units and 113 non-independent process units. Finally, 11,100 minimum physical units that passed full verification were output, formally defined as traceable single points. Typical single points include independent components such as shear wall Q-12-08, frame column KZ-09-12, and foundation CT-03-05.

[0059] Subsequently, a 13-digit hierarchical single-point coding rule was formulated, with the fields being the project identification code, zone code, professional code, component type code, floor code, and serial number, respectively. The code was assigned to all 11,100 traceable single points. After triple verification of format, validity, and uniqueness, the duplicate code rate and omission rate were both 0, and the format compliance rate was 100%. For a single typical traceability point—the 12th-floor frame column KZ-09-12 (section 600mm×600mm, height 3000mm)—its closed contour surface, minimum enclosing cube, and other geometric information were extracted and standardized. A three-dimensional orthogonal mesh was generated using a 100mm×100mm×100mm granularity, producing 1080 initial mesh elements. After boundary trimming, all effective mesh elements were obtained. A secondary meshing with a 50mm granularity was performed on the stress-reinforced zone at the column end, generating 1728 refined mesh elements. Simultaneously, a global right-handed Cartesian coordinate system and a local relative coordinate system for the component were established, completing the dual-coordinate system mapping. After accuracy verification, the maximum coordinate transformation deviation was less than 0.5mm, meeting the engineering accuracy requirements.

[0060] Finally, the 11,100 traceable single points were defined as 11,100 unique topological nodes, with single-point codes as node identifiers. Each node was assigned three types of attributes: basic, spatial, and control. The entire process was broken down into single-point atomic processes, and 42,300 unidirectional upstream and downstream edges were constructed based on immediate predecessor and immediate successor constraints. Force transmission paths were extracted from the structural model, and 18,600 directed structural force-related edges and 3,200 undirected structural indirect-related edges were constructed. 27,800 undirected spatial adjacency edges were constructed based on the intersection of global coordinate intervals. The nodes and all edges were integrated to form a complete topological structure model. After verification, the reachability of all node links was 100%, with no circular dependencies and no logical chain breaks, thus completing the full implementation of this step.

[0061] Step S200: Collect multi-source traceability data for the entire lifecycle of a single point and store it in the traceability database; based on the topology model, mount the traceability database onto the corresponding virtual component to obtain a digital twin;

[0062] Specifically, using the traceability single-point code as the association key, a corresponding mapping relationship is established between the virtual components of the building information model, the nodes of the topology model, and the traceability single point, forming a mounting benchmark. The multi-source traceability data of the single point throughout its entire life cycle in the traceability database is structurally split according to the traceability single-point code to generate a node-level traceability dataset corresponding to the topology node, and the internal grid coordinate mapping data of the single point is synchronously associated. The node system, edge system, and attributes of the topology model are bound to the corresponding mapped virtual components. The node-level traceability dataset is mounted to the topology node attribute library bound to the virtual component, establishing a two-way real-time synchronization mechanism between the traceability database and the digital twin data.

[0063] In one specific embodiment, continuing with the aforementioned 32-story shear wall structure high-rise residential project, the implementation was carried out based on the 11,100 completed traceability points, corresponding topological structure models, and LOD400 precision BIM models. The specific process and experimental data are as follows:

[0064] First, using a 13-bit globally unique single-point code as the unique association key, a one-to-one mapping relationship is established between BIM model virtual components, topology model nodes, and entity traceability single points. After full verification, the mapping matching accuracy is 100%, with no mismatched or missing nodes, forming a stable mounting benchmark.

[0065] Subsequently, for all 11,100 traceability points, multi-source traceability data throughout the entire lifecycle was collected, covering six categories: component design parameters, raw material arrival inspection reports, construction worker identities and process parameters, supervision and acceptance records, full-process construction images, and third-party testing data. All data were strongly bound to the corresponding single-point code and synchronously associated with the internal grid coordinate mapping data of the single point, realizing precise binding of the operation and acceptance responsibility data of each grid unit. A total of 126,000 valid structured traceability data were collected, with an average of 11.3 valid data bound per traceability point, a data integrity rate of 99.8%, no broken links or invalid data, and all data was standardized and stored in a distributed traceability database.

[0066] Then, the entire traceability database was split into node-level traceability datasets corresponding to the topology nodes one-to-one according to the single-point coding. The node system of the topology model, 91,900 associated edges in 4 categories, and all attributes were strongly bound to the corresponding virtual components. The node-level traceability datasets were then accurately mounted to the topology node attribute library bound to the virtual components. A two-way real-time synchronization mechanism between the traceability database and the digital twin was established. The average latency of synchronizing the data collected on site to the twin was less than 2 seconds, and the data synchronization accuracy was 100%.

[0067] Finally, an interactive and traceable engineering digital twin was constructed. After testing, the average response time for single-point code retrieval of the twin was less than 0.3 seconds, the response time for retrieving full traceability data of any virtual component was less than 0.5 seconds, and the accuracy of matching data corresponding to the component grid-level coordinates was 100%, thus completing the entire process of this step.

[0068] Step S300: When a quality anomaly occurs in the physical project, image data of the abnormal area is collected and preprocessed; a pre-trained engineering quality defect detection model is used to infer and identify the image data and output abnormal data; the abnormal data is registered with the digital twin to calculate the global coordinate interval corresponding to the abnormal area; based on the topology model, the target traceability point and the corresponding target grid cell belonging to the global coordinate interval are matched through the topology index to obtain matching data.

[0069] Specifically, the Mask R-CNN instance segmentation algorithm is adopted as the basic algorithm framework for engineering quality defect detection. An image dataset covering all types of components, materials, and categories of quality defects in building engineering is collected. The defect type, instance contour mask, and pixel-level coordinate information of each frame are labeled. The dataset is divided into training, validation, and test sets according to a preset ratio, and standardized preprocessing and multi-scene data augmentation are performed. A Mask R-CNN dual-branch network structure is built using ResNet-FPN as the backbone feature extraction network. The detection branch is responsible for defect classification and bounding box regression, while the mask segmentation branch is responsible for instance-level contour pixel segmentation. Transfer learning is performed based on ImageNet pre-trained weights. Model training is completed using a benchmark dataset. The loss function and model parameters are optimized through cross-validation to eliminate overfitting risks. After the accuracy on the test set reaches the target, the pre-trained main model is solidified. Inference and recognition are performed on the image data, outputting abnormal data, including abnormal type, abnormal region contour, and abnormal region pixel coordinates.

[0070] Furthermore, from the image data, stable rigid geometric feature points of the components are extracted to generate a two-dimensional feature point set; from the digital twin, corresponding three-dimensional feature points with the same name as the two-dimensional feature points in the two-dimensional feature point set are extracted to generate a three-dimensional feature point set; based on the mathematical registration logic of rigid body transformation, the two-dimensional feature point set and the three-dimensional feature point set are matched one by one, and point pairs with incorrect matching and excessive error are eliminated through iterative optimization to obtain a set of corresponding feature point pairs;

[0071] Based on the geometric principles of perspective projection, using a set of corresponding feature point pairs as the solution benchmark and combining the device's intrinsic parameters, the pose parameters of the image acquisition device in the global coordinate system are calculated, including the device's three-dimensional spatial position and the lens's three-dimensional orientation in the global coordinate system. A reprojection error check is performed on the pose parameters by backprojecting all three-dimensional feature points to the image pixel coordinate system using the pose parameters. The deviation between the backprojected coordinates and the actual image feature point coordinates is compared. When the deviation exceeds a preset threshold, the process returns to the registration stage to optimize the matching point pairs until the error meets the accuracy requirements, thus determining the pose parameters.

[0072] Based on the pose parameters involved in the device, the pixel coordinate system is converted into the device imaging plane coordinate system, and then into the device's own coordinate system. Through rigid body transformation of the pose parameters, it is transformed into the engineering global coordinate system to generate a global three-dimensional coordinate point set of the abnormal region contour. Full-dimensional extreme value statistics are performed on the global three-dimensional coordinate point set to obtain the maximum and minimum values ​​of the point set in the X, Y, and Z axes of the engineering global coordinate system, forming the global coordinate interval corresponding to the abnormal region.

[0073] Furthermore, based on the spatial topology index, and through spatial inclusion relationships, the unique traceability point topology node whose global coordinate boundary completely covers the coordinate range of the abnormal area is selected as the target traceability point, and the code and internal grid mapping data of the target traceability point are output. Based on the global coordinate lookup table of the internal grid cells of the target traceability point, and through spatial intersection and inclusion determination logic, the grid cells covering the range of the abnormal area are selected as the target grid cells, and the number, coordinate boundary and associated data index of the target grid cells are output. The output data is integrated as matching data.

[0074] In one specific embodiment, during the inspection of the 12th floor of the project, a suspected honeycomb-like surface defect was found on the surface of the frame column KZ-09-12. Three high-definition images of the defect area were captured using a 4K industrial camera, and lens distortion correction, image denoising, and contrast enhancement were performed as standardized preprocessing to eliminate image interference.

[0075] In the pre-training stage of the engineering quality defect detection model, a high-definition image dataset of 150,000 images covering all types of components, materials, and categories of quality defects was collected. Each frame of the image was labeled with the defect type, instance contour mask, and pixel-level coordinate information. The dataset was divided into a training set of 120,000 images, a validation set of 19,500 images, and a test set of 10,500 images in an 8:1.3:0.7 ratio, and multi-scene data augmentation was performed. A Mask R-CNN dual-branch network was built using ResNet-FPN as the backbone feature extraction network. Transfer learning was performed based on ImageNet pre-trained weights. The loss function and model parameters were optimized through cross-validation to eliminate overfitting risks. After validation on the test set, the model achieved an mAP of 92.3%, meeting the engineering accuracy requirements, and the pre-trained main model was then solidified. The pre-processed abnormal images were input into the model for inference, and the output abnormality types were honeycomb-like textures, abnormal region contour pixel coordinate sets, and bounding box pixel coordinates.

[0076] In the feature registration and coordinate calculation stage, eight stable rigid geometric feature points, including the four corners of the column, the two intersection points of the axes, and the two pre-embedded marker points, are extracted from the abnormal image to generate a two-dimensional feature point set. Eight corresponding three-dimensional feature points are extracted from the digital twin to generate a three-dimensional feature point set. Based on the rigid body transformation mathematical registration logic, each feature point is matched one by one. After iterative optimization, one point pair with excessive error is removed, resulting in seven pairs of corresponding feature points. Using this point pair as the calculation benchmark, the pose parameters of the image acquisition device are calculated in combination with the device's intrinsic parameters. Reprojection error verification is performed, with an average reprojection error of 0.3 mm and a maximum error of 0.4 mm. After meeting the accuracy requirements, the pose parameters are solidified. Through full-link coordinate transformation, a global three-dimensional coordinate point set of the abnormal area contour is generated. After full-dimensional extreme value statistics, the global coordinate interval of the abnormal area is obtained as follows: X-axis 1234.567-1234.623 m, Y-axis 567.890-567.946 m, and Z-axis 89.123-89.215 m.

[0077] In the topology index matching stage, based on the spatial topology index and spatial inclusion relationship determination, with an average response time of 0.2 seconds, the unique traceability single-point topology node whose global coordinate boundary completely covers the anomaly interval is selected as the target traceability single point, coded as 1234567890123 (corresponding to frame column KZ-09-12). Based on the global coordinate lookup table of the internal grid cells of this single point, through spatial intersection and inclusion determination, three encrypted grid cells covering the core range of the anomaly are selected, numbered KZ-09-12-G-087, KZ-09-12-G-088, and KZ-09-12-G-095 respectively. The target traceability single point code, internal grid mapping data, target grid cell number and coordinate boundary are integrated to form standardized matching data, completing the entire process of this step.

[0078] Step S400: Based on the topology model and matching data, retrieve adjacent single points that have spatial and structural associations with the abnormal single point, and analyze the impact range of the abnormality on adjacent components and the overall structure.

[0079] Specifically, based on the topological model, all spatially adjacent points that have spatial boundary contact or coordinate interval adjacency with the anomalous point are retrieved through spatial adjacency edges; all structurally associated points that have direct and indirect force transmission relationships with the anomalous point are traversed and retrieved through structural force-related edges, forming a list of associated adjacent points; based on the list of associated adjacent points, combined with the anomaly type, severity, adjacency level of the associated edges, and force transmission weight, the impact degree of the anomaly on each adjacent point component is graded; the entire force path is traversed along the structural associated edges, and combined with the structural design safety threshold, the force diffusion path of the anomaly is determined, and the affected boundaries and risk levels of the overall structure are delineated.

[0080] In one specific embodiment, a search for associated adjacent single points is conducted based on a topological model. Through spatial adjacency edges, four spatially adjacent single points are retrieved: frame beams L-12-08 and L-12-09 on the same floor that directly contact the spatial boundaries of column body, top, and bottom of column KZ-09-12; and adjacent shear walls Q-12-07 and Q-12-09 on the same floor. Through structural stress-related edges, five structurally related single points are retrieved: a coaxial frame column KZ-11-12 on the 11th floor directly upstream; a coaxial frame column KZ-13-12 on the 13th floor directly downstream; coaxial floor slabs LB-12-12 and LB-13-12 on the 12th and 13th floors indirectly related to stress; and a distant shear wall Q-12-05 connected to the beam on the same floor. These are then integrated to form a list of associated adjacent single points containing seven points. The average search response time is 0.15 seconds, and no associated components are missed.

[0081] Subsequently, based on the anomaly type (honeycomb surface), severity (depth approximately 15mm, accounting for half the thickness of the column section protective layer, belonging to general quality defects), as well as the adjacency level and force transfer weight of the associated edges, the degree of impact was graded and determined: the direct upstream / downstream columns, beams in direct contact with each other on the same floor, and floor slabs on the same axis were determined to have "minor impact, requiring simultaneous observation for 3 days"; the adjacent shear walls on the same floor were determined to have "no direct impact, requiring no special treatment"; and the distant shear walls were determined to have "no impact". The grading and determination logic is completely consistent with the structural design rules.

[0082] Finally, the entire stress path was traversed along the structurally associated edges, from floor 11 KZ-11-12 to floor 12 KZ-09-12 and then to floor 13 KZ-13-12. Based on the structural design safety threshold (the axial compression ratio of this column is designed to be 0.65, and the current honeycomb surface has been preliminarily verified to have an axial compression ratio of 0.62, which is lower than the design value), it was determined that the stress diffusion path is limited to the vertical members of floors 11-13 and the horizontal floor slabs of floors 12-13 on the same axis. The boundaries of the overall structure affected were defined as X-axis 1234.200-1234.900m, Y-axis 567.500-568.200m, and Z-axis 86.000-92.000m. The risk level was "low risk, no overall structural reinforcement is required, but the target single-point defect needs to be repaired and the associated slightly affected members need to be observed simultaneously".

[0083] like Figure 2 As shown, this application provides a visualized digital twin traceability system based on single-point data, comprising:

[0084] The topology model construction module includes: a single-point tracing acquisition unit to acquire the building information model and break down the smallest physical unit within the entire lifecycle of the project, defining it as a single-point tracing; a mapping relationship establishment unit to divide the internal coordinate grid of the physical component corresponding to each single-point tracing and establish the mapping relationship between the local relative coordinate system of the component and the global coordinate system of the project; and a topology model construction unit to construct the topology model of all single-point tracing in the entire project.

[0085] The digital twin construction module includes: a multi-source traceability data acquisition unit that collects multi-source traceability data from a single point throughout its entire lifecycle and stores it in a traceability database; and a digital twin construction unit that, based on a topology model, mounts the traceability database onto the corresponding virtual components to obtain a digital twin.

[0086] The data matching module includes: an image data acquisition unit that acquires image data of the abnormal area when quality anomalies occur in the physical project and performs preprocessing; an abnormal data output unit that pre-trains an engineering quality defect detection model, performs inference and identification on the image data, and outputs abnormal data; a global coordinate interval calculation unit that performs feature registration between the abnormal data and the digital twin, and calculates the global coordinate interval corresponding to the abnormal area; and a data matching unit that, based on a topology model, matches the target traceability point and the corresponding target grid cell belonging to the global coordinate interval through a topology index to obtain matching data.

[0087] Impact Range Analysis Module: Includes: The impact range analysis unit, based on the topological structure model and matching data, retrieves adjacent single points that have spatial and structural associations with the abnormal single point, and analyzes the impact range of the abnormality on adjacent components and the overall structure.

[0088] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, 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 equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A visual digital twin traceability method based on single-point data, characterized in that, Includes the following steps: Obtain the building information model, break down the smallest physical unit within the entire lifecycle of the project, and define it as a traceability point; divide the internal coordinate grid of the physical component corresponding to each traceability point, establish the mapping relationship between the local relative coordinate system of the component and the global coordinate system of the project; and construct the topological structure model of the traceability points of the entire project. Collect multi-source traceability data from a single point throughout its entire lifecycle and store it in the traceability database; Based on the topological model, the traceability database is mounted onto the corresponding virtual components to obtain a digital twin; When a quality anomaly occurs in a physical project, image data of the abnormal area is collected and preprocessed; a pre-trained engineering quality defect detection model is used to infer and identify the image data and output the abnormal data; the abnormal data is then registered with the digital twin to calculate the global coordinate range corresponding to the abnormal area. Based on the topology model, the target trace point and the corresponding target grid cell are matched to the global coordinate range through the topology index to obtain the matching data. Based on the topological model and matching data, we retrieve adjacent single points that have spatial and structural associations with the anomalous single point, and analyze the impact range of the anomalous single point on adjacent components and the overall structure.

2. The visual digital twin traceability method based on single-point data according to claim 1, characterized in that, The acquisition of the building information model, which breaks down the smallest physical unit throughout the entire lifecycle of the project and defines it as a single point of traceability, includes: Obtain the building information model of the target project and perform pre-compliance verification of the model; set the split boundary rules, specifically: the minimum physical entity with independent structural function, capable of independently completing the entire construction process, capable of independently issuing quality acceptance conclusions, and capable of independently binding the operators and acceptance personnel shall be the split limit; it is prohibited to disassemble the integral load-bearing structural components that cannot be split; it is prohibited to merge multiple physical entities with independent processes into the same unit. Based on Building Information Modeling (BIM), a top-down, layer-by-layer decomposition is performed according to professional level, system level, engineering zone level, floor level, and component type level to output an initial component set. For each component in the initial component set, a filtering process is performed based on the decomposition boundary rules to obtain a candidate minimum physical unit set. For each candidate unit in the candidate minimum physical unit set, it is verified whether the unit has a complete, closed, and unambiguous geometric boundary in three-dimensional space, specifically, that it has no spatial overlap, no geometric nesting, and no boundary intersection with other candidate units. It is also verified that the unit corresponds to an independent construction operation procedure, an independent quality acceptance batch, and is independently bound to the operators and inspectors. The minimum physical unit set that passes all verifications is output and defined as a traceability point.

3. The visual digital twin traceability method based on single-point data according to claim 1, characterized in that, The step of dividing the physical component corresponding to each traceability point into an internal coordinate grid and establishing a mapping relationship between the component's local relative coordinate system and the project's global coordinate system includes: The coding fields are defined in a fixed logical order from top to bottom. Each field corresponds to a unique dimension of project management. Each field has a preset fixed length and coding dictionary. Each traceability point is assigned a code. For each traceable point, the geometric information of the corresponding physical component is extracted from the Building Information Model (BIM), including the component's closed contour surface, minimum bounding cube, key feature point set, and entity boundary range. This information is then standardized to output a 3D solid model. Based on the 3D solid model, the minimum bounding cube is used as the mesh to divide the reference space. According to a preset basic granularity standard, the reference space is subjected to 3D orthogonal meshing to generate a 3D mesh unit set. Boundary clipping is performed on the 3D mesh unit set, removing mesh units completely outside the component entity boundary range. For mesh units intersecting the component entity boundary, clipping is performed along the entity boundary. Using the pre-defined origin of the building information model as the coordinate origin, and the north-south direction of the project as the X-axis, the east-west direction as the Y-axis, and the elevation direction as the Z-axis, a right-handed Cartesian global coordinate system is established. The original spatial positions of all components in the entire project are defined based on the global coordinate system. Using the pre-defined reference feature points of the components as the local coordinate origin, and the length direction of the main body of the component as the local X-axis, the width direction as the local Y-axis, and the height direction as the local Z-axis, a right-handed Cartesian local relative coordinate system is established. The mapping relationship between the local relative coordinate system of the components and the global coordinate system of the project is established.

4. The visual digital twin traceability method based on single-point data according to claim 1, characterized in that, The construction of the topology model for tracing a single point in the entire project includes: Each traceability point is defined as a unique topological node, and the node identifier adopts a single-point code; basic attributes, spatial attributes, and control attributes are attached to the node; the process is decomposed into single-point atomic processes, and based on the immediate predecessor and immediate successor constraints, unidirectional upstream and downstream edges of the process are constructed with the immediate predecessor node as the starting point and the immediate successor node as the ending point; the force transmission path is extracted from the 3D solid model, and directed structural force-related edges and undirected structural indirect-related edges are constructed; undirected spatial adjacency edges are constructed based on the intersection of global coordinate intervals; and nodes and edges are integrated to form a topological structure model.

5. The visual digital twin traceability method based on single-point data according to claim 1, characterized in that, The process of mounting the traceability database onto the corresponding virtual components based on the topology model to obtain a digital twin includes: Using the traceability single-point code as the association key, a corresponding mapping relationship is established between the virtual components of the building information model, the nodes of the topology model, and the traceability single point, forming a mounting benchmark. The multi-source traceability data of the single point throughout its entire life cycle in the traceability database is structurally split according to the traceability single-point code to generate a node-level traceability dataset corresponding to the topology node, and the internal grid coordinate mapping data of the single point is synchronously associated. The node system, edge system, and attributes of the topology model are bound to the corresponding mapped virtual components. The node-level traceability dataset is mounted to the topology node attribute library bound to the virtual component, establishing a two-way real-time synchronization mechanism between the traceability database and the digital twin data.

6. The visual digital twin traceability method based on single-point data according to claim 1, characterized in that, The pre-trained engineering quality defect detection model performs inference and identification on image data and outputs abnormal data, including: The Mask R-CNN instance segmentation algorithm is adopted as the basic algorithm framework for engineering quality defect detection. An image dataset covering all types of components, materials, and categories of quality defects in building engineering is collected. The defect type, instance contour mask, and pixel-level coordinate information of each frame are labeled. The dataset is divided into training, validation, and test sets according to a preset ratio, and standardized preprocessing and multi-scene data augmentation are performed. A Mask R-CNN dual-branch network structure is built using ResNet-FPN as the backbone feature extraction network. The detection branch is responsible for defect classification and bounding box regression, while the mask segmentation branch is responsible for instance-level contour pixel segmentation. Transfer learning is performed based on ImageNet pre-trained weights. The model is trained using a benchmark dataset, and the loss function and model parameters are optimized through cross-validation to eliminate overfitting risks. After the accuracy on the test set reaches the target, the pre-trained main model is solidified. Inference and recognition are performed on the image data, outputting abnormal data, including abnormal type, abnormal region contour, and abnormal region pixel coordinates.

7. The visual digital twin traceability method based on single-point data according to claim 1, characterized in that, The step of performing feature registration between abnormal data and digital twins, and calculating the global coordinate interval corresponding to the abnormal region, includes: From the image data, stable rigid geometric feature points of the components are extracted to generate a two-dimensional feature point set; from the digital twin, corresponding three-dimensional feature points with the same name as the two-dimensional feature points in the two-dimensional feature point set are extracted to generate a three-dimensional feature point set; based on the mathematical registration logic of rigid body transformation, the two-dimensional feature point set and the three-dimensional feature point set are matched one by one, and point pairs with incorrect matching and excessive error are eliminated through iterative optimization to obtain a set of corresponding feature point pairs; Based on the geometric principles of perspective projection, using a set of corresponding feature point pairs as the solution benchmark and combining the device's intrinsic parameters, the pose parameters of the image acquisition device in the global coordinate system are calculated, including the device's three-dimensional spatial position and the lens's three-dimensional orientation in the global coordinate system. A reprojection error check is performed on the pose parameters by backprojecting all three-dimensional feature points to the image pixel coordinate system using the pose parameters. The deviation between the backprojected coordinates and the actual image feature point coordinates is compared. When the deviation exceeds a preset threshold, the process returns to the registration stage to optimize the matching point pairs until the error meets the accuracy requirements, thus determining the pose parameters. Based on the pose parameters involved in the device, the pixel coordinate system is converted into the device imaging plane coordinate system, and then into the device's own coordinate system. Through rigid body transformation of the pose parameters, it is transformed into the engineering global coordinate system to generate a global three-dimensional coordinate point set of the abnormal region contour. Full-dimensional extreme value statistics are performed on the global three-dimensional coordinate point set to obtain the maximum and minimum values ​​of the point set in the X, Y, and Z axes of the engineering global coordinate system, forming the global coordinate interval corresponding to the abnormal region.

8. The visual digital twin traceability method based on single-point data according to claim 1, characterized in that, The method, based on a topology model, uses a topology index to match the target traceability point and its corresponding target mesh cell within a global coordinate range to obtain matching data, including: Based on spatial topology indexing, the system filters out unique traceability node topology nodes whose global coordinate boundaries completely cover the coordinate range of the abnormal area through spatial inclusion relationships. These nodes are identified as target traceability nodes, and their codes and internal grid mapping data are output. Based on the global coordinate lookup table of the internal grid cells of the target traceability node, the system filters out grid cells that cover the range of the abnormal area through spatial intersection and inclusion determination logic. These grid cells are identified as target grid cells, and their numbers, coordinate boundaries, and associated data indexes are output. The output data is then integrated as matching data.

9. The visual digital twin traceability method based on single-point data according to claim 1, characterized in that, Based on the topological model and matching data, the process retrieves adjacent single points that have spatial and structural associations with the anomalous single point, and analyzes the impact range of the anomalous point on adjacent components and the overall structure, including: Based on the topological model, all spatially adjacent points that have spatial boundary contact or coordinate interval adjacency with the anomalous point are retrieved through spatial adjacency edges. All structurally associated points that have direct and indirect force transmission relationships with the anomalous point are traversed and retrieved through structural force-related edges, forming a list of associated adjacent points. Based on this list, and considering the anomaly type, severity, adjacency level of the associated edges, and force transmission weight, the impact of the anomaly on each adjacent point component is graded. A full force path traversal is completed along the structural associated edges, and the force diffusion path of the anomaly is determined by combining this with the structural design safety threshold, thus defining the affected boundaries and risk levels of the overall structure.

10. A visualized digital twin traceability system based on single-point data, using the visualized digital twin traceability method based on single-point data according to any one of claims 1-9, characterized in that, include: The topology model construction module includes: a single-point tracing acquisition unit to acquire the building information model and break down the smallest physical unit within the entire lifecycle of the project, defining it as a single-point tracing; a mapping relationship establishment unit to divide the internal coordinate grid of the physical component corresponding to each single-point tracing and establish the mapping relationship between the local relative coordinate system of the component and the global coordinate system of the project; and a topology model construction unit to construct the topology model of all single-point tracing in the entire project. The digital twin construction module includes: a multi-source traceability data acquisition unit that collects multi-source traceability data from a single point throughout its entire lifecycle and stores it in a traceability database; and a digital twin construction unit that, based on a topology model, mounts the traceability database onto the corresponding virtual components to obtain a digital twin. The data matching module includes: an image data acquisition unit that acquires image data of the abnormal area when quality anomalies occur in the physical project and performs preprocessing; an abnormal data output unit that pre-trains an engineering quality defect detection model, performs inference and identification on the image data, and outputs abnormal data; a global coordinate interval calculation unit that performs feature registration between the abnormal data and the digital twin, and calculates the global coordinate interval corresponding to the abnormal area; and a data matching unit that, based on a topology model, matches the target traceability point and the corresponding target grid cell belonging to the global coordinate interval through a topology index to obtain matching data. Impact Range Analysis Module: Includes: The impact range analysis unit, based on the topological structure model and matching data, retrieves adjacent single points that have spatial and structural associations with the abnormal single point, and analyzes the impact range of the abnormality on adjacent components and the overall structure.

Images