Three-dimensional digital management of building pipeline node assisted positioning method

Abstract

The application discloses a three-dimensional digital management building pipeline node auxiliary positioning method, and relates to the technical field of building construction. The method comprises the following steps: step one, collecting multi-source data of a plurality of equipment terminals, fusing the multi-source data through adaptive federated filtering, and outputting three-dimensional coordinates and component parameters of nodes; step two, pre-acquiring a BIM attribute graph of a building pipeline, constructing a measured attribute graph based on the three-dimensional coordinates and the component parameters of the nodes, screening out deviation nodes by matching the spatial position and semantic attributes of the BIM attribute graph and the measured attribute graph through a graph convolution network, correcting the deviation nodes by using reinforcement learning, and outputting a corrected calibration model; and step three, automatically highlighting corresponding pipeline components in the BIM based on the three-dimensional coordinates of the nodes and the calibration model, and superimposedly displaying design parameters and historical maintenance records; and automatically deciding a correction strategy according to the deviation type and grade by using a deep Q network, so that intelligent and iterative correction of the deviation nodes is realized, and manual intervention is greatly reduced.

Description

Technical Field

[0001] This invention relates to the field of building construction technology, specifically to a method for auxiliary positioning of building pipeline nodes using three-dimensional digital management. Background Technology

[0002] In the field of building engineering, pipeline systems are one of the core infrastructures for realizing building functions, such as water supply and drainage, HVAC, and electrical wiring. However, the existing pipeline node management has many pain points: on the one hand, pipelines are mostly buried in walls, ceilings, or underground, and the positioning method relying on two-dimensional drawings is prone to spatial information deviation, resulting in pipeline collisions during the construction phase and difficulty in finding nodes during the operation and maintenance phase; on the other hand, the design parameters and maintenance records of pipeline nodes are mostly stored in paper documents or scattered spreadsheets, making it inconvenient to retrieve information during on-site operations, which can easily lead to problems such as low operation and maintenance efficiency and delayed fault handling.

[0003] Existing building pipeline node auxiliary positioning technology has developed in parallel with 3D digital technology. Although BIM has been widely used in pipeline design, the existing BIM model has a prominent problem of digital and physical disconnect with the actual on-site working conditions. Most BIM models are static data in the design stage, which makes it difficult to match the actual pipeline location and component status after construction in real time. At the same time, single positioning technology is not accurate enough in complex environments such as indoors and underground, and cannot achieve accurate positioning of pipeline nodes. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] Step 1: Collect multi-source data from multiple device terminals, fuse the multi-source data through adaptive federated filtering, and output the three-dimensional coordinates of the nodes and component parameters;

[0006] Step 2: Pre-acquire the BIM attribute map of the building pipelines, construct the measured attribute map based on the 3D coordinates of the nodes and the component parameters, match the spatial location and semantic attributes of the BIM attribute map and the measured attribute map through graph convolutional network to filter out the deviation nodes, use reinforcement learning to correct the deviation nodes, and output the corrected calibration model.

[0007] Step 3: Based on the 3D coordinates of the nodes and the calibration model, the corresponding pipeline components in the BIM are automatically highlighted, and design parameters and historical maintenance records are overlaid and displayed.

[0008] Furthermore, the process of collecting multi-source data from multiple device terminals is as follows:

[0009] Terminals integrating BeiDou differential positioning, MEMS inertial navigation, and electromagnetic signal acquisition modules are deployed in the corresponding areas of building pipeline nodes to acquire the three-dimensional coordinates, position, attitude, and electromagnetic fingerprint of the pipeline nodes, respectively. After preprocessing, multi-source data is obtained.

[0010] Furthermore, the process of outputting the 3D coordinates of the nodes and the component parameters is as follows:

[0011] Using the three-dimensional coordinates of pipeline nodes as state variables, the positioning coordinates and covariance matrix are calculated through Kalman filtering, and the accuracy confidence score is output in combination with the number of satellites and the signal-to-noise ratio. Using the position and attitude of pipeline nodes as state variables, the recursive coordinates and covariance matrix are updated through Kalman filtering, and the confidence score is dynamically adjusted according to the drift characteristics. At the same time, using the three-dimensional coordinates of pipeline nodes as state variables, the state estimate is calculated through least squares matching, and the electromagnetic fingerprint matching degree is used as the confidence score.

[0012] Based on the confidence level, weights are assigned and the final three-dimensional coordinates of the nodes are obtained by fusing the state estimates. The component parameters are then matched by combining the electromagnetic fingerprint with the BIM pre-stored information.

[0013] Furthermore, the process of constructing the measured attribute map is as follows:

[0014] The system acquires the 3D coordinates and component parameters of the nodes, associates the 3D coordinates with the measured nodes acquired on-site, reconstructs the pipeline geometry and topology based on the association, acquires the semantic tags of the measured nodes, and constructs the measured attribute graph by combining the component parameters with the graph structure storage.

[0015] Furthermore, the process of filtering out the deviation nodes is as follows:

[0016] The node attribute fields of the BIM attribute map and the measured attribute map are aligned, and numerical differences are eliminated through min-max normalization. A topological relationship is constructed using an adjacency matrix, and unique node IDs are assigned. A network structure consisting of an input layer, two graph convolutional layers, and an output layer is used to extract high-dimensional feature vectors that fuse spatial, semantic, and topological information from both types of graphs. Spatial location similarity and semantic attribute similarity between nodes are calculated, and a comprehensive matching degree is obtained by weighting the results using spatial and semantic weight coefficients. This comprehensive matching degree is then used to determine which nodes are matched.

[0017] A node is considered a matching node when its overall matching degree is greater than or equal to the matching degree threshold.

[0018] When the overall matching degree is less than the matching degree threshold, it is determined to be an unmatched node. The specific differences in spatial location and semantic attributes are broken down to determine whether it is a deviation node. Deviation types and levels are classified, quantified into deviation levels, and deviation nodes are filtered out.

[0019] Furthermore, the process of breaking down the specific differences between spatial location and semantic attributes is as follows:

[0020] When decomposing single-dimensional differences, for nodes whose overall matching degree does not meet the standard, their spatial location similarity and semantic attribute similarity are checked separately.

[0021] When quantifying the deviation level, the three-dimensional Euclidean distance between the measured node and the corresponding BIM node is calculated for the spatial deviation node, which is the spatial straight-line distance. The deviation level is then classified according to the magnitude of the distance.

[0022] During the deviation node filtering and output process, all nodes identified as spatial deviation, semantic deviation, and composite deviation are summarized, and invalid difference nodes caused by data collection errors are removed.

[0023] Furthermore, the state space is obtained from the BIM attribute map, the measured attribute map, and the list of deviation nodes, while the action space is defined by correcting BIM parameters, re-sampling data, and marking anomalies, with the deviation elimination rate as the reward function. A deep Q-network is used to perform corrections according to the deviation type.

[0024] When spatial deviations exist, minor deviations correct BIM coordinates; moderate deviations are corrected by verifying the data and marking the cause; severe deviations are marked as construction anomalies and rectification suggestions are generated; when semantic deviations exist, minor deviations update BIM semantic attributes; severe deviations trigger a review before updating; when compound deviations exist, the highest-level deviation is prioritized, and minor deviations are corrected simultaneously in both BIM coordinates and semantics; a graph convolutional network is called for rematching, and if the deviation elimination rate is ≥90%, it is considered effective; otherwise, the strategy is adjusted and recorrected; the effectively corrected BIM and measured data are integrated to form a calibration model.

[0025] Furthermore, the process of highlighting the corresponding pipeline components in the BIM is as follows:

[0026] A mobile device equipped with a pre-installed auxiliary positioning system enters the area. The multi-source fusion positioning module configured on the mobile device obtains the three-dimensional coordinates of the personnel and simultaneously calls the calibration model. The three-dimensional coordinates of the personnel are matched with the coordinates of the pipeline nodes in the model to filter nodes. Based on the model's topological relationship, the corresponding pipeline components are associated to form a positioning and component mapping. On the mobile device interface, the calibration model view is positioned to the associated area and enlarged. The target component is filled with high-contrast colors and its outline is thickened and highlighted. The virtual model of the target component is superimposed on the real scene and rendered with luminous highlighting.

[0027] Furthermore, the process of overlaying and displaying design parameters and historical maintenance records is as follows:

[0028] Based on the highlighted component, the semantic attributes of the component are extracted from the calibration model and displayed next to the component as a floating information box; the operation and maintenance database is connected synchronously, and the historical records are retrieved and added to the information box through the component's unique ID.

[0029] The building pipeline node assisted positioning method using three-dimensional digital management provided by this invention has the following beneficial effects:

[0030] (1) This invention integrates Beidou differential positioning, MEMS inertial navigation and electromagnetic signal acquisition multi-source fusion technology to realize high-precision node coordinate acquisition and automatic identification of component parameters of pipeline nodes in electromagnetic environment, improves the robustness and accuracy of line positioning, solves the problem that traditional single positioning method is easily blocked and interfered with, resulting in insufficient accuracy, and provides reliable measured spatial reference data for subsequent BIM model calibration.

[0031] (2) This invention introduces graph convolutional networks to achieve two-dimensional matching of spatial location and semantic attributes between BIM attribute maps and measured attribute maps, and uses deep Q networks to automatically decide on correction strategies based on deviation type and level, thereby realizing intelligent and iterative correction of pipeline deviation nodes, greatly reducing manual intervention, improving model calibration efficiency and consistency, and avoiding the disconnect between the design model and the actual digital and physical data on site.

[0032] (3) This invention integrates the calibrated BIM model with mobile terminals or AR devices to achieve automatic highlighting of pipeline components located by on-site personnel, real-time floating overlay display of design parameters and historical maintenance records, forming a complete closed loop from pipeline data collection, calibration and operation and maintenance, improving the level of intelligent inspection and operation and maintenance, reducing on-site search time, reducing fault handling delays, and supporting traceable record management, thereby digitally managing the entire life cycle of building pipeline systems. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the method of the present invention. Detailed Implementation

[0034] 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.

[0035] Please see Figure 1 This application provides an embodiment of a method for assisted positioning of building pipeline nodes using three-dimensional digital management, the method comprising:

[0036] Step 1: Collect multi-source data from multiple device terminals, fuse the multi-source data through adaptive federated filtering, and output the three-dimensional coordinates of the nodes and component parameters;

[0037] Collection of multi-source data from multiple device terminals:

[0038] At building pipeline nodes, such as the construction or maintenance areas corresponding to valves and joints, the building pipeline nodes are marked by multi-device terminals that integrate Beidou differential positioning module, MEMS inertial navigation module and electromagnetic signal acquisition module. At the same time, electromagnetic signal strength samples of the area are pre-collected to build an electromagnetic environment fingerprint database.

[0039] The BeiDou module receives carrier phase signals from multiple BeiDou satellites and synchronously collects basic information such as timestamps and satellite numbers corresponding to the signals; the inertial navigation module collects the angular velocity and acceleration data of the carrier in real time through built-in gyroscopes and accelerometers; the electromagnetic module collects the electromagnetic signal strength at the current location, such as the RSSI value of Wi-Fi and Bluetooth bands, and associates it with the preliminary coordinates of the collected location.

[0040] Interference filtering is applied to BeiDou signals to separate noise signals in complex electromagnetic environments; zero-bias compensation is applied to inertial navigation data to correct inherent sensor errors; electromagnetic signal data is normalized and matched with a pre-built fingerprint database; BeiDou, inertial navigation, and electromagnetic data at the same timestamp are associated according to the collection location, data with excessive time synchronization errors are removed, and a valid multi-source data set with time alignment is retained.

[0041] The MEMS inertial navigation module collects the angular velocity and acceleration data of the carrier at a high frequency of 100Hz, and records the module's temperature to correct temperature drift errors and timestamps the data. Every second, it recursively calculates the current terminal's position, velocity, and attitude information through the strapdown inertial navigation algorithm as supplementary data when the BeiDou signal is interrupted.

[0042] Low-frequency electromagnetic field signals generated by current in pipelines or external excitation sources include passive electromagnetic signals, active electromagnetic signals, and radio frequency identification or near-field communication signals. Passive electromagnetic signals exist naturally; they are electromagnetic fields generated by the pipeline itself under normal operating conditions without the need for external equipment excitation. For example, the electromagnetic field generated by the pipeline itself under normal operating conditions does not require external equipment excitation. High-voltage power lines in buildings, such as 220V / 380V power supply lines, will naturally radiate low-frequency electromagnetic fields of 50Hz and its harmonics when energized. This signal can be used to: locate live cables, determine whether the line is energized, and estimate the load current, i.e., infer from the magnetic field strength.

[0043] Active electromagnetic signals are artificially excited by applying a specific frequency alternating current to the pipeline using a dedicated transmitting device, causing it to radiate a controllable electromagnetic field. The receiver then collects the data at the ground or at nodes. A pipeline detector applies a kHz-level sinusoidal signal to the metal pipe or cable. This signal propagates along the pipeline and generates an electromagnetic field surrounding it. The receiver detects the strength and direction of this electromagnetic field to locate the pipeline route, estimate its depth, and identify node connections.

[0044] Radio frequency identification (RFID) or near-field communication (NFC) signals are used to install passive RFID tags on pipeline nodes, such as valves, water meters, and electricity meters. When inspection personnel approach the node with a handheld reader, the reader emits radio frequency electromagnetic waves of 13.56MHz or 900MHz. After the tag is activated, it returns the stored ID information, such as pipeline number, material, and installation date.

[0045] The collected raw data contains various noises and outliers. BeiDou data preprocessing is performed. First, satellite signals with a signal-to-noise ratio (SNR) below 30dB are removed, as these signals are easily interfered with. Effective signals with an SNR ≥ 30dB are retained. Second, cycle slip detection and repair are performed on the carrier phase values. Cycle slips are identified using the phase difference method and repaired using polynomial fitting. Preliminary corrections are made to the pseudorange values ​​for ionospheric and tropospheric delays. Based on the model parameters provided by the broadcast ephemeris, the corrected BeiDou pseudorange and carrier phase data are obtained.

[0046] During inertial navigation data preprocessing, firstly, zero bias compensation is performed on the accelerometer data. This is done by correcting the zero bias value obtained from static acquisition and then compensating for temperature drift in the gyroscope data based on a fitting model of temperature and drift. Secondly, the acquired angular velocity and acceleration data are low-pass filtered with a cutoff frequency of 5Hz to remove high-frequency noise. Finally, the recursively obtained position and velocity information is validated for reasonableness. If the velocity exceeds 0.5 m / s, it is considered abnormal and needs to be removed, while normal inertial navigation data is retained.

[0047] Electromagnetic signal data preprocessing: First, invalid signals with a strength below -90dBm are removed, as these signals are susceptible to environmental interference. Valid signals with a strength ≥ -90dBm are retained. Second, a moving average is applied to the signal strength within the same frequency band with a window size of 3 to smooth signal fluctuations. Finally, matching results with a fingerprint matching degree below 0.5 are marked, and their weight is reduced during subsequent fusion. Valid matching data with a matching degree ≥ 0.5 are retained.

[0048] After preprocessing, the three types of data are reordered by timestamp to form a linked dataset of timestamp, BeiDou data, inertial navigation data and electromagnetic data, so that each timestamp contains complete data of the three types.

[0049] The process of outputting the 3D coordinates of the nodes and the component parameters is as follows:

[0050] The BeiDou, inertial navigation, and electromagnetic data are treated as three independent sub-filtering modules, each completing preliminary state estimation based on its own observation model. The BeiDou sub-module state estimation uses the three-dimensional coordinates of the pipeline node as the state variable, constructs the observation equations for the BeiDou carrier phase and pseudorange, and calculates the state estimate value, i.e., the BeiDou positioning coordinates and the corresponding covariance matrix, through Kalman filtering, reflecting the positioning accuracy. At the same time, combined with the number of satellites and the signal-to-noise ratio, the real-time accuracy confidence level of this sub-module is output, ranging from 0 to 1, with higher values ​​indicating more reliable BeiDou positioning.

[0051] The inertial navigation submodule's state estimation uses the node's position, velocity, and attitude as state variables. Based on the recursive equations of inertial navigation, it performs attitude, velocity, and position updates to construct the state equations. Combining observation data from accelerometers and gyroscopes, it updates the state estimates, inertial navigation recursive coordinates, and covariance matrix through Kalman filtering. Simultaneously, based on the inertial navigation drift characteristics, i.e., a drift rate of approximately 0.1° / h, the accuracy confidence level of this submodule is dynamically adjusted; the longer the recursion time, the lower the confidence level.

[0052] The electromagnetic submodule state estimation uses the three-dimensional coordinates of the nodes as state variables. Based on the correspondence between the coordinates and signal features in the electromagnetic fingerprint database, an observation equation for the signal strength is constructed. The state estimate and covariance matrix are calculated through least squares matching. At the same time, the fingerprint matching degree is used as the accuracy confidence of the submodule. The higher the matching degree, the higher the confidence.

[0053] Adaptive weight allocation and main module fusion calculation:

[0054] Based on the accuracy confidence levels of the three sub-modules, information weights are dynamically allocated. Then, the state estimates of each sub-module are fused through the main filtering module. The weight coefficient of each sub-module is calculated, and the weight is positively correlated with the accuracy confidence level of the sub-module. That is, the weight of a sub-module is equal to its own confidence level divided by the sum of its confidence levels and those of the other two modules. The main filtering module calculates the fused state estimate, i.e., the final 3D coordinates of the nodes, based on the state estimates and covariance matrix of each sub-module using the information fusion formula. The accuracy of the fused coordinates is determined by the covariance and weights of each sub-module. Sub-modules with higher confidence levels have a greater impact on the fusion result.

[0055] Combining the signal characteristics collected by the electromagnetic submodule with the pre-stored information of the BIM model, the pipeline component parameters corresponding to the current node are matched, such as pipe diameter, material and design elevation. By fusing coordinates to associate the components in the BIM model, the preset parameters of the component are extracted, and the consistency is verified with the component characteristics corresponding to the electromagnetic signals collected on site. Finally, the component parameters are output.

[0056] Post-validation is performed on the output 3D coordinates and component parameters. The fused coordinates are compared with the known reference point coordinates of the area. If the deviation is ≤2 cm, the coordinates are considered valid. If the deviation is >2 cm, the state estimation of the submodule is retried, the weights are adjusted, and the fusion is performed again to verify the coordinate accuracy. The component parameters output by fusion are compared with the pre-stored parameters of the BIM model. If the parameter consistency rate is ≥90%, such as matching pipe diameter and material, the parameters are considered valid. If the consistency rate is <90%, the parameters are marked as abnormal, and a prompt message for on-site manual review is added to verify the component parameters.

[0057] Step 2: Pre-acquire the BIM attribute map of the building pipelines, construct the measured attribute map based on the 3D coordinates of the nodes and the component parameters, use graph convolutional network to match the spatial location and semantic attributes of the BIM attribute map and the measured attribute map to filter out the deviation nodes, use reinforcement learning to correct the deviation nodes, and output the corrected calibration model.

[0058] The original model files of building pipelines were exported from the project's BIM platform and imported into the model processing module of the auxiliary positioning system. Then, using the component classification tags of the BIM model, all pipeline-related components were selected, including core nodes such as pipe bodies, valves, joints, branch points, and flanges. Non-pipeline components such as building structures and electromechanical equipment were removed, resulting in a preliminary model set containing only the pipeline system. For the selected pipeline components, their geometric information was extracted and topological relationships were constructed, forming the spatial dimension attribute basis. Geometric information extraction was based on the project's local engineering coordinate system, extracting the three-dimensional coordinates of each pipeline component, including the component's center point coordinates, endpoint coordinates, length, pipe diameter, elevation, and direction. For node-type components such as valves and joints, the precise coordinates of their installation locations were additionally extracted.

[0059] The topology relationship is constructed based on the endpoint connection relationship of the components to build the topology network of the pipeline system: if the endpoint coordinates of two components coincide and the error is ≤5 mm, it is determined that there is a connection relationship between them. Based on this, the adjacency relationship table of the pipeline is generated, the upstream and downstream connection objects of each component are clarified, and a topology network structure of components, connection relationships and spatial locations is formed.

[0060] Based on geometric information, semantic attributes of components are added to form a complete BIM attribute map. Existing semantic information of components in the BIM model is extracted, including preset attributes such as component type, material, specifications, design pressure, design temperature, construction team, and installation time. For auxiliary positioning needs, custom attribute labels are added, including unique component ID, node importance (marked as core node or general node according to maintenance priority), and allowable deviation threshold. The geometric information, topological relationships, and semantic attributes of all components are integrated and stored in the form of a graph structure. Each pipeline component is a node of the graph, and the connection relationship between components is an edge of the graph. Node attributes contain both geometric and semantic information, ultimately forming a structured BIM attribute map.

[0061] The process of constructing the measured attribute map is as follows:

[0062] Based on the results of multi-source fusion, the three-dimensional coordinates and component parameters of pipeline nodes are extracted to complete the preliminary classification of data and node association. Node coordinates are extracted from each pipeline node, such as valves and joints, and the fused three-dimensional coordinates are transformed into the project's local engineering coordinate system. Simultaneously, coordinate accuracy indicators, such as centimeter-level error range, are recorded. The three-dimensional coordinates are mapped to the physical nodes marked on-site to ensure that each measured node is associated with a specific pipeline component. Component parameters are extracted from the fusion results, including the pipe diameter, material, and current status measured on-site. Material is supplemented through visual recognition or handheld device detection. Current status includes, for example, valve on / off status. On-site information such as parameter acquisition time and personnel is also recorded.

[0063] Based on the coordinates and relationships of the measured nodes, the geometry and topology of the pipelines on site are reconstructed, forming the measured spatial attributes. The geometric information integration is based on the three-dimensional coordinates of the measured nodes, supplemented by geometric data such as pipeline segment length, direction, and elevation deviation measured on site. For example, the pipeline segment length between node A and node B is 5.2 meters, and the elevation is 2 centimeters lower than the design. For node-type components, the actual offset of their installation position is marked.

[0064] Topology restoration reconstructs the topology network of the measured pipeline by analyzing the physical connections of the nodes on site. If the components corresponding to two measured nodes have a physical connection, such as valve and pipe welding, the connection relationship between the two is established in the measured drawing, generating an adjacency list of the measured pipeline, clarifying the actual connection objects of each node, and forming a correspondence with the topology structure of the BIM attribute map.

[0065] Based on spatial information, semantic attributes of the measured nodes are supplemented to ultimately construct a complete measured attribute graph. During the semantic attribute supplementation process, semantic tags are added to each measured node, including a unique node ID, acquisition timestamp, parameter confidence level, and on-site anomaly markers. The graph structure is integrated and stored by treating each measured node as a node in the graph, with physical connections between nodes as edges. Node attributes include 3D coordinates, geometric parameters, topological relationships, and semantic tags, in conjunction with BIM attributes. Figure 1 The measured attribute map is stored in a consistent graph structure format, forming a measured attribute map of actual nodes, connections, and site attributes. This ensures that its structure is fully compatible with the BIM attribute map, providing a unified graph data foundation for subsequent matching. The completed measured attribute map is verified to check the consistency of the coordinates of each node with the physical location on site, obtain the measured 3D coordinates of the nodes, and check the matching degree between the topological relationships and the actual connections on site to confirm the integrity of the semantic attributes. After the verification is passed, the measured attribute map can be used for the matching process with the BIM attribute map.

[0066] The process of filtering out deviation nodes is as follows:

[0067] By matching BIM attribute maps with measured attribute maps using graph convolutional networks, precise alignment of spatial location and semantic attributes is achieved. The feature learning capability of graph structures is used to mine the deep correlation between the two types of attribute maps, and pipeline nodes with spatial or semantic differences are located. Deviation nodes are filtered out through graph data preprocessing, graph feature extraction, two-dimensional matching calculation, and deviation node determination.

[0068] During the format unification process, although BIM attribute maps and measured attribute maps are both graph structures, the number and units of node attributes may differ. For example, the pipe diameter unit in BIM is millimeters, while in actual measurement it may be centimeters. Therefore, it is necessary to first align the node attribute fields of the two types of maps. Core common fields such as 3D coordinates, pipe diameter, material, component type, and topological connection relationship are retained, while non-critical fields unique to a certain type of map are removed. Attributes with inconsistent units are uniformly converted, and the measured pipe diameter is unified to millimeters to ensure that the attribute dimensions are completely matched. At the same time, considering that the numerical range of different attributes varies greatly, directly inputting them into the network will lead to an imbalance of feature weights. The min-max normalization method is adopted to map the attribute values ​​of all nodes in the two types of maps to the [0,1] interval, eliminating the impact of numerical range differences on feature learning, and ensuring that the topological representation format of the two types of maps is consistent. Both use an adjacency matrix to describe the connection relationship between nodes. An element of 1 in the matrix indicates that there is a connection between the two corresponding nodes, and an element of 0 indicates that there is no connection. At the same time, a unique identifier ID is assigned to each node to facilitate the traceability of subsequent matching results.

[0069] Leveraging the core advantages of graph convolutional networks, this method mines the correlation information between the attributes of nodes themselves and those of neighboring nodes. Feature extraction is performed on both the BIM attribute map and the measured attribute map to obtain high-dimensional node feature vectors containing spatial, semantic, and topological information. A graph convolutional network structure is constructed. The network adopts a lightweight structure with an input layer, two graph convolutional layers, and an output layer, adaptable to the node scale of pipeline attribute maps, typically ranging from hundreds to thousands of building pipeline nodes. The input layer receives standardized node attribute features, such as feature vectors composed of common attributes like 3D coordinates and pipe diameter. The two graph convolutional layers are responsible for progressively mining features. The first graph convolutional layer focuses on the local correlation features between the node's own attributes and directly adjacent nodes, such as the attribute matching relationship between a valve node and its upstream and downstream pipelines. The second graph convolutional layer further integrates local features with the global correlation features of indirectly adjacent nodes, such as the overall attribute features of the pipeline branch where the valve node is located. The output layer outputs a fixed-dimensional high-dimensional feature vector, such as 64-dimensional or 128-dimensional, which integrates the spatial location, semantic attributes, and topological relationships of the node.

[0070] When performing feature learning through graph segmentation, the preprocessed BIM attribute map and the measured attribute map are input into the trained graph convolutional network. The network is pre-trained using a large amount of attribute map data of similar building pipelines to ensure that it has general feature extraction capabilities. During the feature learning process, the network perceives the neighborhood environment of each node through the adjacency matrix. For example, for a pipe node in the BIM attribute map, the network will combine its own pipe diameter, material attributes, and the attribute information of adjacent valves and joints to generate a comprehensive feature vector that can characterize the spatial rationality, semantic attribute integrity, and topological connection correctness of the node. The feature extraction process of the measured attribute map is completely consistent with this, and finally, a set of high-dimensional feature vectors corresponding to all nodes in both types of maps is obtained.

[0071] Based on the extracted high-dimensional feature vectors, the spatial location matching degree and semantic attribute matching degree between nodes of the BIM attribute map and the measured attribute map are calculated respectively. The comprehensive matching degree is obtained by weighted fusion to achieve two-dimensional collaborative matching. In the spatial location matching degree calculation, the feature components related to spatial location in the feature vectors of the two types of graph nodes are selected and automatically learned and filtered by the graph convolutional network. Corresponding to the three-dimensional coordinate attributes of the original input, the cosine similarity calculation method is used to measure the spatial location similarity between BIM nodes and measured nodes. The core logic of cosine similarity is to calculate the cosine value of the angle between the two feature vectors to judge the consistency of the vector directions. The more consistent the directions are, the higher the similarity. The value range is [0,1].

[0072] If the spatial similarity between a BIM node and a measured node is ≥0.85, it is preliminarily determined that they may match in spatial location. Next, semantic attribute matching is calculated, selecting feature components related to semantic attributes from the feature vector, corresponding to attributes such as pipe diameter, material, and component type in the original input. The cosine similarity calculation method is also used to measure the semantic consistency between the two types of nodes. For example, if both the BIM node and the measured node are DN100 galvanized steel pipes, the semantic similarity is close to 1; if the measured node is a DN80 galvanized steel pipe, the semantic similarity is significantly reduced. A semantic similarity ≥0.8 is set as the initial threshold for semantic matching. Threshold; finally, a weighted fusion of comprehensive matching degree is used. Considering the different importance of spatial location and semantic attributes to the matching results, the weights are determined by the analytic hierarchy process: the weight of spatial location is set to 0.6, and the weight of semantic attributes is set to 0.4. The core of pipeline positioning is spatial alignment, while semantic attributes need to help verify the rationality of the matching. The formula is: comprehensive matching degree = (spatial location similarity × 0.6) + (semantic attribute similarity × 0.4). A comprehensive matching degree ≥ 0.82 is set as the final matching threshold. If the comprehensive matching degree of a BIM node and a measured node reaches this threshold, they are determined to be matching nodes; if not, the deviation node screening process is entered.

[0073] For node pairs that do not meet the comprehensive matching threshold, the specific differences in spatial location and semantic attributes are further analyzed to determine whether they are deviation nodes, and the deviation type and level are classified to provide a basis for subsequent correction.

[0074] S201: When decomposing single-dimensional differences, for node pairs whose overall matching degree does not meet the standard, check their spatial location similarity and semantic attribute similarity respectively: if the spatial location similarity is <0.85 and the semantic attribute similarity is ≥0.8, it is judged as a spatial deviation node, that is, the semantic attributes such as component type and material are consistent, but the actual installation location deviates from the BIM design location; if the semantic attribute similarity is <0.8 and the spatial location similarity is ≥0.85, it is judged as a semantic deviation node, that is, the installation location is consistent with the design location, but the component specifications, materials, etc. do not match the BIM design, and there may be construction mismatch; if both similarities are <threshold, it is judged as a composite deviation node, that is, there is both spatial location deviation and semantic attribute deviation.

[0075] S202: When quantifying deviation levels, for spatial deviation nodes, calculate the Euclidean distance (i.e., the straight-line distance) between the measured node and the corresponding BIM node in three-dimensional coordinates. The deviation level is determined according to the distance: distance ≤ 3 cm is a slight deviation, 3 cm < distance ≤ 10 cm is a moderate deviation, and distance > 10 cm is a severe deviation. For semantic deviation nodes, the level is determined according to the degree of difference in semantic attributes: for example, a pipe diameter deviation ≤ 10 mm and the material is a substitute of the same type, such as galvanized steel pipe and stainless steel pipe, is a slight deviation; a pipe diameter deviation > 10 mm and the material is a substitute of the same type is a severe deviation. For composite deviation nodes, the level is determined according to the principle of choosing the more severe deviation type.

[0076] S203: During the deviation node filtering and output process, all nodes judged as spatial deviation, semantic deviation, and composite deviation are summarized, and invalid difference nodes caused by data collection errors, such as errors in actual data entry, are removed. Finally, a deviation node list is formed, which includes the node's unique ID, deviation type, deviation level, and specific difference parameters, such as spatial deviation distance and semantic attribute difference content.

[0077] Output the corrected calibration model:

[0078] By constructing a reinforcement learning environment for deviation correction and initializing the agent's decision-making strategy, the BIM attribute map, measured attribute map, and deviation node list are used as the environment state space during environment modeling. The state information includes the type of deviation node, deviation level, current BIM model parameters, and measured data parameters. Correcting BIM model parameters, re-collecting measured data, and marking construction anomalies are used as the agent's action space. The deviation elimination rate and the reduction ratio of deviation nodes after correction are used as the core indicators of the reward function. If the deviation of the node is eliminated after correction, a positive reward is given; if the deviation is amplified or the correction is ineffective, a negative reward is given. During the strategy initialization process, a deep Q-network is used as the agent's decision-making model. Based on historical deviation correction examples, such as the pipeline deviation processing data pre-trained model of similar projects, the agent initially has the basic ability to match correction actions according to the deviation type. For example, for nodes with slight spatial deviations, the initial strategy tends to correct the BIM model coordinates.

[0079] The agent autonomously selects and executes the optimal correction action based on the type and level of the deviation node during the correction process of the spatial deviation node:

[0080] If the spatial deviation is slight (≤3 cm): the agent selects to correct the BIM model parameters, adjusts the coordinate parameters of the corresponding nodes in the BIM attribute map according to the three-dimensional coordinates of the measured nodes, aligns the BIM model with the measured position, and records the corrected coordinate deviation.

[0081] If the spatial deviation is moderate, between 3 and 10 centimeters: the agent first triggers the action of secondary verification of measured data, calls multiple device terminals to re-collect the measured data of the node, and if the deviation still exists after the secondary collection, then the action of correcting the cause of the BIM model coordinate and annotation deviation is executed.

[0082] If the spatial deviation is greater than 10 cm: the agent selects to mark the construction abnormality and generate rectification suggestions, marks the node as a construction deviation, and outputs a rectification instruction for repositioning and installation.

[0083] During the correction of semantic deviation nodes, if the semantic deviation is slight, such as pipe diameter deviation ≤10 mm: the agent selects to update BIM semantic attribute action, replaces the material, specifications and other parameters of the corresponding node in the BIM attribute map with measured data, and simultaneously supplements the attribute update record.

[0084] If there is a serious semantic deviation, such as a pipe diameter deviation > 10 mm: the intelligent agent will select to trigger a manual review action, push the deviation information to the site management personnel, and wait for the manual confirmation of the cause of the deviation, such as incorrect construction materials, before executing the BIM attribute update or construction rectification action.

[0085] During the correction of composite deviation nodes, the agent makes decisions based on the principle of prioritizing the highest deviation level: if the semantic deviation is severe, manual review and construction rectification are prioritized; if the spatial deviation is severe, marking construction anomalies is prioritized; if both are minor, BIM coordinate and semantic attribute corrections are performed simultaneously.

[0086] After the action is executed, the correction effect is verified, and the BIM model is updated to obtain a calibration model. The agent is then re-invoked through the graph convolutional network to match the corrected BIM data with the measured data again, and the deviation elimination rate is calculated. If the deviation elimination rate is ≥90%, the correction is deemed effective; if it is <90%, the decision-making process is restarted, and the action strategy is adjusted, such as changing the BIM model correction to re-collecting data. The calibration model is then integrated, fusing all effective corrected BIM attribute maps with the measured attribute maps, preserving the original BIM parameters, correction traces, and correlation information of the measured data to form a calibration model that includes spatial alignment and semantic consistency. At the same time, correction record attributes are supplemented, including correction time, correction action, and executor, to ensure model traceability. The final output calibration model aligns with the physical spatial information of the measured data while preserving the semantic and topological relationships of the BIM model.

[0087] Step 3: Based on the 3D coordinates of the nodes and the calibration model, the corresponding pipeline components in the BIM are automatically highlighted, and design parameters and historical maintenance records are overlaid and displayed.

[0088] Real-time correlation between on-site positioning and calibration model:

[0089] When a mobile device or AR device with integrated positioning function and pre-installed auxiliary positioning system enters the pipeline area, the device acquires the current three-dimensional coordinates of the personnel in real time through the multi-source fusion positioning module and simultaneously calls the calibration model in the system. The device matches the real-time collected personnel position coordinates with the pipeline node coordinates in the calibration model, and uses Euclidean distance to filter and lock the pipeline node closest to the personnel's current position. When the distance is ≤1 meter, the association is triggered. Based on the topological relationship of the calibration model, the device automatically associates the pipeline components corresponding to the node, including the node itself and the upstream and downstream connected pipes, valves, etc., to form an association mapping between the current position and the target component.

[0090] Automatic highlighting and visualization rendering of BIM components:

[0091] After completing the location and model association, the system highlights the target component in the mobile or AR interface: If using a mobile device, the system positions the BIM view of the calibration model to the currently associated component area, automatically zooms in on the view of that area, and highlights the target component with high-contrast color filling and bold outline, such as filling the target component with orange by default, to distinguish it from the gray display of other unassociated components; If using an AR device, the system recognizes the site environment through the camera, overlays the target component in the calibration model onto the real scene as a virtual model, and simultaneously performs highlight rendering on the virtual model, such as adding a glowing effect, to ensure that on-site personnel can intuitively identify the physical location of the target component; To avoid interference from multiple components, the system only highlights the core component currently located and associated, while other unassociated components are displayed with low transparency to focus the visual focus.

[0092] Overlaying and retrieving design parameters and historical information:

[0093] Based on the highlighted components, the system automatically extracts related information from the calibration model and project database, and overlays it on the visualization interface: it extracts the semantic attributes of the component from the calibration model, including basic parameters such as component type, pipe diameter, material, design pressure, design elevation, and installation location, and displays them next to the highlighted component in the form of a floating information box; the system synchronously connects to the project operation and maintenance database, retrieves the component's historical records based on the component's unique ID, including maintenance time, maintenance content, such as valve seal replacement on May 12, 2024, maintenance personnel, and cause of failure, and adds this information to the information box;

[0094] The information display supports interactive operations: on-site personnel can click on the information box to expand or collapse the content, or switch between design parameters or history tabs to view detailed information as needed.

[0095] During the display process, the system will make dynamic adjustments based on on-site operations. If on-site personnel move, the system will update the positioning coordinates in real time, re-associate the nearest pipeline components, and synchronously switch the highlighted objects and overlay information.

[0096] If on-site personnel make new inspection records for components, such as marking a minor valve leak, the system will automatically associate the record with the component ID and update the historical record content in the information box. For different scenarios, such as the dim lighting in underground utility tunnels, the system supports manually adjusting the brightness of the highlight color and the font size of the information box to ensure that the displayed content is clearly visible.

[0097] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0098] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0099] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for assisted positioning of building pipeline nodes using three-dimensional digital management, characterized in that, The method includes: Step 1: Collect multi-source data from multiple device terminals, fuse the multi-source data through adaptive federated filtering, and output the three-dimensional coordinates of the nodes and component parameters; Step 2: Pre-acquire the BIM attribute map of the building pipelines, construct the measured attribute map based on the 3D coordinates of the nodes and the component parameters, match the spatial location and semantic attributes of the BIM attribute map and the measured attribute map through graph convolutional network to filter out the deviation nodes, use reinforcement learning to correct the deviation nodes, and output the corrected calibration model. The process of constructing the measured attribute map is as follows: The system acquires the 3D coordinates and component parameters of the nodes, associates the 3D coordinates with the measured nodes acquired on site, reconstructs the pipeline geometry and topology based on the association, acquires the semantic tags of the measured nodes, and constructs the measured attribute graph by combining the component parameters with graph structure storage. The process of filtering out deviation nodes is as follows: The node attribute fields of the BIM attribute map and the measured attribute map are aligned, and numerical differences are eliminated through min-max normalization. A topological relationship is constructed using an adjacency matrix, and unique node IDs are assigned. A network structure consisting of an input layer, two graph convolutional layers, and an output layer is used to extract high-dimensional feature vectors that fuse spatial, semantic, and topological information from both types of graphs. Spatial location similarity and semantic attribute similarity between nodes are calculated, and a comprehensive matching degree is obtained by weighting the results using spatial and semantic weight coefficients. This comprehensive matching degree is then used to determine which nodes are matched. A node is considered a matching node when its overall matching degree is greater than or equal to the matching degree threshold. When the overall matching degree is less than the matching degree threshold, it is determined to be an unmatched node. The specific differences in spatial location and semantic attributes are broken down to determine whether it is a deviation node. Deviation types and levels are divided, quantified into deviation levels, and deviation nodes are filtered out. The process of breaking down the specific differences between spatial location and semantic attributes is as follows: When decomposing single-dimensional differences, for nodes whose overall matching degree does not meet the standard, their spatial location similarity and semantic attribute similarity are checked separately. When quantifying the deviation level, the three-dimensional Euclidean distance between the measured node and the corresponding BIM node is calculated for the spatial deviation node, which is the spatial straight-line distance. The deviation level is then classified according to the magnitude of the distance. During the deviation node filtering and output process, all nodes judged as spatial deviation, semantic deviation, and composite deviation are summarized, and invalid difference nodes caused by data collection errors are removed. The process of outputting the corrected calibration model is as follows: The state space is obtained from the BIM attribute map, the measured attribute map, and the list of deviation nodes. The action space is to correct BIM parameters, re-acquire data, and mark anomalies. The reward function is the deviation elimination rate. A deep Q-network is used to perform corrections according to the deviation type. When spatial deviations exist, minor deviations correct BIM coordinates; moderate deviations are corrected by verifying the data and marking the cause; severe deviations are marked as construction anomalies and rectification suggestions are generated; when semantic deviations exist, minor deviations update BIM semantic attributes; severe deviations trigger a review before updating; when compound deviations exist, the highest level deviation is prioritized, and minor deviations are corrected simultaneously in both BIM coordinates and semantics; the graph convolutional network is called for rematching, and if the deviation elimination rate is ≥90%, it is considered effective; otherwise, the strategy is adjusted and recorrected; the effectively corrected BIM and measured data are integrated to form a calibration model; Step 3: Based on the 3D coordinates of the nodes and the calibration model, the corresponding pipeline components in the BIM are automatically highlighted, and design parameters and historical maintenance records are overlaid and displayed.

2. The building pipeline node auxiliary positioning method using three-dimensional digital management according to claim 1, characterized in that, The process of collecting multi-source data from multiple device terminals is as follows: Terminals integrating BeiDou differential positioning, MEMS inertial navigation, and electromagnetic signal acquisition modules are deployed in the corresponding areas of building pipeline nodes to acquire the three-dimensional coordinates, position, attitude, and electromagnetic fingerprint of the pipeline nodes, respectively. After preprocessing, multi-source data is obtained.

3. The building pipeline node auxiliary positioning method using three-dimensional digital management according to claim 2, characterized in that, The process of outputting the 3D coordinates of the nodes and the component parameters is as follows: Using the three-dimensional coordinates of pipeline nodes as state variables, the positioning coordinates and covariance matrix are calculated through Kalman filtering, and the accuracy confidence is output by combining the number of satellites and the signal-to-noise ratio. Using the position and attitude of pipeline nodes as state variables, the recursive coordinates and covariance matrix are updated through Kalman filtering, and the confidence level is dynamically adjusted according to the drift characteristics. At the same time, using the three-dimensional coordinates of pipeline nodes as state variables, the state estimate is calculated through least squares matching, and the electromagnetic fingerprint matching degree is used as the confidence level. Based on the confidence level, weights are assigned and the final three-dimensional coordinates of the nodes are obtained by fusing the state estimates. The component parameters are then matched by combining the electromagnetic fingerprint with the BIM pre-stored information.

4. The building pipeline node auxiliary positioning method using three-dimensional digital management according to claim 1, characterized in that, The process of highlighting the corresponding pipeline components in the BIM is as follows: A mobile positioning device equipped with a pre-installed auxiliary positioning system enters the area. The multi-source fusion positioning module configured on the mobile positioning device is used to obtain the three-dimensional coordinates of the personnel and simultaneously call the calibration model. The three-dimensional coordinates of the personnel are matched with the coordinates of the pipeline nodes in the model to filter nodes. Based on the topological relationship of the model, the corresponding pipeline components are associated to form a positioning and component mapping. On the mobile interface, the calibration model view is positioned to the associated area and zoomed in. The target component is filled with high-contrast colors and its outline is thickened and highlighted. The virtual model of the target component is superimposed on the real scene and rendered with luminous highlights.

5. The method for assisted positioning of building pipeline nodes using three-dimensional digital management according to claim 1, characterized in that, The process of overlaying design parameters and historical maintenance records is as follows: Based on the highlighted component, the semantic attributes of the component are extracted from the calibration model and displayed next to the component as a floating information box; the operation and maintenance database is connected synchronously, and the historical records are retrieved and added to the information box through the component's unique ID.

Images