Method for automatically generating multi-professional pipe network BIM model based on CAD drawing of graph neural network

Abstract

The invention relates to a method for automatically generating a multi-professional pipe network BIM model based on a CAD drawing of a graph neural network, comprising: pipe network CAD drawing information analysis and multi-professional pipe network BIM model generation, the pipe network CAD drawing information analysis comprising graph neural network model construction, CAD drawing analysis based on the graph neural network model, legend table analysis and multi-graph information aggregation; the multi-professional pipe network BIM model is generated based on the extracted pipe network component information, and different reconstruction strategies are adopted for different types of pipe network components. According to the method, an automatic building pipe network CAD drawing analysis technical route combining deep learning driving and knowledge rule constraint is constructed, symbol primitive semantic extraction is carried out based on a graph neural network, high-accuracy symbol and layer automatic identification is realized, a pipe network system expressed by multiple CAD drawings is converted into a graph structure, and the method is suitable for large-scale popularization and application. And entity matching and information aggregation are carried out on a plurality of drawings and a plurality of graph structures, so that an automatic, efficient and high-stability building pipe network system information model reconstruction task is realized.

Description

Technical Field

[0001] The present invention relates to a method for automatically generating a multi-disciplinary pipe network BIM model, and in particular to a method for automatically generating a multi-disciplinary pipe network BIM model from a CAD drawing based on a graph neural network. Background Art

[0002] Building pipe networks provide water, electricity, heating, and ventilation services to residents, ensuring the proper functioning of these primary urban hubs. In recent years, countless fires and explosions have been attributed to urban pipe networks. A key reason for this is that existing management of urban pipe networks is largely limited to two-dimensional drawings, failing to intuitively reflect the three-dimensional spatial information of the pipe network system and thus failing to support the comprehensive management and maintenance needs of urban pipe networks. BIM-based building pipe networks can effectively support intelligent management, enabling managers to promptly identify potential problems in the system and quickly and accurately locate pipes or facilities requiring repair. BIM modeling during the design and planning phase can also effectively support the smooth construction of building pipe networks. Therefore, achieving efficient and accurate three-dimensional modeling of building pipe networks is of great significance.

[0003] Building pipe network CAD drawings record in detail the semantics, location, geometry, attributes, and element relationships of building pipe network entities. With comprehensive information and high expression accuracy, they are a high-quality data source for 3D modeling of building pipe networks. Drawing-based 3D pipe network modeling methods use pipe network CAD drawings as their data source and extract relevant information from the drawings to perform 3D modeling of building pipe networks. In the early stages of development, this type of method was mainly based on manual interaction. That is, based on existing drawings, professional 3D modeling software was used to create and edit various pipes and place corresponding pipe points and equipment in corresponding locations to complete the construction of the 3D model. Although it has a high degree of sophistication, the modeling process is relatively cumbersome and requires the input of a large amount of parameters, attributes, and other information. This requires a high level of operator expertise and has high modeling costs, making it unsuitable for large-scale pipe network modeling.

[0004] Because CAD drawings depict three-dimensional architectural entities in two dimensions, the information they contain is inevitably distributed across multiple drawings of varying types. This is particularly evident in CAD drawings of building pipe networks. While some relevant research has addressed this issue and proposed targeted solutions, a universal method for integrating information from multiple drawings has yet to be achieved. This ultimately results in insufficient utilization of drawing information and limited efficiency improvements.

[0005] In recent years, with the development of artificial intelligence technology, intelligent modeling methods based on deep learning have received widespread attention. Graph Neural Networks (GNNs) have advantages in processing graph-structured data and can be used to analyze the topological relationships and semantic information of building components in CAD drawings, providing a new solution for automated BIM modeling. In the field of BIM analysis applications for building pipe networks, some scholars have structured the BIM model graphs of building pipe networks and used subgraph matching technology to conduct regulatory checks on the pipe network system. Therefore, for the BIM conversion and modeling of CAD drawings for multi-disciplinary building pipe networks, research based on a deep learning-driven technology based on graph methods to reduce the need for manual intervention and improve modeling efficiency has important engineering application value. Summary of the Invention

[0006] Purpose of the invention: The purpose of the present invention is to propose a method for automatically generating multi-professional pipe network BIM models from CAD drawings based on graph neural networks, which can efficiently and automatically generate multi-professional pipe network BIM models from CAD drawings.

[0007] Technical solution: The present invention includes: pipe network CAD drawing information analysis and multi-disciplinary pipe network BIM model generation. The pipe network CAD drawing information analysis includes graph neural network model construction, CAD drawing analysis based on the graph neural network model, legend table analysis and multi-graph information aggregation; the multi-disciplinary pipe network BIM model generation is based on the extracted pipe network component information, and adopts different reconstruction strategies for different types of pipe network components.

[0008] The graph neural network model construction specifically includes:

[0009] Step 1: Specify sample labels;

[0010] Step 2: Construct the model input layer, use two separate multi-layer perceptrons MLP to receive the initial node features V and edge features E respectively, and stack them separately into and

[0011] Step 3: Construct the GATv2 stacking layer of the model. The initial GATv2 layer receives the stacked node features V′ and edge features E′. Each subsequent GATv2 layer receives the node features and edge features output by the previous GATv2 layer.

[0012] Step 4: Construct the output layer of the model. An MLP receives the node features V″ output from the GATv2 stacked layer and maps them to the probability distribution vector p of the classification prediction result; the output layer MLP consists of two fully connected layers, connected by a ReLU activation function in the middle, and finally mapped to the category probability vector p.

[0013] When constructing the GATv2 stacking layer of the model, the GATv2 version of the graph attention network is used as the backbone of the graph neural network model.

[0014] The CAD drawing parsing based on the graph neural network model specifically includes:

[0015] Step 1: Split a single drawing file containing multiple drawing sheets into individual drawing sheets;

[0016] Step 2: For all the plan view sheets obtained by dividing the drawing, select points with the same vertical position as the registration points, and adjust the position of each plan view sheet by horizontal translation;

[0017] Step 3: Explode all drawing entities contained in each drawing sheet multiple times until only basic entities remain.

[0018] Step 4: For each exploded plan view, extract the geometric properties of all basic primitives and calculate the topological connection relationship between each primitive.

[0019] Step 5: For each exploded planar image, generate an undirected graph with each basic element as a node and the topological connections between elements as edges;

[0020] Step 6: For each undirected graph corresponding to a floor plan, input it into the constructed graph neural network model and output the building component type classification results of all nodes;

[0021] Step 7: Map the node classification results of the undirected graph back to the plane graph corresponding to the undirected graph to obtain the semantic type classification results of all graph elements.

[0022] The specific process of the legend table parsing is as follows:

[0023] Step 1: Traverse the legend table obtained by drawing segmentation. If it is a CAD table object, execute step 2; otherwise, execute steps 3-4.

[0024] Step 2: Obtain and record the column representing the type and the column representing the legend symbol from the CAD table object, and record the correspondence between the type and the legend symbol;

[0025] Step 3: Get all horizontal and vertical lines from the legend table and build the table structure by the left-turn closed topological faceting algorithm;

[0026] Step 4: Reconstruct the CAD table object by closing the table cells where the text and symbols in the original legend table fall according to the center point position, and perform the same operation as step 2.

[0027] The specific process of multi-graph information aggregation is as follows:

[0028] Step 1: Traverse all the sub-drawings and group them according to the pipe network system they belong to;

[0029] Step 2: For each set of drawings in step 1, search for entities with the entity type of annotation and text;

[0030] Step 3: For all the labeled primitives obtained in step 2, group them according to whether they have topological connections with each other;

[0031] Step 4: Traverse all text primitives obtained in step 2 and link them to the nearest annotation primitive group obtained in step 3;

[0032] Step 5: For each set of drawings in step 1, traverse each drawing, obtain the semantic classification of all the graphics elements, extract all the pipe segment component graphics elements and pipe point / equipment graphics elements, and update the semantic types of the pipe segment component graphics elements and pipe point / equipment graphics elements based on the extraction results of the legend table;

[0033] Step 6: For all horizontal pipe segment primitives obtained in step 5, group these primitives according to whether they have direct or indirect topological connection relationships with each other;

[0034] Step 7: Take all the elbows, reducers, risers, pipe points, and equipment obtained in step 5, and all the horizontal pipe segment groups obtained in step 6. Treat each elbow, reducer, riser, pipe point / equipment, or horizontal pipe segment group as a graph node. If there is a direct topological connection between two graph elements, establish an edge to build the graph structure.

[0035] Step 8: Traverse the annotation primitive group that has been linked to text in step 4, traverse all nodes of the graph structure established in step 7, and check whether the pipe segment primitive group corresponding to the node is topologically connected to any annotation primitive in the annotation primitive group. If so, assign the number to the pipe segment primitive group;

[0036] Step 9: For each group of drawings processed in step 8, traverse the drawings in the group two by two, compare the nodes in their respective corresponding graph structures, find the pipe segment element groups with the same number attribute, and mark them as entities with the same name;

[0037] Step 10: For each set of drawings processed in step 8, traverse the drawings D1 and D2 in the set in pairs, and take the graph structures corresponding to drawings A and B as G1 respectively;

[0038] Step 11: For each set of drawings processed through steps 5 to 10, merge the graph structures corresponding to all drawings based on the nodes that have completed entity matching with the same name, and obtain the aggregate graph structure of the pipe network system corresponding to the set of drawings;

[0039] Step 12: Traverse each pipe network system aggregation graph structure obtained in step 11 and execute steps 13 to 17 respectively;

[0040] Step 13: In the pipe network system aggregation graph structure, treat the riser type nodes as non-connected nodes, search for each connected subgraph in the graph, and unify the height attributes of all nodes in each connected subgraph;

[0041] Step 14: In the pipe network system aggregation graph structure, search for all riser type nodes. If they do not contain a height attribute, search for other pipe segment nodes connected to their upper and lower ends and obtain their height attributes, and assign values to their upper and lower height attributes.

[0042] Step 15: In the pipe network system aggregation graph structure, treat the nodes of the variable diameter pipe type as non-connected nodes, search for each connected subgraph in the graph, and unify the pipe diameter attributes of all nodes in each connected subgraph;

[0043] Step 16: In the pipe network system aggregation graph structure, search for all variable diameter pipe segment type nodes. If they do not contain the pipe diameter attributes at both ends, search for other pipe segment nodes connected to them at both ends and obtain the pipe diameter attributes, and assign values to the pipe diameter attributes at both ends.

[0044] Step 17: In the pipe network system aggregation graph structure, find the specific combination of intersecting pipe segments. If there is no pipe point or equipment node between two nodes that meet the conditions, insert a corresponding node of elbow or tee type, and infer the position and size of the elbow or tee based on the position and size of the pipe segments on both sides. Among them, elbows and tees belong to the pipe point node type.

[0045] The pipe segment components include elbows, diameter-reducing pipe segments, risers and horizontal pipe segments.

[0046] The generation of the multi-disciplinary pipe network BIM model includes modeling of pipe segment components and modeling of pipe points and equipment.

[0047] The modeling process for the pipe segment component is as follows:

[0048] Step 1: Traverse each generated pipe network system aggregation graph structure and obtain all pipe segment type nodes from it;

[0049] Step 2: Generate a circle of corresponding radius as its cross section based on the pipe diameter property. Use the line connecting the center points of the cross sections at both ends of the pipe segment as the scanning path. Stretch the cross section along this path to form a cylinder, which is the 3D model of the pipe segment.

[0050] The modeling process for pipe points and equipment is as follows:

[0051] Step 1: Traverse each generated pipe network system aggregation graph structure, obtain all pipe points and equipment type nodes, and read their semantic type attributes;

[0052] Step 2: For the tube points with simple geometric shapes obtained in step 1, a parametric modeling method is used to generate a stretched body model, and the model position is adjusted based on its posture information;

[0053] Step 3: For the pipe points / equipment with complex geometric shapes obtained in step 1, the model replacement method is used to retrieve the corresponding preset model in the model library according to its semantic type, and the preset model is transformed based on its size information and posture information to obtain the corresponding pipe point / equipment model.

[0054] Beneficial effects: The present invention has the following advantages:

[0055] 1) This paper constructs a technical route for automated building pipe network CAD drawing parsing that combines deep learning-driven and knowledge-rule constraints. It extracts symbolic primitive semantics based on graph neural networks, achieves highly accurate automatic recognition of symbols and layers, converts pipe network systems represented by multiple CAD drawings into graph structures, and performs entity matching and information aggregation across multiple drawings and graph structures, enabling automatic, efficient, and stable reconstruction of building pipe network system information models.

[0056] 2) The automated BIM modeling technology for multi-disciplinary pipe network systems is based on and extends the IFC (Industry Foundation Classes) standard. It adopts different modeling strategies, including parametric modeling and model replacement, for different disciplines and types of pipe network component entities. This outputs complete BIM data for multi-disciplinary building pipe networks. The modeling results strongly support the management and maintenance of urban building facilities.

[0057] 3) It significantly improves the intelligence level of BIM model generation of building pipe networks, reduces dependence on complex knowledge rule bases and manual intervention, reduces system maintenance and labor costs, and improves modeling efficiency. Compared with traditional human-computer interaction modeling, the present invention can efficiently and accurately complete the automatic construction of BIM models of building pipe networks, and has broad application value in the fields of construction engineering informatization, digital twins, and city information models (CIM). BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is a flow chart of the present invention;

[0059] Figure 2 Network architecture for graph node classification and recognition tasks;

[0060] Figure 3 Annotate primitive aggregation;

[0061] Figure 4 Generate a single graph structure of the pipe network system and match the graph nodes;

[0062] Figure 5 Pipeline network system single sheet conversion diagram structure;

[0063] Figure 6 Illustration of step 20;

[0064] Figure 7 Multi-image aggregation structure of the pipe network system;

[0065] Figure 8 Illustration of riser height calculation;

[0066] Figure 9 Illustration of missing pipe fitting points;

[0067] Figure 10 Complete pipeline system aggregation graph structure;

[0068] Figure 11 Generation of pipe segment 3D model;

[0069] Figure 12 Bend pipe fittings generation diagram;

[0070] Figure 13 Equipment model replacement and adjustment. DETAILED DESCRIPTION

[0071] The present invention will be further described below with reference to the accompanying drawings.

[0072] like Figure 1 As shown, this embodiment of the method automatically generates a multi-disciplinary pipe network BIM model from CAD drawings based on a graph neural network. This method combines deep learning-driven and knowledge-based rule constraints to construct a graph neural network model to parse and identify drawing primitives and layer organization in pipe network CAD drawings. By converting multiple drawings representing the same pipe network system into graph structures, it performs entity matching and information aggregation across multiple drawings and graph structures, achieving comprehensive utilization of drawing information. Furthermore, it infers and completes missing information in the drawings, automatically generating a complete and realistic building pipe network BIM model. This method includes parsing pipe network CAD drawing information and generating a multi-disciplinary pipe network BIM model.

[0073] During the pipeline network CAD drawing information parsing and processing phase, a graph neural network-based approach parses the pipeline network CAD drawings, identifies pipeline component symbols, and extracts information such as their location, geometry, and attributes. This phase is divided into four sub-phases: graph neural network model construction, graph neural network-based CAD drawing parsing, legend table parsing, and multi-image information aggregation.

[0074] The specific process of building a graph neural network model is as follows:

[0075] Step 1: Specify sample labels, i.e., semantic annotations of CAD entities, including pipe segments (including elbows, reducers, risers, and horizontal segments), pipe points, equipment, and annotations (including index annotations, leader annotations, dimension annotations, height annotations, and slope annotations);

[0076] Step 2: Use the GATv2 version of the Graph Attention Network (GAT) as the backbone of the graph neural network model to build a graph neural network model. Its architecture is as follows: Figure 2 As shown, it includes three modules: model input layer, GATv2 stacking layer and output layer. The specific construction method of each module is described in steps 3 to 5.

[0077] Step 3: Construct the model input layer, use two separate multilayer perceptrons (MLP) to receive the initial node features V and edge features E respectively, and stack them into and The two MLPs in the input layer have the same structure, each consisting of two fully connected layers connected by a ReLU activation function. For the input value x, the ReLU function is calculated as follows:

[0078] ReLU(x)=max(0,x)

[0079] Step 4: Construct the GATv2 stacking layer of the model. The initial GATv2 layer receives the stacked node features V′ and edge features E′. Each subsequent GATv2 layer receives the node features and edge features output by the previous GATv2 layer.

[0080] Step 5: Construct the output layer of the model. An MLP receives the node features V″ output from the GATv2 stacked layer and maps them to the probability distribution vector p of the classification prediction result. The output layer MLP consists of two fully connected layers, connected by a ReLU activation function in the middle, and finally mapped to the category probability vector p through the Softmax activation function. The Softmax function is often used in the output layer of multi-class classification tasks. It can convert a real number vector into a probability distribution. For any real number vector z = [z1,z2,…,z C], where C is the total number of categories and the Softmax function is defined as:

[0081]

[0082] The specific process of CAD drawing parsing based on the graph neural network model is as follows:

[0083] Step 1: Open the engineering drawing file (.dwg format) in AutoCAD, and split the single architectural drawing file containing multiple drawing sheets into drawing files (.dwg format) for each drawing sheet.

[0084] Step 2: For all the plan view sheets obtained by dividing the drawing, select points with the same vertical position as the registration points, and adjust the position of each plan view sheet by horizontal translation;

[0085] Step 3: Explode all drawing entities contained in each plan view sheet multiple times until the entities only contain line segments, circles, arcs, ellipses, and text;

[0086] Step 4: For each exploded plan view, extract the geometric properties of all its primitives and calculate the topological connection relationship between each primitive;

[0087] Step 5: For each exploded plan view, an undirected graph is generated with each element as a node and the topological connection relationship between elements as an edge. The characteristics of each graph node include the type and geometric properties of the element corresponding to the node, and the characteristics of each graph edge include the type and properties of the topological connection relationship corresponding to the edge.

[0088] Step 6: For each undirected graph corresponding to a plane map sheet, input it into the constructed graph neural network model and output the semantic classification results of all nodes;

[0089] Step 7: Map the node classification results of the undirected graph back to the plane graph corresponding to the undirected graph to obtain the semantic classification results of all graph elements.

[0090] The specific process of legend table parsing is as follows:

[0091] Step 1: Traverse the legend table obtained by drawing segmentation. If it is a CAD table object, execute step 2; otherwise, execute steps 3-4.

[0092] Step 2: Obtain and record the column representing the type and the column representing the legend symbol from the CAD table object, and record the correspondence between the type and the legend symbol;

[0093] Step 3: Get all horizontal and vertical lines from the legend table and build the table structure by the left-turn closed topological faceting algorithm;

[0094] Step 4: Reconstruct the CAD table object by closing the table cells where the text and symbols in the original legend table fall according to the center point position, and perform the same operation as step 2.

[0095] The specific process of multi-graph information aggregation is as follows:

[0096] Step 1: Traverse all the sub-drawings and group them according to the pipe network system they belong to;

[0097] Step 2: For each set of drawings in step 1, search for entities with the entity type of annotation and text;

[0098] Step 3: For all the labeled primitives obtained in step 2, group them according to whether they have topological connections with each other;

[0099] Step 4: Traverse all text primitives obtained in step 2 and link them to the nearest annotation primitive group obtained in step 3, such as Figure 3 As shown;

[0100] Step 5: For each set of drawings in step 1, traverse each drawing, obtain the semantic classification of all the elements in it, extract all pipe segment component elements (including elbows, reducers, risers, and horizontal pipe segments) and pipe point / equipment elements, and update the semantic types of pipe segment component elements and pipe point / equipment elements based on the extraction results of the legend table;

[0101] Step 6: For all horizontal pipe segment primitives obtained in step 5, group these primitives according to whether they have direct or indirect topological connection relationships with each other;

[0102] Step 7: Take all the elbows, reducers, risers, pipe points and equipment obtained in step 5 and all the horizontal pipe segment groups obtained in step 6, and treat a elbow, reducer, riser, pipe point / equipment or a horizontal pipe segment group as a graph node. If there is a direct topological connection between two graph elements (groups), establish an edge to build a graph structure as follows: Figure 4 and Figure 5 shown.

[0103] Step 8: Traverse the annotation primitive group that has been linked to text in step 4, traverse all nodes of the graph structure established in step 7, and check whether the pipe segment primitive group corresponding to the node is topologically connected to any annotation primitive in the annotation primitive group. If so, assign the number to the pipe segment primitive group, such as Figure 6 As shown;

[0104] Step 9: For each group of drawings processed in step 8, traverse the drawings in the group two by two, compare the nodes in their respective corresponding graph structures, find the pipe segment element groups with the same number attribute, and mark them as entities with the same name;

[0105] Step 10: For each set of drawings processed in step 8, traverse drawings D1 and D2 in the set, taking the graph structures G1 and G2 corresponding to drawings A and B respectively. Search for node A with the number attribute processed in step 9 in G1, and find the corresponding matching node B in G2. If the node A or B that meets the conditions does not exist, continue traversing the next pair of drawings.

[0106] Step 11: Find the next unmatched node C connected to node A and the next unmatched node D connected to node B;

[0107] Step 12: Determine whether the entity attributes of nodes C and D are the same, whether the number of connected edges is the same, and whether the edges have the same characteristics. If all are the same, then nodes C and D are entities with the same name. Otherwise, search for the next unmatched node connected to node B as node D.

[0108] Step 13: After completing the matching of node C and node D, mark the current nodes C and D as the same-name pipe segment group, and use them as node A and node B respectively. Repeat steps 11 and 12 until all nodes are matched. Figure 4 As shown;

[0109] Step 14: For each set of drawings processed through steps 5 to 13, merge the graph structures corresponding to all drawings based on the nodes that have completed entity matching with the same name, and obtain the aggregate graph structure of the pipe network system corresponding to the set of drawings, such as Figure 7 As shown;

[0110] Step 15: Traverse each pipe network system aggregation graph structure obtained in step 14 and execute steps 16 to 20 respectively;

[0111] Step 16: In the pipe network system aggregation graph structure, treat the riser type nodes as non-connected nodes, search for each connected subgraph in the graph, and unify the height attributes of all nodes in each connected subgraph;

[0112] Step 17: In the pipe network system aggregation graph structure, search for all riser type nodes. If they do not contain height attributes, search for other pipe segment nodes connected to their upper and lower ends and obtain their height attributes. Assign values to their upper and lower height attributes, such as Figure 8 As shown;

[0113] Step 18: In the pipe network system aggregation graph structure, treat the nodes of the variable diameter pipe type as non-connected nodes, search for each connected subgraph in the graph, and unify the pipe diameter attributes of all nodes in each connected subgraph;

[0114] Step 19: In the pipe network system aggregation graph structure, search for all variable diameter pipe segment type nodes. If they do not contain the pipe diameter attributes at both ends, search for other pipe segment nodes connected to them at both ends and obtain the pipe diameter attributes, and assign values to the pipe diameter attributes at both ends.

[0115] Step 20: In the pipe network system aggregation graph structure, find the specific combination of intersecting pipe segments. If there is no pipe point or equipment node between the two nodes that meet the conditions, such as Figure 9 As shown, the corresponding insertion type is a node of elbow or tee, and the position and size of the elbow or tee are calculated based on the position and size of the pipe segments on both sides. Among them, elbow and tee belong to the pipe point node type.

[0116] During the processing stage of the multi-disciplinary pipeline network BIM model generation device, different reconstruction strategies are adopted for different types of pipeline network components based on the extracted pipeline network component information, specifically including modeling of pipe section components and modeling of pipe points and equipment.

[0117] The modeling process for pipe segment components is as follows:

[0118] Step 1: Traverse each generated pipe network system aggregation graph structure, such as Figure 10 As shown, all pipe segment type nodes are obtained from it;

[0119] Step 2: Generate a circle of corresponding radius as its cross section according to the pipe diameter property. Use the line connecting the center points of the cross sections at both ends of the pipe section as the scanning path. Stretch the cross section along the path to form a cylinder, which is the 3D model of the pipe section. Figure 11 shown.

[0120] The modeling process for pipe points and equipment is as follows:

[0121] Step 1: Traverse each generated pipe network system aggregation graph structure, obtain all pipe points and equipment type nodes, and read their semantic type attributes;

[0122] Step 2: For the pipe points with simple geometric shapes obtained in step 1, a parametric modeling method is used to generate a stretching body model, and the model position is adjusted based on its posture information. Taking the generation of curved pipes as an example, Figure 12 As shown, the model generation process is described in detail in steps 3 and 4;

[0123] Step 3: For the two pipe segments connected by the elbow fitting, let the two endpoints that are farther apart be P3 and P4, the intersection of the two pipe segments be S, point C be the intersection of the generated pipe fitting arc and line segment SS1, S1 be the center of the arc P1P2, and the calculated result for point S1 is as follows:

[0124]

[0125] Where, |SP2| is the length of line segment SP2, and |SP1+SP2| is the length of the vector added by SP1 and SP2. p1 ,Y p1 ,Z p1 ) and P2(X p2 ,Y p2 ,Z p2 ) is the perpendicular intersection of point S1 with SP3 and SP4. Their coordinates can be calculated as follows:

[0126]

[0127] In a pipe bend, S1 is the center of the arc corresponding to the pipe fitting, d is the diameter of the pipe segment, and S is the vertical distance from the center point S1 to the pipe segment. Draw an arc with point S1 as the center and |S1P1| as the radius. Point C is the intersection of line SS1 and the arc. Since the length of |CS1| is equal to |S1P1|, the coordinates of point C are calculated as follows:

[0128]

[0129] Step 4: Let the endpoints of the pipe segment be P3(X p3 ,Y p3 ,Z p3 ) and P4(X p4 ,Y p4 ,Z p4 ), point S(X s ,Y s ,Z s ) is the connection point of the two pipe segments, and the length L of |SP1| and |SP2| is calculated as follows:

[0130] L=|SP1|=|SP2|=|S1P2| / tan(θ / 2)=(d / 2+s) / tan(θ / 2)

[0131] Step 5: For the pipe points / equipment with complex geometric shapes obtained in step 1, the corresponding preset model is retrieved from the model library according to its semantic type by using the model replacement method. The preset model is transformed based on its size information and posture information to obtain the corresponding pipe point / equipment model. Taking a valve model as an example, Figure 13As shown, steps 6 to 9 give the calculation process of model replacement and adjustment;

[0132] Step 6: Retrieve the corresponding valve model from the model library based on the device semantics, and adjust the spatial transformation matrix of the model as follows:

[0133] M modelmatrix =M scaling *M rotation *M translation

[0134] Where M modelmatrix is the transformation matrix applied to the 3D model, M scaling is the scaling matrix, M rotation is the rotation matrix, M translation is the translation matrix.

[0135] Step 7: Calculate the scaling matrix as follows:

[0136]

[0137] Where d is the adjusted size parameter, corresponding to Figure 13 The diameter of the middle pipe sections A and B; d1 is the size parameter before adjustment, corresponding to Figure 13 The initial interface size parameters of a valve model;

[0138] Step 8: Calculate the rotation matrix as follows:

[0139]

[0140] Where α, β, and γ are the angles between the straight line formed by the model center and the origin and the X-axis, Y-axis, and Z-axis, respectively;

[0141] Step 9: Calculate the translation matrix as follows:

[0142]

[0143] Where x p1 、y p1 and z p1 They are Figure 13 The displacement of the valve model on the X-axis, Y-axis and Z-axis when it is translated from its initial model center point to the midpoint of the line connecting the center points of pipe segments A and B at one end connected to the valve.

Claims

1. A method for automatically generating a multi-disciplinary pipe network BIM model from a CAD drawing based on a graph neural network, characterized in that: include: Pipeline network CAD drawing information analysis and multi-disciplinary pipeline network BIM model generation. The pipeline network CAD drawing information analysis includes graph neural network model construction, CAD drawing analysis based on the graph neural network model, legend table analysis and multi-graph information aggregation; the multi-disciplinary pipeline network BIM model generation is based on the extracted pipeline network component information and adopts different reconstruction strategies for different types of pipeline network components.

2. The method for automatically generating a multi-disciplinary pipe network BIM model from a CAD drawing based on a graph neural network according to claim 1 is characterized in that: The graph neural network model construction specifically includes: Step 1: Specify sample labels; Step 2: Construct the model input layer, use two separate multi-layer perceptrons MLP to receive the initial node features V and edge features E respectively, and stack them separately into and Step 3: Construct the GATv2 stacking layer of the model. The initial GATv2 layer receives the stacked node features V ′ and edge feature E ′ , each subsequent GATv2 layer receives the node features and edge features output by the previous GATv2 layer. Step 4: Construct the output layer of the model, an MLP receives the node features V output from the GATv2 stacked layer ″ , and maps it to the probability distribution vector p of the classification prediction result; the output layer MLP consists of two fully connected layers, connected by the ReLU activation function in the middle, and finally mapped to the probability vector p of the category.

3. The method for automatically generating a multi-disciplinary pipe network BIM model from a CAD drawing based on a graph neural network according to claim 2 is characterized in that: When constructing the GATv2 stacking layer of the model, the GATv2 version of the graph attention network is used as the backbone of the graph neural network model.

4. The method of automatically generating a multi-disciplinary pipe network BIM model from a CAD drawing based on a graph neural network according to claim 2, characterized in that: The CAD drawing parsing based on the graph neural network model specifically includes: Step 1: Split a single drawing file containing multiple drawing sheets into individual drawing sheets; Step 2: For all the plan view sheets obtained by dividing the drawing, select points with the same vertical position as the registration points, and adjust the position of each plan view sheet by horizontal translation; Step 3: Explode all drawing entities contained in each drawing sheet multiple times until only basic entities remain. Step 4: For each exploded plan view, extract the geometric properties of all basic primitives and calculate the topological connection relationship between each primitive; Step 5: For each exploded planar image, generate an undirected graph with each basic element as a node and the topological connections between elements as edges; Step 6: For each undirected graph corresponding to a floor plan, input it into the constructed graph neural network model and output the building component type classification results of all nodes; Step 7: Map the node classification results of the undirected graph back to the plane graph corresponding to the undirected graph to obtain the semantic type classification results of all graph elements.

5. The method of automatically generating a multi-disciplinary pipe network BIM model from a CAD drawing based on a graph neural network according to claim 4 is characterized in that: The specific process of the legend table parsing is as follows: Step 1: Traverse the legend table obtained by drawing segmentation. If it is a CAD table object, execute step 2; otherwise, execute steps 3-4. Step 2: Obtain and record the column representing the type and the column representing the legend symbol from the CAD table object, and record the correspondence between the type and the legend symbol; Step 3: Get all horizontal and vertical lines from the legend table and build the table structure by the left-turn closed topological faceting algorithm; Step 4: Reconstruct the CAD table object by closing the table cells where the text and symbols in the original legend table fall according to the center point position, and perform the same operation as step 2.

6. The method for automatically generating a multi-disciplinary pipe network BIM model from a CAD drawing based on a graph neural network according to claim 5, characterized in that: The specific process of multi-graph information aggregation is as follows: Step 1: Traverse all the sub-drawings and group them according to the pipe network system they belong to; Step 2: For each set of drawings in step 1, search for entities with the entity type of annotation and text; Step 3: For all the labeled primitives obtained in step 2, group them according to whether they have topological connections with each other; Step 4: Traverse all text primitives obtained in step 2 and link them to the nearest annotation primitive group obtained in step 3; Step 5: For each set of drawings in step 1, traverse each drawing, obtain the semantic classification of all the elements in it, extract all the pipe segment component elements and pipe point / equipment elements, and update the semantic types of the pipe segment component elements and pipe point / equipment elements based on the extraction results of the legend table; Step 6: For all horizontal pipe segment primitives obtained in step 5, group these primitives according to whether they have direct or indirect topological connection relationships with each other; Step 7: Take all the elbows, reducers, risers, pipe points, and equipment obtained in step 5, and all the horizontal pipe segment groups obtained in step 6. Treat each elbow, reducer, riser, pipe point / equipment, or horizontal pipe segment group as a graph node. If there is a direct topological connection between two graph elements, establish an edge to build the graph structure. Step 8: Traverse the annotation primitive group that has been linked to text in step 4, traverse all nodes of the graph structure established in step 7, and check whether the pipe segment primitive group corresponding to the node is topologically connected to any annotation primitive in the annotation primitive group. If so, assign the number to the pipe segment primitive group; Step 9: For each group of drawings processed in step 8, traverse the drawings in the group two by two, compare the nodes in their respective corresponding graph structures, find the pipe segment element groups with the same number attribute, and mark them as entities with the same name; Step 10: For each set of drawings processed in step 8, traverse the drawings D1 and D2 in the set in pairs, and take the graph structures corresponding to drawings A and B as G1 respectively; Step 11: For each set of drawings processed through steps 5 to 10, merge the graph structures corresponding to all drawings based on the nodes that have completed entity matching with the same name, and obtain the aggregate graph structure of the pipe network system corresponding to the set of drawings; Step 12: Traverse each pipe network system aggregation graph structure obtained in step 11 and execute steps 13 to 17 respectively; Step 13: In the pipe network system aggregation graph structure, treat the riser type nodes as non-connected nodes, search for each connected subgraph in the graph, and unify the height attributes of all nodes in each connected subgraph; Step 14: In the pipe network system aggregation graph structure, search for all riser type nodes. If they do not contain a height attribute, search for other pipe segment nodes connected to their upper and lower ends and obtain their height attributes, and assign values to their upper and lower height attributes. Step 15: In the pipe network system aggregation graph structure, treat the nodes of the variable diameter pipe type as non-connected nodes, search for each connected subgraph in the graph, and unify the pipe diameter attributes of all nodes in each connected subgraph; Step 16: In the pipe network system aggregation graph structure, search for all variable diameter pipe segment type nodes. If they do not contain the pipe diameter attributes at both ends, search for other pipe segment nodes connected to them at both ends and obtain the pipe diameter attributes, and assign values to the pipe diameter attributes at both ends. Step 17: In the pipe network system aggregation graph structure, find the specific combination of intersecting pipe segments. If there is no pipe point or equipment node between two nodes that meet the conditions, insert a corresponding node of elbow or tee type, and infer the position and size of the elbow or tee based on the position and size of the pipe segments on both sides. Among them, elbows and tees belong to the pipe point node type.

7. The method of automatically generating a multi-disciplinary pipe network BIM model from a CAD drawing based on a graph neural network according to claim 6, characterized in that: The pipe segment components include elbows, diameter-reducing pipe segments, risers and horizontal pipe segments.

8. The method of automatically generating a multi-disciplinary pipe network BIM model from a CAD drawing based on a graph neural network according to claim 1, characterized in that: The generation of the multi-disciplinary pipe network BIM model includes modeling of pipe segment components and modeling of pipe points and equipment.

9. The method for automatically generating a multi-disciplinary pipe network BIM model from a CAD drawing based on a graph neural network according to claim 8, characterized in that: The modeling process for the pipe segment component is as follows: Step 1: Traverse each generated pipe network system aggregation graph structure and obtain all pipe segment type nodes from it; Step 2: Generate a circle of corresponding radius as its cross section based on the pipe diameter property. Use the line connecting the center points of the cross sections at both ends of the pipe segment as the scanning path. Stretch the cross section along this path to form a cylinder, which is the 3D model of the pipe segment.

10. The method of automatically generating a multi-disciplinary pipe network BIM model from a CAD drawing based on a graph neural network according to claim 8, characterized in that: The modeling process for pipe points and equipment is as follows: Step 1: Traverse each generated pipe network system aggregation graph structure, obtain all pipe points and equipment type nodes, and read their semantic type attributes; Step 2: For the tube points with simple geometric shapes obtained in step 1, a parametric modeling method is used to generate a stretched body model, and the model position is adjusted based on its posture information; Step 3: For the pipe points / equipment with complex geometric shapes obtained in step 1, the model replacement method is used to retrieve the corresponding preset model in the model library according to its semantic type, and the preset model is transformed based on its size information and posture information to obtain the corresponding pipe point / equipment model.

Images