BIM automatic modeling optimization method based on structured data

Abstract

The invention discloses a BIM automatic modeling optimization method based on structured data, and the method comprises the following steps: S1, constructing a BIM modeling system based on a cloud platform, supporting the input and processing of the structured data and the generation of a model, and integrating the structured data to a BIM tool; s2, using a bimodal fusion intelligent identification method and a flood filling method to extract information from the vector drawing, and converting the information into structured data; s3, based on a parametric modeling technology, designing a rule-driven modeling algorithm to build a model, and quickly generating building components of the BIM model; and S4, according to the design logic and the dependency relationship, carrying out automatic modeling according to a sequence of structural columns, structural beams, structural walls, building wall bodies, building rooms, building doors and windows, parking spaces, electromechanical water pipes, electromechanical air pipes, electromechanical bridges, electromechanical equipment and electromechanical point locations, optimizing a modeling process, and improving model precision and generation efficiency. According to the invention, the working efficiency and the modeling accuracy are improved.

Description

Technical Field

[0001] This invention belongs to the field of BIM modeling technology, specifically relating to a BIM automatic modeling optimization method based on structured data. Background Technology

[0002] Building Information Modeling (BIM), as a core technology for building lifecycle management, integrates geometric, physical, and functional data to achieve efficient integration of multi-disciplinary collaborative design, construction, and operation and maintenance. The application of BIM technology has significantly improved the overall efficiency and quality of building projects, becoming an indispensable technological tool in the modern construction industry. However, the traditional BIM modeling process heavily relies on manual operation and has the following shortcomings:

[0003] 1. Inefficiency: Relying on manual input of geometric and attribute information not only leads to low modeling efficiency but also easily results in poor data consistency and insufficient optimization capabilities. For example, in large and complex architectural projects, designers need to spend a lot of time creating and adjusting building components one by one, resulting in long modeling cycles and difficulty in meeting the needs of rapid design and change. Although some parametric modeling tools can quickly generate building components through predefined rules and parameters, existing tools usually require users to have a high level of technical skills and experience, and still have limitations when dealing with complex geometries and multidisciplinary collaborative design.

[0004] 2. Poor data consistency

[0005] In manual modeling, modifications and updates to the model by different professionals at different stages can easily lead to data inconsistencies. For example, there may be differences in geometric dimensions and material properties between the model in the design phase and the model in the construction phase. This not only increases the complexity of project management but may also lead to construction errors and increased costs. Although rule-based reasoning techniques can automatically generate models that conform to design specifications using knowledge bases, maintaining data consistency remains a challenge in multi-disciplinary collaboration and dynamic change scenarios.

[0006] 3. Insufficient optimization capabilities

[0007] Existing optimization algorithms suffer from low accuracy when dealing with complex geometric relationships and multiple constraints, failing to meet the requirements of high-precision modeling. In Building Information Modeling (BIM), geometric and parametric accuracy are crucial. Insufficient optimization accuracy in existing algorithms can lead to inaccurate connections between components and unreasonable spatial layouts, impacting the model's usability and the accuracy of subsequent construction.

[0008] In summary, existing BIM modeling technologies have many shortcomings in terms of efficiency, data consistency, and optimization capabilities. Summary of the Invention

[0009] The purpose of this invention is to provide a BIM automatic modeling optimization method based on structured data, which improves work efficiency and modeling accuracy.

[0010] To achieve the above objectives, the present invention adopts the following technical solution:

[0011] A method for automatic BIM modeling optimization based on structured data includes the following steps:

[0012] S1: Construct a cloud-based BIM modeling system that supports the input, processing, and model generation of structured data, and integrate the BIM modeling system into BIM tools to provide a visual interface that allows users to define modeling rules, parameters, and constraints.

[0013] S2: Based on the BIM modeling system, geometric and textual information are extracted from vector drawings using a dual-modal fusion intelligent recognition method, key geometric information related to architectural space identification is extracted from vector drawings using a flood filling method, and then converted into structured data.

[0014] S3: Based on parametric modeling technology, design a rule-driven modeling algorithm to build a model, and use the structured data in S2 to quickly generate building components of the BIM model. The building components include standard components, non-standard components, and combinations of the two.

[0015] S4: Adopting an adaptive optimization algorithm, it automatically models the structural columns, structural beams, structural walls, building walls, building rooms, building doors and windows, parking spaces, electromechanical water pipes, electromechanical air ducts, electromechanical cable trays, electromechanical equipment, and electromechanical points in sequence according to the design logic and dependencies. This optimizes the modeling process and improves model accuracy and generation efficiency.

[0016] The beneficial effects of this invention are:

[0017] (i) High efficiency: Based on parametric modeling technology, the design of rule-driven modeling algorithms to build models realizes automated modeling algorithms, which significantly reduces manual intervention and shortens the modeling time by more than 50% compared with traditional methods. It is particularly suitable for complex projects and large-scale buildings.

[0018] (ii) Flexibility: It adopts a dynamic adjustment mechanism to support real-time response to design changes, adapt to diverse design needs, and is applicable to different types of buildings, such as residential buildings and commercial complexes. It meets the personalized needs of complex projects and improves the efficiency of design exploration.

[0019] (III) Standardization and Accuracy: The rule-based reasoning module ensures that the model conforms to industry standards, such as IFC and GB standards, with a geometric conflict detection accuracy of over 95%. The adaptive optimization algorithm improves the modeling accuracy of non-standard components and complex geometric structures, reducing the model error rate to below 5%, which is significantly better than traditional methods.

[0020] (iv) Data consistency and interoperability: Data preprocessing design addresses compatibility issues with heterogeneous data sources, improving data consistency to 98% and reducing information silos. It supports multi-platform data exchange, reducing data loss rates in cross-software collaboration, improving project collaboration efficiency, and lowering the learning curve for new technologies.

[0021] In summary, this invention significantly improves the efficiency of BIM application in design, construction, and operation and maintenance, providing a reliable digital solution for complex projects. By reducing modeling time and improving model quality, it lowers project costs and enhances the overall efficiency of building lifecycle management, demonstrating significant theoretical and practical value. Attached Figure Description

[0022] Figure 1 This is a flowchart of a specific embodiment 1 of the present invention;

[0023] Figure 2 This is a flowchart of the dual-modal fusion intelligent recognition method in specific embodiment 1 of the present invention;

[0024] Figure 3 This is a flowchart of the process of building the CG recognition model in specific embodiment 1 of the present invention;

[0025] Figure 4 This is a flowchart of building an AI recognition model in specific embodiment 1 of the present invention;

[0026] Figure 5 This is a flowchart of the flooding method in specific embodiment 1 of the present invention. Detailed Implementation

[0027] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0028] Example 1:

[0029] See Figures 1 to 5 As shown, a BIM automatic modeling optimization method based on structured data includes the following steps:

[0030] Step S1: Construct a cloud-based BIM modeling system that supports the input, processing, and model generation of structured data. Integrate this system into BIM tools, providing a visual interface that allows users to define modeling rules, parameters, and constraints. Specifically, this includes:

[0031] S101: Develop a cloud-based BIM modeling system that supports the input, processing, and model generation of structured data, and integrate the BIM modeling system into BIM tools to achieve seamless integration and facilitate user use in existing work environments;

[0032] Cloud platform development: Develop a cloud-based BIM modeling system that supports the input, processing, and model generation of structured data, and provides powerful computing capabilities and storage resources;

[0033] System Integration: Integrate the BIM modeling system into existing BIM tools such as Revit, ArchiCAD, and SketchUp. Support multi-platform data exchange to achieve seamless integration and facilitate user use in existing work environments.

[0034] Cloud platform performance optimization: Optimize the performance of the cloud platform to ensure the stability and response speed of the BIM modeling system and improve the user experience.

[0035] S102: Design a user-visualized interface that allows users to define modeling rules, parameters, and constraints, develop interactive functions, and support users in viewing and adjusting the model in real time.

[0036] User interface design: Provide a user-friendly visual interface that allows users to define modeling rules, parameters and constraints, reducing the operational threshold and making it easy for non-professional users to use;

[0037] Interactive feature development: Develop rich interactive features to support users in viewing and adjusting the model in real time, improving the convenience and flexibility of user operation.

[0038] Step S2: Based on the BIM modeling system, extract geometric and textual information from vector drawings using a dual-modal fusion intelligent recognition method; extract key geometric information related to architectural space identification from vector drawings using a flood filling method, and convert it into structured data.

[0039] The method of extracting geometric and textual information from vector graphics using a dual-modal fusion intelligent recognition method specifically includes:

[0040] S201: Construct a CG recognition model for extracting geometric features of components in architectural vector drawings. Use this CG recognition model to extract and parse information about components with strong geometric characteristics but weak semantic meaning in the architectural vector drawings to be identified, obtaining the first set of component recognition data. Specific steps:

[0041] S2011: Extract geometric element information of components with strong geometric characteristics and weak semantic meaning from vector drawings. These components include building walls, structural walls, structural columns, structural beams, electromechanical water pipes, electromechanical air ducts, and electromechanical cable trays.

[0042] S2012: Utilizing geometric principles, a CG recognition model is built using the CG computer graphics algorithm corresponding to the components in S2011. CG computer graphics algorithms are designed for building walls, structural walls, structural columns, structural beams, electromechanical water pipes, electromechanical air ducts, and cable trays.

[0043] The CG computer graphics algorithms used for the building walls and the structural walls are the same:

[0044] Step 1: Geometric element collection

[0045] In architectural engineering vector drawings, collect geometric elements related to building walls and structural walls, including four categories: straight lines, arcs, polylines, and block references;

[0046] Step 2: Decomposition of Geometric Elements

[0047] The collected geometric elements, categorized into four types—straight lines, arcs, polylines, and block references—are all decomposed into the two most basic geometric shapes, straight lines and arcs, using the Explode API.

[0048] Step 3: Geometric Set Decomposition

[0049] All straight lines are grouped into a set, and then split into multiple new sets of straight lines according to the collinearity algorithm; all arcs are grouped into a set, and then split into multiple new sets of arcs according to the collinearity algorithm.

[0050] Collinearity Algorithm: Traverse all lines. Using the collinearity algorithm of three points on a line with a tolerance of 50, obtain all collinear lines. Then, using the collinearity distance algorithm, filter out lines whose distance does not meet the distance limit disx according to the distance disx between collinear lines. The distance limit disx for building walls is 1800mm, and the distance limit disx for structural walls is 1500mm. Based on this line, recursively loop until all lines are found, forming a set of collinear lines.

[0051] The algorithm for collinearity of three points on a straight line is as follows: Obtain the starting points sp1 and sp2 and the ending points ep1 and ep2 of the two lines to be judged. If the absolute values ​​of the differences between the pairwise vectors of the three points sp1, sp2, ep1 and sp1, sp2, ep2 and sp2, sp2, ep2 are all less than the tolerance value, then they are collinear.

[0052] Collinear line distance algorithm: Obtain the midpoints cp1 and cp2 of two collinear lines to be determined, and the line lengths length1 and length2. Subtract half of the sum of length1 and length2 from the distance between cp1 and cp2 to obtain the distance value dis. If dis is positive, the distance result is dis; if the value is negative, the distance result is 0.

[0053] Collinear Arc Algorithm: Traverse all arcs. Using the collinearity algorithm of arc center with a tolerance of 10mm, obtain all collinear arcs. Then, using the straight-line distance algorithm of arcs, filter out arcs whose distance does not meet the distance limit disx requirement (disx is 10mm) based on the distance between collinear arcs to obtain collinear arcs. Recursively loop based on these arcs until all arcs are found, forming a set of collinear arcs.

[0054] Collinearity algorithm for arc centers: Obtain the centers op1 and op2 and radii r1 and r2 of the two arcs to be determined. When the absolute value of the difference between op1 and op2 and the absolute value of the difference between r1 and r2 are both less than the tolerance value, they are collinear.

[0055] Arc-to-line distance algorithm: Take the starting point sp or ending point ep of one arc, calculate its projection point tp with the other arc, and the distance between the two points is the distance dis between the two arcs.

[0056] Step 4: Merging Collinear Sets

[0057] Each set of collinear lines is merged into a new line using a line merging algorithm; multiple sets of collinear lines result in multiple merged new lines, which are then combined into a new set of lines. Similarly, each set of collinear arcs is merged into a new arc using an arc merging algorithm; multiple sets of collinear arcs result in multiple merged new arcs, which are then combined into a new set of arcs.

[0058] Line merging algorithm: Take the starting points sp1, sp2... and ending points ep1, ep2... of all lines to form a point set. Use bubble sort to obtain the two farthest points p1 and p2. Use these two points to generate a new line, which is the merged line.

[0059] Arc merging algorithm: Take the starting angles sAngle1, sAngle2... and the ending angles eAngle1, eAngle2... of all arcs to form an angle set. Use bubble sort to obtain the maximum and minimum angle values ​​angleMax and angleMin. Use the center op0 and radius r0 of one of the arcs as the center op and radius radiu, and the maximum and minimum angle values ​​angleMax and angleMin as the starting and ending angles to generate a new arc, i.e., the merged arc.

[0060] Step 5: Constructing the binary tuple

[0061] From the merged set of lines, the nearest parallel line for each line is found using the nearest parallel line algorithm, forming a set of binary data; from the merged set of arcs, the nearest parallel line for each arc is found using the nearest parallel line algorithm, forming a set of binary data.

[0062] The algorithm for finding the closest parallel line is as follows: Traverse all lines, use the line parallel vector algorithm with a tolerance of 1° to obtain all parallel lines, then calculate the distance between all parallel lines and the distance between the center points of any pair of lines, and use bubble sort to find the closest parallel line.

[0063] Line parallel vector algorithm: Obtain vectors v1 and v2 of the two lines to be judged, calculate the angle between them, and if the absolute value of the angle is less than the tolerance value, or if the absolute value of the angle between the angle and 180° is less than the tolerance value, then the two lines are considered to be parallel.

[0064] The algorithm for finding the closest parallel line to an arc is as follows: Traverse all arcs, and obtain all parallel arcs by using the distance between their centers (with a tolerance of 10mm). Then calculate the distance between all parallel arcs and the distance between the center points of any two arcs. Use bubble sort to find the parallel arc that is closest to the center.

[0065] Step 6: Calculate the center line

[0066] The centerline algorithm is used to calculate the centerline line and width of each straight line pair; the centerline arc and width of each arc pair are calculated using the arc centerline algorithm.

[0067] Centerline Algorithm: Take two lines from the pair of lines, select the shorter line, and project its starting point sp and ending point ep onto another line to obtain projection points tsp and tep. Take the midpoint cp1 between the starting point sp and its projection point tsp as the starting point sp0 of the centerline, and take the midpoint cp2 between the ending point ep and its projection point tep as the ending point ep0 of the centerline. A new line is formed by these two center points sp0 and ep0, which is the centerline. The distance between the starting point sp and its projection point tsp is taken as the width w of the centerline.

[0068] Arc centerline algorithm: Take two arcs from the arc pair, select the shorter arc, and obtain its center op for later use. Project its starting point sp and ending point ep onto another arc to obtain projection points tsp and tep. Take the midpoint cp1 between the starting point sp and its projection point tsp as the starting point sp0 of the centerline arc, and take the midpoint cp2 between the ending point ep and its projection point tep as the ending point ep0 of the centerline arc. With the center op as the center of the arc, sp0 and ep0 are the starting and ending points of the arc, respectively. The distance between sp0 and op is the radius, forming a new arc, which is the centerline arc. Take the distance between the starting point sp and its projection point tsp as the width w of the centerline arc.

[0069] Step 7: Filtering based on criteria

[0070] The calculated centerline straight line and width are filtered according to wall filtering rules to obtain centerline straight lines and widths that conform to the characteristics of building walls and structural walls. The calculated centerline arc and width are also filtered according to wall filtering rules to obtain centerline arc and widths that conform to the characteristics of building walls and structural walls.

[0071] Wall filtering rules: The straight line length is greater than 100mm, and the width is not less than wxMin and not greater than wxMax. For building walls, wxMin is 100mm and wxMax is 400mm; for structural walls, wxMin is 200mm and wxMax is 600mm, and the width value must be a module of 50.

[0072] CG computer graphics algorithm for structural columns:

[0073] Step 1: Geometric element collection

[0074] In architectural engineering vector drawings, collect geometric elements related to structural columns, including five categories: straight lines, arcs, circles, polylines, and block references;

[0075] Step 2: Decomposition of Geometric Elements

[0076] The collected geometric elements, categorized into four types—straight lines, arcs, circles, polylines, and block references—are all decomposed into the three most basic geometric shapes: straight lines, arcs, and circles, using the Explode API.

[0077] Step 3: Geometric Set Decomposition

[0078] All straight lines are grouped into a set, and then split into multiple new sets of straight lines by connecting the beginning and end of the lines; arcs and circles are not split into sets, but are directly grouped into another set for later use.

[0079] Algorithm for connecting the first and last lines: Traverse all lines and use the line connection algorithm with a tolerance of 10mm to obtain all connected lines. Recursively loop based on the obtained lines until all lines are found; forming a set of lines connected end to end.

[0080] Algorithm for connecting straight lines: Obtain sp1, sp2 and endpoints ep1, ep2 of the two lines to be judged. If the distance between sp1 and sp2, or sp1 and ep2, or sp2 and ep1, or ep2 and ep1 is less than the tolerance value, then the two lines are considered to be connected.

[0081] Step 4: Generate the outline

[0082] The set of lines connected end to end obtained in the previous step is used to calculate multiple enclosing rectangular contour lines using the convex hull algorithm; the set of arcs and circles prepared in the previous step are all regenerated into circular contour lines using the center and radius of the circle.

[0083] Step 5: Filtering based on criteria

[0084] All obtained rectangular outlines are filtered according to the rectangular column outline filtering rules to obtain rectangular outlines that conform to the characteristics of a rectangular column structure. All obtained circular outlines are filtered according to the circular column outline filtering rules to obtain circular outlines that conform to the characteristics of a rectangular column structure.

[0085] Rectangular column outline filtering rules: The area of ​​the rectangular outline is not less than 0.04m² and not more than 4m²; and the ratio of the shortest side to the longest side is not less than 0.18.

[0086] Rectangular column outline filtering rules: The diameter of the circular outline shall not be less than 100mm and not more than 2000mm.

[0087] CG computer graphics algorithm for structural beams:

[0088] The identification of the centerline and width of structural beams is the same as that of building walls and structural walls using CG computer graphics algorithms, so it will not be elaborated here; however, structural beams also need to have their corresponding cross-sectional dimensions and reinforcement information identified.

[0089] The cross-sectional dimensions and reinforcement information of structural beams are all marked using the structural beam planar method. Based on this marking method, the specific identification method is as follows:

[0090] Step 1: Collection of Annotation Elements for Flat Slab Construction

[0091] In architectural engineering vector drawings, collect elements related to the structural beam flat method annotation, including two categories: lines and text; where lines are the leader lines for structural beam flat method annotation, and text is the specific flat method annotation content.

[0092] Step 2: Classification of elements marked using the flat surface method

[0093] The flat surface annotation of structural beams is divided into two types: centralized annotation and in-situ annotation. The centralized annotation is found by using a centralized annotation finding algorithm, and the remaining text is in-situ annotation.

[0094] Centralized annotation self-finding algorithm: Traverse all leader lines and find the corresponding text through the vertical character search algorithm. In this specific embodiment, the angle tolerance is 5° and the distance tolerance is 200mm as an example. The data is stored in dictionary form as dictionary dic1 for later use, where the leader line is the dictionary key value and the text is the dictionary value value.

[0095] Vertical line character finding algorithm: Obtain the vector v1 of the leader line, obtain the character angle to obtain the character direction vector v2, and use the angle between the two vectors. When the absolute value of the difference between angle and 90° or 270° is ≤ angle tolerance of 5°, the leader line is considered to be perpendicular to the character. Take the character perpendicular to the leader line, obtain the position point p1 of the character, and then obtain the projection point p2 from p1 to the leader line. The distance dis between p1 and p2 is used as the distance between the leader line and the character. When this distance dis ≤ distance tolerance of 200mm, the character is considered to be the character corresponding to this leader line.

[0096] Step 3: Centralized element matching

[0097] From the second step, a set of all leader lines is obtained from the dictionary dic1 containing the centralized annotation elements. A leader-based beam-finding algorithm is used to match each leader line with the centerline of the structural beam. In this specific embodiment, an angle tolerance of 5° and a distance tolerance of 200mm are used as an example. This data is stored in dictionary dic2 for later use, with the leader line as the dictionary key and the structural beam centerline as the dictionary value. Using the leader line key values ​​of dictionaries dic1 and dic2 as an intermediary, the structural beam centerline is matched one-to-one with the centralized annotation text to form a new dictionary dic3, with the structural beam centerline as the dictionary key and the centralized annotation text as the dictionary value.

[0098] Lead-line beam finding algorithm: Obtain the starting point sp1 and ending point ep1 of the lead-line and the vector v1. Obtain the centerline vector v2 of the structural beam. When the absolute value of the difference between the angle angle of v1 and v2 and 90° or 270° is ≤ the tolerance angle of 5°, the lead-line is considered to be perpendicular to the centerline of the structural beam. Find the centerline of the structural beam perpendicular to the lead-line. Obtain the projection points tp1 and tp2 of sp1 and ep1 with the centerline of the structural beam. When the distance diS2 between sp1 and tp1 or the distance dis2 between ep1 and tp2 is ≤ the distance tolerance of 200mm, the centerline of the structural beam is considered to be the centerline of the structural beam corresponding to this lead-line.

[0099] Step 4: In-situ annotation element matching

[0100] From the in-situ annotations obtained in the second step, the corresponding text is found using a straight-line parallel character search algorithm. In this specific embodiment, an angle tolerance of 5° and a distance tolerance of 1200mm are used as an example. The text is stored in dictionary data as dictionary dic4 for later use, with the center line of the structural beam as the dictionary key value and the in-situ annotation text as the dictionary value value.

[0101] Parallel Line Character Search Algorithm: Obtain the vector v1 of the structural beam centerline, obtain the character angle to obtain the character direction vector v2, and use the angle between the two vectors (angle) to determine the character's direction. If the absolute value of the difference between angle and 0°, 180°, or 360° is less than or equal to the angle tolerance of 5°, then the structural beam centerline is considered parallel to the character. Take the character parallel to the structural beam centerline, obtain the character's position point p1, and then obtain the projection point p2 from p1 to the structural beam centerline. The distance dis between p1 and p2 is taken as the distance between the structural beam centerline and the character. If this distance dis is less than or equal to the distance tolerance of 1200mm, then the character is considered to correspond to the structural beam centerline.

[0102] Step 5: Annotation element parsing

[0103] The centralized and in-situ annotations of the structural beam centerlines can be obtained from dictionaries dic3 and dic4. The beam number, width, height, and reinforcement information contained in the centralized and in-situ annotation text applied to the centerline of each structural beam can be parsed through the structural beam flat method annotation logic.

[0104] CG computer graphics algorithm for the electromechanical water pipe:

[0105] Step 1: Collection of electromechanical and water pipe elements

[0106] In architectural engineering vector graphics, collect geometric and text elements related to mechanical, electrical, and water pipes, including two categories: lines, polylines, and text.

[0107] Step 2: Elemental Decomposition of Mechanical, Electrical, and Water Pipes

[0108] The collected geometric elements, categorized into three types—lines, polylines, and text—are all broken down into the most basic two elements, lines and text, using the Explode API.

[0109] Step 3: Classification of Mechanical, Electrical, and Water Pipe Elements

[0110] The layers containing the linear and text elements of the electromechanical water pipes are classified; the electromechanical system of the water pipes is obtained by using a layer electromechanical system matching algorithm.

[0111] Layer electromechanical system matching algorithm: Each character of the layer name is broken down and matched one by one with the fixed standard name of electromechanical system. The layer with the most matching characters is the electromechanical system matched by that layer.

[0112] Step 4: Matching of electromechanical and water pipe elements

[0113] The classified electromechanical and water pipe elements are then matched one by one with their corresponding label text using the parallel line matching algorithm described above. In this specific embodiment, taking an angle tolerance of 5° and a distance tolerance of 1800mm as an example, each electromechanical and water pipe line is matched with its corresponding label text.

[0114] Step 4: Analysis of Mechanical, Electrical, and Water Pipe Elements

[0115] By using the labeling logic of the electromechanical water pipe design standards, the size and slope information of each electromechanical water pipe can be analyzed.

[0116] Mechanical and electrical ducts and cable trays:

[0117] Mechanical and electrical ducts and cable trays are designed with double lines in the design drawings, similar to structural beams. However, their labeling method is similar to that of mechanical and electrical water pipes. Therefore, during identification, the center line is first obtained using the same method as for structural beams. The specific method is the same as the CG computer graphics algorithm for structural beams, so it will not be elaborated further. Then, the dimensional data is obtained using the same method as for mechanical and electrical water pipes, and the specific method is the same as the CG computer graphics algorithm for mechanical and electrical water pipes, so it will not be elaborated further.

[0118] S2013: The CG recognition model is used to extract and parse the information of the components with strong geometric characteristics but weak semantic features to obtain the first set of component recognition data.

[0119] This algorithm demonstrates high efficiency and stability when processing large-scale, complex drawing data. It extracts detailed parameter information of geometric components from vector drawings, such as the length, height, thickness, start coordinates, and end coordinates of walls. The extracted geometric information is then preliminarily analyzed and organized to generate structured geometric feature data. Finally, the extracted geometric feature data undergoes quality checks and verification to ensure its accuracy and completeness.

[0120] For efficient processing of complex structures

[0121] To address the complex architectural layouts in the drawings, the CG algorithm was further optimized to enable it to quickly parse the geometric relationships within the drawings.

[0122] Structural decomposition: Using hierarchical processing and recursive decomposition techniques, the complex building structure is decomposed into multiple relatively simple substructures or basic geometric elements, and geometric information is extracted and processed separately.

[0123] Data integration: Integrating and reconstructing the geometric information of the decomposed substructures or basic geometric elements to ensure accurate identification and understanding of complex building structures.

[0124] S202: Build an AI recognition model for recognizing the semantic features of components in architectural vector drawings. Use the AI ​​recognition model to extract and parse the information of components with strong semantic features and weak geometric features in the architectural vector drawings to be recognized, and obtain a second set of component recognition data.

[0125] Detailed operation steps:

[0126] S2021: Select a deep learning model;

[0127] A deep learning framework based on PyTorch + Detectron2 was selected, combined with the Mask R-CNN deep learning model, to construct a recognition model suitable for components with strong semantic features in architectural vector drawings. These components include building doors, windows, parking spaces, rooms, electromechanical equipment, and electromechanical points.

[0128] S2022: Collect massive datasets;

[0129] 1. Collect a dataset of architectural drawings, including images and annotation files in COCO format, and denoise and binarize the images to enhance clarity;

[0130] 2. Register the training and validation sets with the detection and segmentation framework Detectron2, specify the image paths and annotation files, and set the metadata for each category;

[0131] S2023: Model Configuration and Training;

[0132] (1) Dataset partitioning: The preprocessed dataset is divided into a training set, a validation set, and a test set, allocated in a 7:2:1 ratio. The deep learning model is trained using the training set, and the model parameters are continuously adjusted using the backpropagation algorithm. The model is periodically validated using the validation set, and hyperparameters such as the learning rate and optimizer are adjusted based on the validation results. The training set and validation set are registered with the detection and segmentation framework Detectron2, specifying the image paths and annotation files, and setting metadata for each category: such as "door", "window", "parking", "room", "equipment", and "component". When the model's performance on the validation set stabilizes and reaches the expected accuracy, the model is finally evaluated using the test set to ensure that the model has good generalization ability and recognition accuracy.

[0133] Categories will be configured based on building doors, windows, parking spaces, rooms, electromechanical equipment, and electromechanical points. The number of categories is 6. The input resolution is 800x1333, adapted to high-resolution drawings. The learning rate is 0.00025, the maximum number of iterations is 10000, and the learning rate decay step size is 6000 and 8000 iterations. COCO pre-trained weights are loaded to accelerate training. The batch size is set to 2, and the number of samples per image for ROI Heads is 128.

[0134] (2) Use the DefaultTrainer module of Detectron2 for training, which automatically handles data loading, optimizer configuration, and model saving. Observe the classification loss, boundingbox loss, and mask loss, and visualize the training process using the TensorBoard visualization tool. After training, the model weights are saved in their respective output directories. S2024: Inference and Visualization;

[0135] Use the DefaultPredictor function to perform inference on the test image, outputting detection boxes, segmentation masks, and categories. Use Detectron2's Visualizer tool to visualize the detection results on a drawing and save them as an image file.

[0136] S203: Construct a dual-modal collaborative fusion framework

[0137] The first set of component recognition data obtained by the CG recognition model in S201 and the second set of component recognition information obtained by the AI ​​recognition model in S202 are integrated according to the association relationship of building components to form complete and interconnected component recognition data of the building vector drawing.

[0138] Leveraging the strengths of both AI and CG recognition models, a dual-modal fusion strategy is designed to achieve comprehensive and accurate recognition of drawings. This involves combining the advantages of both models. For example, the CG recognition module is first used to extract geometric information, and then the AI ​​recognition model is used to identify semantic information.

[0139] During the drawing recognition process, the CG recognition model and the AI ​​recognition model work together to perform data fusion, association and application based on the structure obtained from the CG recognition model and the AI ​​recognition model modules, so as to achieve efficient recognition of the drawings.

[0140] (1) Data fusion: The second set of component recognition data obtained by the AI ​​recognition model and the second set of component recognition data obtained by the CG recognition model are integrated into a unified structured dataset;

[0141] (2) Data association: Based on the spatial relationship, architectural logic and functional attributes of components, establish the association between components, such as: 1. The subordinate relationship between doors and windows and walls, and the affiliation relationship between rooms and floors; 2. The subordinate and deduction relationship between doors and windows and walls on the same floor; 3. The relationship between building walls, structural columns and structural walls enclosing the outline of rooms on the same floor; 4. The deduction relationship between building walls and structural columns, structural walls and structural beams on the same floor; 5. The vertical alignment relationship of structural columns on adjacent floors; 6. The overlapping and deduction relationship between structural beams, structural columns and structural walls on the same floor; 7. The connection relationship between the beginning and end of mechanical and electrical water pipes of the same system on the same floor; 8. The connection relationship between the beginning and end of mechanical and electrical air ducts of the same system on the same floor; 9. The connection relationship between the beginning and end of mechanical and electrical cable trays of the same system on the same floor.

[0142] The dual-modal collaborative fusion framework supports continuous optimization and adaptation, automatically adjusting and optimizing the system based on new data and feedback.

[0143] Data collection: Establish a feedback mechanism to continuously collect new drawing data and user feedback information.

[0144] System optimization: The dual-modal fusion system is automatically adjusted and optimized based on the new data and feedback collected.

[0145] Pattern mining: Using machine learning and data mining techniques, we analyze and mine a large amount of drawing data and user feedback to discover potential patterns and rules, providing a basis for system optimization and improvement.

[0146] It also includes S204: Creating a data storage architecture that supports structured data storage, specifically including:

[0147] (1) The first set of component identification data extracted in S201 and the second set of component identification data obtained in S202 are respectively processed into structured data and stored as different types of structured data according to the building component category;

[0148] Because the content and length of the identification data for each type of component are different, a custom format (*.cqa) is used. The identified data is stored as a separate cqa file according to the component category. The specific data content is as follows:

[0149] Building walls: Specialty, path points and convexity, width, length, type, name, shape, floor name.

[0150] Building doors and windows: Specialty, location, orientation, direction, angle, door width, door height, type, name, incoming room name, outgoing room name, fire resistance rating, access type, floor name. The incoming room name indicates which room the exit is from, and the outgoing room name indicates which room the exit is to.

[0151] Parking space: specialty, location, angle, path point and convexity, width, length, area, type, name, parking space type, floor name.

[0152] Building room: specialty, location, path point and convexity, area, perimeter, type, name, function, fire protection function, elevation, floor name.

[0153] Structural wall: specialty, path point and convexity, width, length, type, name, shape, floor name.

[0154] Structural column: specialty, outline points and convexity, width, length, area, perimeter, type, name, shape, floor name.

[0155] Structural beams: specialty, path point and convexity, width, height, length, type, name, shape, floor name.

[0156] Mechanical and electrical water pipes: specialty, route points and convexity, diameter, length, mechanical and electrical system, type, name, shape, floor name.

[0157] Mechanical and electrical ductwork: specialty, path points and convexity, width, height, length, mechanical and electrical system, type, name, shape, floor name.

[0158] Mechanical and electrical cable trays: specialty, path points and convexity, width, height, length, mechanical and electrical system, type, name, shape, floor name.

[0159] Mechanical and electrical equipment: specialty, location, angle, type, name, equipment type, floor name.

[0160] Electromechanical points: Specialty, location, angle, type, name, point type, floor name.

[0161] (2) Use sequential storage structure to ensure data integrity and availability.

[0162] (3) Data in each storage file is encrypted separately using a symmetric encryption algorithm. The AES advanced encryption standard symmetric encryption algorithm is used, with the CBC cipher block chaining mode used by default. Files are encrypted using a 128, 192, or 256-bit key and a 16-byte initialization vector (IV), combined with PKCS7 padding to ensure efficient and secure data protection. The following efficiency is achieved:

[0163] 1. Optimization of recognition algorithm in complex drawing environments

[0164] In complex drawing environments, improving recognition accuracy requires comprehensive consideration of multiple factors, such as the diversity of graphic elements, the complexity of text annotations, and the differences in drawing quality. Solving this problem not only requires powerful image processing algorithms but also necessitates striking a balance between the algorithm's robustness and efficiency.

[0165] 2. Highly efficient data storage architecture design

[0166] It provides a data storage architecture that can meet the needs of large-scale data storage and support efficient data retrieval and analysis, solving the problems of data redundancy and lack of correlation.

[0167] 3. Breakthrough in multimodal data fusion technology

[0168] The effective integration of textual and graphic information in vector graphics paper was achieved, which required solving technical challenges such as data format differences, establishing data correlation, and maintaining data consistency. In addition, the accuracy and integrity of the data were guaranteed during the data integration process.

[0169] The flooding method identifies key geometric information related to architectural space from vector graphics.

[0170] S205: Used to classify entity elements in vector graphics by layer, group different types of geometric elements into corresponding layer sets, and filter out key geometric information related to architectural space recognition. Specifically, this includes:

[0171] a. Classify entity elements in the vector drawing by layer, categorizing them into walls, doors and windows, structural columns, and room names. Perform layer classification on the input vector drawing, assigning different types of geometric elements to the corresponding layer sets for easier subsequent processing.

[0172] Layer recognition: Used to read the layer information of the captured vector graphics paper and identify the names and attributes of different layers.

[0173] Layer Classification: Entities in the vector graphics are categorized into four layers: walls, doors and windows, structural columns, and room names, forming four layer sets. For example, layers containing geometric elements related to walls are categorized into the "Walls" layer set, layers containing geometric elements related to doors and windows are categorized into the "Doors and Windows" layer set, layers containing geometric elements related to structural columns are categorized into the "Structural Columns" layer set, and layers containing text elements related to room names are categorized into the "Room Names" layer set.

[0174] Layer reorganization: Reorganize or merge the categorized layers as needed to ensure that key geometric element information is concentrated in the corresponding layer set, which facilitates subsequent processing.

[0175] b. Remove noise data, such as redundant lines and dots, from the vector graphics after the layers have been categorized to reduce interference and improve recognition accuracy.

[0176] Noise detection: Existing geometric analysis algorithms are used to detect abnormal geometric elements in the vector drawing. In this specific embodiment, all geometric elements in the vector drawing are traversed, and detection is performed based on conditions such as length equal to 0, coordinates being infinite, and coordinates being empty, to identify abnormal geometric elements. Abnormal geometric elements include isolated points, extremely short line segments, etc.

[0177] Noise Removal: Based on the test results, delete or correct noise data to ensure the cleanliness of the vector graphics.

[0178] Data smoothing: Smooths the remaining geometric elements to reduce minor fluctuations caused by drawing errors.

[0179] c. Filter out key geometric information related to architectural space identification, including walls, doors and windows, structural columns, and room names, to provide basic data for subsequent flood filling.

[0180] Key element identification: Based on the layer classification results, key geometric elements are identified and filtered out. Specifically, the four categories of walls, doors and windows, structural columns, and room names summarized by the layer classification are used to filter out key geometric component information and text information related to architectural space identification.

[0181] Information extraction: Extract the geometric component information and text information of key geometric elements, namely the geometric information and text information of four categories of entity components, including straight lines and curves of walls; straight lines and curves of doors and windows; straight lines, curves and circles of structural columns; text position and text content.

[0182] Data transformation: The geometric and textual information of each entity component extracted from the information is stored as structured data for easy subsequent processing.

[0183] S206: Divide the building space into multiple independent zones based on key geometric information and automatically select appropriate seed points to improve the robustness and accuracy of the flooding method. Seed point selection should ensure that they are located within independent spatial zones, avoiding cross-zone selection.

[0184] Zone division: Based on key geometric information, such as walls, doors and windows, structural columns, and room names, the building space is divided into multiple independent zones.

[0185] Seed point generation: Based on the division of each independent area, a seed point is automatically selected within each independent area. The seed point uses the room text.

[0186] Seed point verification: Verify the selected seed point to ensure that it is located within an independent spatial region and avoid cross-region selection.

[0187] S207: Based on the selected seed point and custom tolerance value, execute the flood filling algorithm to identify independent spatial regions.

[0188] a. Based on the selected seed points and a custom tolerance value, execute the flood fill algorithm to identify independent spatial regions. The flood fill algorithm gradually expands the seed points to fill the entire spatial region until it encounters a boundary.

[0189] Algorithm initialization: Set the initial parameters of the flood filling algorithm, including seed point and tolerance value.

[0190] Region filling: Starting from a seed point, the filling area is gradually expanded until a boundary is encountered. The boundary can be a key geometric element such as a wall, door, window, or structural column.

[0191] Region marking: Mark the filled area to distinguish different spatial regions.

[0192] b. Adjust the tolerance value of the flood filling algorithm according to actual needs to adapt to different building spaces and geometric features.

[0193] Tolerance Value Setting: Based on the complexity and geometric characteristics of the architectural space, an appropriate tolerance value is set. In this specific embodiment: During the drawing process of architectural engineering design floor plans, design errors are unavoidable due to manual drafting, typically around 10mm. Therefore, setting an appropriate tolerance value is crucial for accurate division and generation of architectural spaces during architectural space identification. In architectural components, wall dimensions are typically between 200mm and 400mm, and structural column dimensions are between 400mm and 1000mm, all using a 50mm module. This modular design provides a reference for tolerance setting. If the tolerance value is too large, exceeding 50mm, the generated room outline may deviate significantly from the actual boundaries of the walls or structural columns, resulting in inaccurate space division and affecting subsequent functionality. Conversely, if the tolerance is too small, less than 10mm, incompletely closed areas in the drawings may prevent the generation of architectural spaces due to errors, or the spatial outline may incorrectly penetrate adjacent areas, causing identification failure or spatial confusion. Taking all factors into consideration, 50mm is a reasonable and effective tolerance value: it not only covers the approximately 10mm error commonly found in manual drafting, ensuring that most non-closed areas are correctly identified, but also aligns with the 50mm module standard for building components, helping to maintain the geometric accuracy and design compliance of spatial identification results. Furthermore, a 50mm tolerance value balances accuracy and robustness in practical applications, avoiding loose boundaries due to excessive tolerance or generation failure due to insufficient tolerance. Therefore, the tolerance value in this embodiment is set to 50mm.

[0194] Tolerance value optimization: Through multiple experiments and verifications, the tolerance value was optimized to ensure the accuracy and robustness of the flood filling algorithm.

[0195] S208: Optimize and reprocess the results of flood filling to remove any possible errors or unreasonable area divisions, ensuring the accuracy and completeness of the identification results.

[0196] A. Optimize the results of flood filling and remove any possible errors or unreasonable area divisions.

[0197] Region merging: Merging adjacent regions with similar characteristics reduces the fragmentation of region division.

[0198] Region segmentation: Excessively large or unreasonable areas are segmented to ensure the rationality and independence of each region. In this specific embodiment, the rules for defining excessively large or unreasonable regions are: 1) The region is not enclosed; 2) The region area is less than 1.5㎡; 3) The region area is greater than 1.5㎡, but the minimum width of the region is less than or equal to 600mm.

[0199] Boundary correction: Correcting the region boundary, that is, optimizing the region boundary by removing intersecting lines, duplicate points, and collinear points.

[0200] b. Verify the optimized results of the region boundaries to ensure the accuracy and completeness of the identification results.

[0201] Consistency check: Check the consistency between the area boundary identification results and the original drawings to ensure that all spatial areas are correctly identified. Display the area boundaries in situ on the drawings for manual verification.

[0202] Integrity check: The integrity of the identification results is checked manually to ensure that no spatial areas are missed.

[0203] Result Correction: Based on the verification results, the identification results obtained from the consistency check and integrity check are corrected to ensure the accuracy and integrity of the final result.

[0204] Step S3: Based on parametric modeling technology, design a rule-driven modeling algorithm to build the model, and quickly generate building components of the BIM model using the structured data processed in S2. These building components include standard components, non-standard components, and combinations of both; specifically including:

[0205] S301: Perform parametric modeling

[0206] Rule definition: Based on the modeling rules of the BIM modeling software, it covers the data of each component obtained in S1, including structural columns, structural beams, structural walls, building walls, building rooms, building doors and windows, parking spaces, electromechanical water pipes, electromechanical air ducts, electromechanical cable trays, electromechanical equipment, and electromechanical points;

[0207] Data-driven modeling: Utilizing data from structured data and the modeling rules of BIM modeling software, building component models can be generated quickly, achieving automated modeling.

[0208] The parametric model is optimized by continuously improving its performance, increasing the speed and accuracy of model generation, and ensuring that the model accurately reflects the design intent.

[0209] S302: Rule-Driven Reasoning

[0210] Building design code integration: Integrate national building standards and building design codes into the modeling rules of the BIM modeling software to ensure that the generated BIM model meets relevant requirements;

[0211] Application of the rule reasoning engine: By utilizing the rule reasoning engine, a BIM model that meets the requirements is automatically generated based on the input structured data and the modeling rules of the BIM modeling software, thereby improving the accuracy and efficiency of modeling.

[0212] Update and maintain the modeling rules of the BIM modeling software: Regularly update and maintain building design codes to ensure that the modeling algorithm always complies with the latest industry standards.

[0213] Step S4: Adaptive optimization algorithm is used to automatically model the structural columns, structural beams, structural walls, building walls, building rooms, building doors and windows, parking spaces, electromechanical water pipes, electromechanical air ducts, electromechanical cable trays, electromechanical equipment and electromechanical points in sequence according to the design logic and dependencies. The modeling process is optimized to improve model accuracy and generation efficiency.

[0214] Structural columns are the core components of a building's load-bearing system, supporting the entire building's load. Automatic modeling must prioritize completing the structural columns to determine the building's framework and geometric relationships.

[0215] Structural beams and columns together form the load-bearing framework of a building, and their position depends on the columns. Beam modeling must follow column modeling to ensure the accuracy of node connections.

[0216] Structural walls, such as shear walls, are load-bearing and lateral force-resisting components that rely on a frame of columns and beams. When modeling them, it is essential to ensure geometric compatibility with the columns and beams.

[0217] Building walls, including non-load-bearing exterior walls and interior partition walls, rely on the completion of the structural framework and are used to enclose spaces. The boundaries of structural walls and columns must be referenced during modeling.

[0218] The definition of building rooms is based on the layout of building walls and must be done after wall modeling. The algorithm optimizes room division according to functional requirements, such as area and ventilation, and provides spatial basis for subsequent door, window and mechanical and electrical layout.

[0219] Building doors and windows are attached to the building walls, and their location and size need to be determined after the walls and rooms are modeled. At the same time, geometric compatibility with the walls must be considered.

[0220] The layout of parking spaces depends on the building's walls and room layout, and is typically located in the basement or a specific area. Structural columns and beams must be considered during modeling.

[0221] Mechanical and electrical pipes, such as water supply and drainage pipes, need to be laid after the main building structure is completed, depending on the spatial distribution of walls, rooms, and parking spaces. The modeling sequence should consider the pipe routing to reduce wall penetrations and conflicts.

[0222] Mechanical and electrical ducts, such as ventilation and air conditioning systems, typically require more space than water pipes. They need to be modeled after water pipes are laid out to avoid space conflicts.

[0223] Mechanical and electrical cable trays, also known as electrical cable supports, depend on the layout of water pipes and air ducts. They need to be modeled after both are completed to utilize the remaining space. Conflicts with the mechanical and electrical systems must be considered during modeling.

[0224] Mechanical and electrical equipment, such as air conditioning units and pumping stations, need to be modeled after the pipeline layout is completed to ensure accurate connection points.

[0225] The electromechanical points are the final components to be modeled, depending on the completion of all electromechanical systems and building spaces.

[0226] This invention provides a new approach and method for solving existing technical problems by employing an automatic modeling and optimization approach based on structured data. Through efficient algorithms and rules, this method can automatically generate high-quality BIM models, significantly improving modeling efficiency and flexibility while ensuring that the models conform to industry standards. This invention also overcomes problems related to technology integration, algorithm optimization, and user adaptability, facilitating its widespread application.

[0227] The technical solution provided by this invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. It should be noted that those skilled in the art can make several improvements and modifications to this invention without departing from the principles of this invention, and these improvements and modifications also fall within the protection scope of the claims of this invention.

Claims

1. A BIM automatic modeling optimization method based on structured data, characterized in that, Comprise the following steps: S1: Construct a cloud platform-based BIM modeling system that supports the input, processing, and model generation of structured data, and integrate the BIM modeling system into BIM tools to provide a visual interface that allows users to define modeling rules, parameters, and constraints; S2: Based on the BIM modeling system, use a dual-modal fusion intelligent recognition method to extract geometric information and text information from vector drawings, use the flood fill method to extract key geometric information related to building space identification from vector drawings, and convert it into structured data; S3: Based on parametric modeling technology, design a rule-driven modeling algorithm to build a model, and use the structured data from S2 to quickly generate building components for the BIM model, including standard components, non-standard components, and combinations of the two; S4: Use an adaptive optimization algorithm to automatically model in the order of structural columns, structural beams, structural walls, building walls, building rooms, building doors and windows, parking spaces, mechanical and water pipes, mechanical and air pipes, mechanical bridges, mechanical equipment, and mechanical points, according to design logic and dependencies.

2. The BIM automatic modeling optimization method based on structured data according to claim 1, characterized in that: The S1 specifically comprises: S101: Develop a cloud platform-based BIM modeling system that supports the input, processing, and model generation of structured data, and integrate the BIM modeling system into BIM tools to achieve seamless integration and facilitate user use in existing work environments; S102: Design a user visual interface that allows users to define modeling rules, parameters, and constraints, and develop interactive functions to support users in real-time viewing and adjusting the model.

3. The BIM automatic modeling optimization method based on structured data according to claim 1, characterized in that: The dual-modal fusion intelligent recognition method used in S2 to extract geometric information and text information from vector drawings specifically comprises: S201: Build a CG recognition model for extracting component geometric features from building vector drawings, use the CG recognition model to extract and analyze component information with strong geometric characteristics and weak semantic significance in the building vector drawings to be recognized, and obtain a first set of component recognition data; S202: Build an AI recognition model for identifying component semantic features in building vector drawings, use the AI recognition model to extract and analyze component information with strong semantic features and weak geometric features in the building vector drawings to be recognized, and obtain a second set of component recognition data; S203: Build a dual-modal collaborative fusion framework, Integrate the first set of component recognition data obtained from the CG recognition model in S201 and the second set of component recognition information obtained from the AI recognition model in S202 into a complete and interrelated component recognition data set for building vector drawings according to the association relationship of building components.

4. The BIM automatic modeling optimization method based on structured data according to claim 3, characterized in that: Also includes S204: Create a data storage architecture that supports structured data storage, specifically comprising: (1) Structurally process the first set of component recognition data extracted in S201 and the second set of component recognition data obtained in S202, and store them as different types of structured data according to building component categories; (2) Store using a sequential storage structure; (3) Encrypt the data of each storage file separately, and use a symmetric encryption algorithm to encrypt the file.

5. The method of claim 3, wherein: The S201 specifically comprises: S2011: Extract geometric element information of components with strong geometric characteristics and weak semantic meaning from vector drawings. These components include building walls, structural walls, structural columns, structural beams, electromechanical water pipes, electromechanical air ducts, and electromechanical cable trays. S2012: Using geometric principles, a CG recognition model is constructed using the CG computer graphics algorithm corresponding to the components in S2011. S2013: The CG recognition model is used to extract and parse the information of the components with strong geometric characteristics but weak semantic features to obtain the first set of component recognition data.

6. The BIM automatic modeling optimization method based on structured data according to claim 3, characterized in that: S202 specifically includes: S2021: Select a deep learning model; A deep learning framework based on PyTorch + Detectron2 was selected, combined with the Mask R-CNN deep learning model, to construct a recognition model suitable for components with strong semantic features in architectural vector drawings. These components include building doors, windows, parking spaces, rooms, electromechanical equipment, and electromechanical points. S2022: Collect massive datasets; 1. Collect a dataset of architectural drawings, including images and annotation files in COCO format, and denoise and binarize the images to enhance clarity; 2. Register the training and validation sets with the detection and segmentation framework Detectron2, specify the image paths and annotation files, and set the metadata for each category; S2023: Model Configuration and Training; 1. Based on the categories of building doors, building windows, parking spaces, building rooms, electromechanical equipment, and electromechanical point configuration, load COCO pre-trained weights to accelerate training; extract and parse the component information with strong semantic features and weak geometric features in the building vector drawings to obtain the second set of component recognition data; 2. Use the DefaultTrainer training module of Detectron2 to train the model, observe the classification loss, bounding box loss and mask loss, and use the TensorBoard visualization tool to visualize the training process. After training is completed, the model weights are saved in the specified output directory. S2024: Reasoning and Visualization; Use the DefaultPredictor function to perform inference on the test image, outputting detection boxes, segmentation masks, and categories. Use Detectron2's Visualizer tool to visualize the detection results on a drawing and save them as an image file.

7. The BIM automatic modeling optimization method based on structured data according to claim 1, characterized in that: The flooding method in S2 extracts key geometric information related to architectural space identification from vector graphics, specifically including: S205: Using vector graphics, classify the entity elements in the vector graphics by layer, categorize different types of geometric elements into the corresponding layer sets, and filter out key geometric information related to architectural space recognition; S206: Divide the building space into multiple independent areas based on key geometric information, and automatically select appropriate seed points. The selection of seed points should ensure that they are located within independent spatial areas, and avoid cross-area selection. S207: Based on the selected seed point and custom tolerance value, execute the flood filling algorithm to separate independent spatial regions; S208: Optimize and reprocess the results of flood filling to remove any errors or unreasonable area divisions, ensuring the accuracy and completeness of the identification results.

8. The BIM automatic modeling optimization method based on structured data according to claim 1, characterized in that: S3 specifically includes: S301: Perform parametric modeling Rule definition: Based on the modeling rules of BIM modeling software, it covers the data of each component including structural columns, structural beams, structural walls, building walls, building rooms, building doors and windows, parking spaces, mechanical and electrical water pipes, mechanical and electrical air ducts, mechanical and electrical cable trays, mechanical and electrical equipment, and mechanical and electrical points; Data-driven modeling: Utilizing data from structured data and the modeling rules of BIM modeling software, building component models can be generated quickly, achieving automated modeling. The parametric model is optimized by continuously improving its performance, increasing the speed and accuracy of model generation, and ensuring that the model accurately reflects the design intent. S302: Rule-Driven Reasoning Building design code integration: Integrate national building standards and building design codes into the modeling rules of the BIM modeling software to ensure that the generated BIM model meets relevant requirements; Application of the rule reasoning engine: By utilizing the rule reasoning engine, a BIM model that meets the requirements is automatically generated based on the input structured data and the modeling rules of the BIM modeling software, thereby improving the accuracy and efficiency of modeling. Update and maintain the modeling rules of the BIM modeling software: Regularly update and maintain building design codes to ensure that the modeling algorithm always complies with the latest industry standards.

Images