Computable structural model reconstruction method based on structural construction drawing raster images

Abstract

The present application relates to the technical field of graphic modeling, in particular to a computable structure model reconstruction method based on structural construction drawing raster images, which comprises the following steps: obtaining structural construction drawing raster image data of a target building, and identifying structural general description drawing raster image data, structural plan drawing raster image data and component reinforcement detail drawing raster image data according to a multi-modal large language model; obtaining single-layer structural drawing data according to the structural plan drawing raster image data, extracting component parameter information according to the component reinforcement detail drawing raster image data, and extracting global material information from the structural general description drawing raster image data; and generating a computable structure model of the target building based on the global material information, the single-layer structural drawing data and the corrected component parameter information, thereby solving the problems of low building structure modeling efficiency and poor applicability of model structure calculation and analysis in related technologies, improving the modeling efficiency and the applicability of the computable structure model.

Description

Technical Field

[0001] This invention relates to the field of graphic modeling technology, and in particular to a method for reconstructing a computable structural model based on raster images of structural construction drawings. Background Technology

[0002] With the rapid development of digital analysis and intelligent design of building structures, constructing computable structural models based on structural construction drawings is the core foundation for realizing building mechanics simulation and safety verification. In many engineering scenarios, structural construction drawings exist in the form of raster images, containing key information such as component layout, reinforcement details, and material requirements. How to efficiently parse these images and generate structural models that meet calculation requirements has become a key technical need in the industry.

[0003] In related technologies, the main approaches rely on manual modeling based on drawings or automated modeling based on vector graphics. Manual modeling requires engineers to extract component parameters and establish topological relationships manually, while methods based on vector graphics can directly parse CAD primitive information to generate models; alternatively, they can attempt to perform simple component contour recognition on raster images.

[0004] However, manual modeling in related technologies is inefficient, error-prone, and difficult to adapt to large-scale projects. Vector graphics modeling methods are incompatible with existing raster image drawings. Raster image analysis can only identify component outlines and cannot construct component topological relationships. The generated models are not suitable for structural calculation and analysis, which urgently needs to be solved. Summary of the Invention

[0005] This invention provides a method for reconstructing a computable structural model based on raster images of structural construction drawings, in order to solve the problems of low efficiency in building structure modeling and poor applicability of model structure calculation and analysis in related technologies, thereby improving modeling efficiency and enhancing the applicability of computable structural models.

[0006] To achieve the above objectives, a first aspect of the present invention provides a method for reconstructing a computable structural model based on raster images of structural construction drawings, comprising the following steps: acquiring raster image data of structural construction drawings of a target building, and generating raster image data of a general structural description drawing, a structural plan drawing, and detailed reinforcement drawings of components based on the raster image data of the structural construction drawings; obtaining single-layer structural drawing data based on the structural plan drawing raster image data, extracting component parameter information through the detailed reinforcement drawings of components, performing cross-sectional geometric verification and position correction on the relative coordinates of the reinforcing bars in the component parameter information to obtain corrected component parameter information, and extracting global material information from the general structural description drawing raster image data; and generating a computable structural model of the target building based on the global material information, the single-layer structural drawing data, and the corrected component parameter information.

[0007] Furthermore, in some embodiments, the step of generating structural general description raster image data, structural plan raster image data, and component reinforcement detail raster image data based on the structural construction drawing raster image data includes: preprocessing the structural construction drawing raster image data to obtain preprocessed structural construction drawing raster images; performing type recognition on the preprocessed structural construction drawing raster images according to preset structured prompt words to obtain structural construction drawing raster image classification results; and outputting the structural construction drawing raster image classification results as structural general description raster image data, structural plan raster image data, and component reinforcement detail raster image data according to a preset structured format.

[0008] Furthermore, in some embodiments, obtaining single-layer structure map data based on the structure planar raster image data includes: performing semantic segmentation on the structure planar raster image data to obtain a semantic segmentation result of the structure planar image; and generating the single-layer structure map data based on the semantic segmentation result of the structure planar image.

[0009] Furthermore, in some embodiments, the step of semantic segmentation of the structural planar raster image data to obtain the semantic segmentation result of the structural planar image includes: locating and masking non-structural primitives in the structural planar raster image data, and combining a preset image inpainting algorithm to smooth the background and complete the connectivity of broken structural lines caused by the masking area to obtain a purified structural base image; performing pixel-level feature extraction and classification on the purified structural base image to obtain a structural semantic segmentation classification result, wherein the structural semantic segmentation classification result is mapped to a preset RGB pixel value to obtain the semantic segmentation result of the structural planar image.

[0010] Further, in some embodiments, generating the single-layer structural diagram data based on the semantic segmentation result of the structural plan view includes: obtaining the column component connected domain, beam component, and wall component skeleton / axis from the semantic segmentation result of the structural plan view; using the geometric centroid of the column component connected domain, the endpoints and intersections of the skeleton / axis of the beam component and the wall component as graph nodes, and constructing an initial graph node set; merging the coordinates of all beam endpoints whose distance from the center of the column component is less than a preset threshold to the corresponding column component center coordinates, using the merged column component center coordinates as a single graph node, and updating the initial graph node set according to the single graph node to obtain a target graph node set; splitting the beam component skeleton based on the target graph node set to obtain split beam segments and wall segments, using the two ends of the split beam segments as graph nodes, constructing a target graph edge set according to the split beam segments, and generating the single-layer structural diagram data.

[0011] Further, in some embodiments, the step of performing cross-sectional geometric verification and position correction on the relative coordinates of the reinforcing bars in the component parameter information to obtain the corrected component parameter information includes: extracting component information from the raster image data of the component reinforcement detail drawing to obtain the parameterized information of the component reinforcement detail drawing, wherein the parameterized information includes the component number, component cross-sectional dimensions, and component reinforcement list; calculating the component cross-sectional boundary and geometric center position based on the component cross-sectional dimensions, and extracting the reinforcing bar diameter, protective layer thickness, and stirrup diameter from the component reinforcement list; determining whether the distance from the center coordinate of the reinforcing bar in the component reinforcement list to the edge of the component cross-section is less than the sum of the protective layer thickness, stirrup diameter, and longitudinal reinforcing bar radius; if it is less, then translating the center coordinate of the reinforcing bar inward along the normal of the component cross-section until the boundary construction requirements are met to obtain the corrected component parameter information.

[0012] Further, in some embodiments, the step of correcting the raster image data of the component reinforcement detail to obtain the corrected component parameter information includes: extracting component information from the raster image data of the component reinforcement detail to obtain parameterized information of the component reinforcement detail, wherein the parameterized information includes component number, component cross-sectional dimensions, and component reinforcement list; determining the component cross-sectional boundary and the geometric center position of the component cross-section based on the component cross-sectional dimensions, and extracting the rebar diameter, protective layer thickness, and stirrup diameter recorded in the component reinforcement list; determining whether the distance from the rebar center coordinates in the component reinforcement list to the edge of the component cross-section is less than the sum of the protective layer thickness, stirrup diameter, and longitudinal rebar radius; if it is less, then translating the rebar center coordinates inward along the component cross-section normal until the geometric center position coordinates of the component cross-section meet the preset boundary construction requirements, and outputting the corrected rebar design position coordinates and the parameterized information together as the corrected component parameter information.

[0013] Furthermore, in some embodiments, the step of extracting global material information from the structural general description diagram raster image data includes: identifying the concrete strength, design standards, and general construction requirements in the structural general description diagram raster image data, and outputting the global material information according to a preset structured format.

[0014] Furthermore, in some embodiments, generating a computable structural model of the target building based on the global material information, the single-layer structural diagram data, and the corrected component parameter information includes: embedding the global material information and the corrected component parameter information into the single-layer structural diagram data according to a preset component number matching method to obtain single-layer structural diagram data with embedded information; performing topological verification and completion on the single-layer structural diagram data with embedded information to obtain single-layer structural diagram data with complete attributes; and performing spatial alignment and combination on the single-layer structural diagram data with complete attributes to generate a computable structural model reconstruction result of the target building.

[0015] The computable structural model reconstruction method based on structural construction drawing raster images provided by the present invention takes the structural construction drawing raster image data of the target building as input, and generates three types of drawing data through classification. Then, it generates single-layer structural drawing data from the structural plan drawing, extracts and corrects component parameter information from the component reinforcement detail drawing, and extracts global material information from the structural general description drawing. Finally, it integrates the three types of data to generate a computable structural model of the target building. This method solves the problems of low efficiency in building structure modeling and poor applicability of model structure calculation and analysis in related technologies, improves modeling efficiency, and enhances the applicability of computable structural models.

[0016] To achieve the above objectives, a second aspect of the present invention provides a computable structural model reconstruction device, comprising: an acquisition module, configured to acquire raster image data of structural construction drawings of a target building, and generate raster image data of a general structural description drawing, a structural plan drawing, and a raster image data of detailed reinforcement drawings of components based on the raster image data of the structural construction drawings; an analysis module, configured to obtain single-layer structural drawing data based on the raster image data of the structural plan drawing, extract component parameter information through the raster image data of the detailed reinforcement drawings of components, perform cross-sectional geometric verification and position correction on the relative coordinates of the reinforcing bars in the component parameter information to obtain corrected component parameter information, and extract global material information from the raster image data of the general structural description drawing; and a construction module, configured to generate a computable structural model of the target building based on the global material information, the single-layer structural drawing data, and the corrected component parameter information.

[0017] Furthermore, in some embodiments, the acquisition module is specifically used for: preprocessing the structural construction drawing raster image data to obtain a preprocessed structural construction drawing raster image; performing type recognition on the preprocessed structural construction drawing raster image according to preset structured prompt words to obtain a classification result of the structural construction drawing raster image; and outputting the classification result of the structural construction drawing raster image into the structural general description drawing raster image data, the structural plan drawing raster image data, and the component reinforcement detail drawing raster image data according to a preset structured format.

[0018] Furthermore, in some embodiments, the analysis module is specifically used for: performing semantic segmentation on the structural planar raster image data to obtain the semantic segmentation result of the structural planar map; and generating the single-layer structural map data based on the semantic segmentation result of the structural planar map.

[0019] Furthermore, in some embodiments, the analysis module is also used to: locate and mask non-structural primitives in the structural planar raster image data, and combine a preset image inpainting algorithm to smooth the background and complete the connectivity of the broken structural lines caused by the masking area to obtain a purified structural base map; perform pixel-level feature extraction and classification on the purified structural base map to obtain a structural semantic segmentation classification result, and map the structural semantic segmentation classification result to preset RGB pixel values ​​to obtain the semantic segmentation result of the structural planar image.

[0020] Furthermore, in some embodiments, the analysis module is also used to: obtain the column component connected domain, beam component, and wall component skeleton / axis from the semantic segmentation result of the structural plan view; take the geometric centroid of the column component connected domain, the endpoints and intersections of the skeleton / axis of the beam component and the wall component as graph nodes, and construct an initial graph node set; merge the coordinates of all beam endpoints whose distance from the center of the column component is less than a preset threshold to the corresponding column component center coordinates, take the merged column component center coordinates as a single graph node, and update the initial graph node set according to the single graph node to obtain a target graph node set; split the beam component skeleton based on the target graph node set to obtain split beam segments and wall segments, take the two ends of the split beam segments as graph nodes, construct a target graph edge set according to the split beam segments, and generate the single-layer structural diagram data.

[0021] Furthermore, in some embodiments, the analysis module is also used to: extract component information from the raster image data of the component reinforcement detail drawing to obtain parametric information of the component reinforcement detail drawing, wherein the parametric information includes component number, component cross-sectional dimensions, and component reinforcement list; calculate the component cross-sectional boundary and geometric center position based on the component cross-sectional dimensions, and extract the rebar diameter, protective layer thickness, and stirrup diameter from the component reinforcement list; determine whether the distance from the rebar center coordinates in the component reinforcement list to the edge of the component cross-section is less than the sum of the protective layer thickness, stirrup diameter, and longitudinal rebar radius; if it is less, then translate the rebar center coordinates inward along the component cross-section normal until the boundary construction requirements are met, to obtain the corrected component parameter information.

[0022] Furthermore, in some embodiments, the analysis module is also used to: extract component information from the raster image data of the component reinforcement detail drawing to obtain parametric information of the component reinforcement detail drawing, wherein the parametric information includes component number, component cross-sectional dimensions, and component reinforcement list; determine the component cross-sectional boundary and the geometric center position of the component cross-section based on the component cross-sectional dimensions, and extract the rebar diameter, protective layer thickness, and stirrup diameter recorded in the component reinforcement list; determine whether the distance from the rebar center coordinate in the component reinforcement list to the edge of the component cross-section is less than the sum of the protective layer thickness, stirrup diameter, and longitudinal rebar radius; if it is less, then translate the rebar center coordinate inward along the component cross-section normal until the geometric center position coordinate of the component cross-section meets the preset boundary construction requirements, and output the corrected rebar design position coordinate and the parametric information together as the corrected component parameter information.

[0023] Furthermore, in some embodiments, the analysis module is also used to: identify the concrete strength, design standards, and general construction requirements in the raster image data of the general structural description diagram, and output the global material information in a preset structured format.

[0024] Furthermore, in some embodiments, the construction module is specifically used for: embedding the global material information and the corrected component parameter information into the single-layer structural diagram data according to a preset component number matching method to obtain single-layer structural diagram data with embedded information; performing topological verification and completion on the single-layer structural diagram data with embedded information to obtain single-layer structural diagram data with complete attributes; and performing spatial alignment and combination on the single-layer structural diagram data with complete attributes to generate a computable structural model of the target building.

[0025] The computable structural model reconstruction device based on structural construction drawing raster images provided in this embodiment of the invention takes the structural construction drawing raster image data of the target building as input, and generates three types of drawing data through classification. Then, it generates single-layer structural drawing data from the structural plan drawing, extracts and corrects component parameter information from the component reinforcement detail drawing, and extracts global material information from the structural general description drawing. Finally, it integrates the three types of data to generate a computable structural model of the target building. This solves the problems of low efficiency in building structure modeling and poor applicability of model structure calculation and analysis in related technologies, improves modeling efficiency, and enhances the applicability of computable structural models.

[0026] To achieve the above objectives, a third aspect of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the computable structural model reconstruction method based on structural construction drawing raster images as described in the above embodiments.

[0027] Therefore, the present invention has the following beneficial effects: (1) This invention directly processes raster images of structural construction drawings without relying on original CAD vector layer information or primitive information, and can be applied to various scenarios such as scanned copies of paper drawings, photographed images and image-based PDFs.

[0028] (2) This invention achieves the joint extraction of structural layout information, component attribute information and material attribute information through the collaborative processing of multiple drawings, including the general structural description drawing, structural plan drawing and component reinforcement detail drawing. Furthermore, through the construction of single-layer structural drawing data and floor assembly, it achieves the automatic conversion from two-dimensional structural construction drawing raster image to a computable structural model, thereby improving the efficiency of automatic acquisition of structural information and BIM model generation.

[0029] (3) This invention transforms the component layout, component attributes and material properties in the structural construction drawings into an electronic data format with geometric and physical meaning that is easy for computers to recognize and calculate, thereby providing a unified data foundation for subsequent structural analysis, model management, dataset construction and intelligent computing. Attached Figure Description

[0030] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A flowchart of a computable structural model reconstruction method based on structural construction drawing raster images provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a two-stage semantic segmentation process according to a specific embodiment of the present invention; Figure 3 A schematic diagram illustrating the construction of a single-layer structure diagram data according to a specific embodiment of the present invention; Figure 4 This is a schematic diagram of the semantic segmentation result to image data flow according to a specific embodiment of the present invention; Figure 5 This is a schematic diagram of a multi-drawing collaborative automatic reconstruction method based on structural construction drawing raster images according to a specific embodiment of the present invention; Figure 6 A block diagram of a computable structural model reconstruction device based on a structural construction drawing raster image provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present invention. Detailed Implementation

[0031] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0032] The following describes, with reference to the accompanying drawings, a method for reconstructing a computable structural model based on a structural construction drawing raster image according to an embodiment of the present invention. First, the method for reconstructing a computable structural model based on a structural construction drawing raster image according to an embodiment of the present invention will be described with reference to the accompanying drawings.

[0033] Figure 1 This is a flowchart of a method for reconstructing a computable structural model based on a structural construction drawing raster image, according to an embodiment of the present invention.

[0034] like Figure 1 As shown, the method for reconstructing a computable structural model based on raster images of structural construction drawings includes the following steps: In step S101, structural construction drawing raster image data of the target building is obtained, and structural general description drawing raster image data, structural plan drawing raster image data, and component reinforcement detail drawing raster image data are generated based on the structural construction drawing raster image data.

[0035] Specifically, in this embodiment of the invention, raster image data of construction drawings for a multi-story reinforced concrete frame structure teaching building, which contains only scanned copies of paper-based archived blueprints and has lost its original electronic drawings, is obtained.

[0036] Furthermore, in some embodiments, generating structural general description raster image data, structural plan raster image data, and component reinforcement detail raster image data based on structural construction drawing raster image data includes: preprocessing the structural construction drawing raster image data to obtain preprocessed structural construction drawing raster images; performing type recognition on the preprocessed structural construction drawing raster images according to preset structured prompt words to obtain structural construction drawing raster image classification results; and outputting the structural construction drawing raster image classification results as structural general description raster image data, structural plan raster image data, and component reinforcement detail raster image data according to a preset structured format.

[0037] Among them, the structural general description raster image data refers to the bitmap data of the architectural structural general description drawings, which records global information such as overall material parameters, design standards, and general construction requirements of the project; the structural plan raster image data refers to the bitmap data of the structural plan layout drawings of each floor of the building, which shows the plan position and arrangement relationship of components such as beams, columns, and shear walls; and the component reinforcement detail raster image data refers to the bitmap data of the component reinforcement detail drawings, which marks parameters such as component cross-sectional dimensions, steel bar specifications, and arrangement positions.

[0038] As one possible implementation, this embodiment of the invention preprocesses the acquired raster image, uses adaptive Gaussian filtering for noise reduction, then detects the outer frame of the drawing based on Hough Transform, calculates the tilt angle and performs affine rotation correction, and finally uses CLAHE (contrast-limited adaptive histogram equalization) to enhance the contrast of primitive lines, making degraded grid lines and text clearer. The preprocessed structural construction drawing raster image is then input into a multimodal large language model, such as a visual language pre-training model based on the Transformer architecture. To prevent the model from outputting redundant explanatory text that would prevent subsequent programs from parsing, strong constraint prompts are injected at the system level, such as "You are a construction engineer. Please identify the type of this drawing. Only one of the following three enumerated values ​​is allowed to be output: 'Structural General Description Drawing', 'Structural Plan Drawing', 'Component Reinforcement Details'. Please return in JSON format: {"paper_type": The "drawing category"} is automatically categorized by parsing the returned JSON data and processing it into three different pipelines: structural general description raster image data, structural plan raster image data, and component reinforcement detail raster image data.

[0039] In step S102, single-layer structural diagram data is obtained based on the structural plan raster image data. Component parameter information is extracted from the component reinforcement detail raster image data. The relative coordinates of the reinforcing bars in the component parameter information are checked for cross-sectional geometry and corrected for position to obtain the corrected component parameter information. Global material information is also extracted from the structural general description diagram raster image data.

[0040] Furthermore, in some embodiments, obtaining single-layer structure map data based on structure planar raster image data includes: performing semantic segmentation on the structure planar raster image data to obtain semantic segmentation results of the structure planar map; and generating single-layer structure map data based on the semantic segmentation results of the structure planar map.

[0041] The process involves extracting global material information from the raster image data of the structural general description diagram, including identifying the concrete strength, design standards, and general construction requirements in the raster image data of the structural general description diagram, and outputting the global material information in a preset structured format.

[0042] Furthermore, the semantic segmentation result of the structural plan view refers to the tagged image output after semantic segmentation processing. Different component regions are assigned unique identifiers. The single-layer structural diagram data refers to the structured topological data composed of graph nodes and graph edges, which fully records the coordinates, attributes, and connection relationships between all components in a single floor.

[0043] In some embodiments, semantic segmentation is performed on the structural planar raster image data to obtain the semantic segmentation result of the structural planar image. This includes: locating and masking non-structural primitives in the structural planar raster image data, and combining a preset image inpainting algorithm to smooth the background and complete the connectivity of broken structural lines caused by the masking area to obtain a purified structural base image; performing pixel-level feature extraction and classification on the purified structural base image to obtain a structural semantic segmentation classification result, and mapping the structural semantic segmentation classification result to preset RGB pixel values ​​to obtain the semantic segmentation result of the structural planar image.

[0044] It should be noted that in actual engineering drawings, due to the dense annotation information, text and leader lines often overlap or intersect with the outlines of beams and columns. If a single end-to-end segmentation network is used directly, it is easy to cause problems such as component boundary adhesion or topological breakage. Therefore, the embodiments of the present invention adopt a two-stage semantic segmentation strategy.

[0045] Specifically Figure 2 This is a schematic diagram of a two-stage semantic segmentation process according to a specific embodiment of the present invention, such as... Figure 2 As shown, this embodiment of the invention first performs the first stage (unstructured primitive stripping), using the DBNet text detection model to identify all character bounding boxes and thin lead-out lines, setting their pixels as the background color. For the line gaps left after removing the text, the system calls an image inpainting algorithm based on Generative Adversarial Network (GAN) or Mask Autoencoder (MAE), performing tensor interpolation based on the line direction at both ends of the gap to restore the occluded continuous structural lines and obtain a purified structural base map. Subsequently, the second stage (component extraction and feature coloring) is performed, inputting the purified structural base map into U-Net deep learning. The learning network, by eliminating interference from text and lines, can focus on the geometric feature recognition of closed regions. After the model outputs the feature map, the system uses a post-processing channel to uniformly assign the connected component pixel matrix classified as "beam" to a preset first RGB vector (e.g., red [255,0,0]), the "column" region to a second RGB vector (e.g., green [0,255,0]), and the "shear wall" region to a third RGB vector (e.g., blue [0,0,255]), while the rest are set to black background, thus obtaining a high-fidelity mask image, i.e., the semantic segmentation result of the structural planar diagram.

[0046] Further, in some embodiments, generating single-layer structural diagram data based on the semantic segmentation results of the structural plan view includes: obtaining the connected domains of column members, and the skeletons / axis lines of beam members and wall members from the semantic segmentation results of the structural plan view; using the geometric centroids of the connected domains of column members, and the endpoints and intersections of the skeletons / axis lines of beam members and wall members as graph nodes, and constructing an initial set of graph nodes; merging the coordinates of all beam endpoints whose distance from the center of the column member is less than a preset threshold to the corresponding center coordinates of the column member, using the merged center coordinates of the column member as a single graph node, and updating the initial set of graph nodes based on the single graph node to obtain a target set of graph nodes; splitting the skeleton of the beam member based on the target set of graph nodes to obtain split beam segments and wall segments, using the two ends of the split beam segments as graph nodes, constructing a target set of graph edges based on the split beam segments, and generating single-layer structural diagram data.

[0047] As one possible way to achieve this, Figure 3 As a schematic diagram of single-layer structural diagram data construction according to a specific embodiment of the present invention, the pixel region of the beam / wall semantic segmentation result diagram is skeletonized to obtain a skeleton centerline with a width of one pixel. The skeleton pixel set is denoted as... S Traverse the skeleton pixel set S For each pixel, count its neighboring pixels. If the number of neighboring pixels is equal to 1, it is identified as a skeleton endpoint; if the number of neighboring pixels is greater than 2, it is identified as a skeleton intersection. All endpoints and intersections are then used as a preliminary candidate node set. V c Perform connected component analysis on the pixel connected regions of the column semantic segmentation result image; calculate the geometric centroid of each connected region as the set of column center nodes. V p For each skeleton endpoint v e ∈ V c If the distance from the center of any pillar v p ∈ V p If the number of pixels is less than or equal to the preset snapping threshold (e.g., 15 pixels), then... v e Move to v p If multiple endpoints are attracted to the same pillar center, they are merged into a single node; the pillar center node is then... V p The nodes are merged at skeleton intersections and after adsorption; clustering algorithms (such as DBSCAN, with the cluster radius set manually, such as 10 pixels) are used to merge nearby nodes to obtain the final node set. VEach node is assigned a unique number; the skeleton is split into edges along the pixel path, with the two ends of each edge being endpoints or intersections; then the two ends of each edge are mapped to the nearest node, forming a set of undirected edges. E Each edge records the start and end node IDs and the skeleton length (in pixels); using the drawing reference point or the lower left corner as the local coordinate origin, and combining the drawing scale text or manually set scale, the pixel coordinates are mapped to physical plane coordinates; each resulting drawing node... V i Each possesses a true physical plane coordinate characteristic. X i =[ x real , y real ], each edge E ij Save the node indices and component names at both ends; finally, use a large language model to identify the component names and the pixel coordinates corresponding to each component name, and assign the component names to graph edges as initial features according to the principle of proximity, thereby realizing the generation and construction of the target graph edge set.

[0048] Furthermore, in some embodiments, the relative coordinates of the reinforcing bars in the component parameter information are subjected to cross-sectional geometric verification and position correction to obtain corrected component parameter information. This includes: extracting component information from the raster image data of the component reinforcement detail drawing to obtain parameterized information of the component reinforcement detail drawing, wherein the parameterized information includes the component number, component cross-sectional dimensions, and component reinforcement list; calculating the component cross-sectional boundary and geometric center position based on the component cross-sectional dimensions, and extracting the reinforcing bar diameter, protective layer thickness, and stirrup diameter from the component reinforcement list; determining whether the distance from the center coordinate of the reinforcing bar in the component reinforcement list to the edge of the component cross-section is less than the sum of the protective layer thickness, stirrup diameter, and longitudinal reinforcing bar radius; if it is less, then translating the center coordinate of the reinforcing bar inward along the normal direction of the component cross-section until the boundary construction requirements are met to obtain the corrected component parameter information.

[0049] Among them, parametric information refers to converting the graphics and annotations in the drawings into digital parameters that can be recognized, stored and retrieved by computers; component number refers to the code in the drawings used to uniquely identify a single building component; component cross-sectional dimensions refer to the geometric dimensions such as the length, width and thickness of the cross-section of components such as beams and columns; and component reinforcement list refers to the collection of reinforcement-related data that records the specifications, quantity, arrangement and location of the steel bars inside the component.

[0050] It should be noted that traditional large language models are prone to producing "artificial intelligence illusions," which can cause the output steel bar coordinates to violate physical reality. Therefore, this embodiment of the invention introduces a post-processing layer for cross-sectional geometry verification rules to eliminate the influence of artificial intelligence illusions.

[0051] As one possible implementation, this embodiment of the invention uses an open-source large language model to extract the information of the frame column "KZ1": cross-sectional dimensions 600×600mm, stirrup diameter 10mm, protective layer thickness 30mm, and the relative axis coordinates of a certain longitudinal steel bar are identified as (290, 290) through the large language model; let the left and right net boundary coordinates of the component cross-section be respectively... x l and x r The coordinates of the upper and lower net boundaries are respectively y b and y t Then the geometric center coordinates of the component section can be expressed as: x c =( x l + x r ) / 2, y c =( y b + y t ) / 2. Let the first digit be 2. i The radius of the root longitudinal reinforcement is r (For example, if the diameter is 25 mm, then r = 12.5 mm). The minimum allowable distance constraint value between the center of the rebar and the boundary of the section is: protective layer thickness + stirrup diameter + longitudinal reinforcement radius = 30 + 10 + 12.5 = 52.5 mm. Half the width of the column section is 300 mm. The longitudinal reinforcement coordinates output by the model are 290 mm, which is only 10 mm away from the edge, less than the 52.5 mm limit required by the structure (300 - 52.5 = 247.5 mm). Therefore, the abnormal coordinates (290, 290) are automatically forcibly corrected along the normal direction to the safe critical coordinates (247.5, 247.5) that meet the boundary constraints. This achieves the coordinate correction and outputs the corrected component parameter information.

[0052] In step S103, a computable structural model of the target building is generated based on global material information, single-layer structural diagram data, and corrected component parameter information.

[0053] Furthermore, in some embodiments, generating a computable structural model of the target building based on global material information, single-layer structural diagram data, and corrected component parameter information includes: embedding global material information and corrected component parameter information into single-layer structural diagram data according to a preset component number matching method to obtain single-layer structural diagram data with embedded information; performing topological verification and completion on the single-layer structural diagram data with embedded information to obtain single-layer structural diagram data with complete attributes; and performing spatial alignment and combination on the single-layer structural diagram data with complete attributes to generate a computable structural model reconstruction result of the target building.

[0054] As one possible implementation, this embodiment of the invention uses global material information and a string precise / fuzzy matching algorithm to match and interface the corrected component parameter information (JSON data) with the target graph edge set (component number). To address the identification errors caused by text occlusion that often occurs in real engineering drawings, a topological connectivity verification and adjacent component attribute interpolation mechanism is set up. That is, the adjacency matrix A is traversed. If the graph edge Eij (the beam between nodes i and j) fails to match the cross-sectional dimensions, the system will retrieve the previous span adjacent edge Eki of node i. According to the prior physical rule that continuous span beams in building structures usually have equal cross-sections, the cross-sectional attributes of edge Eki are temporarily assigned to Eij and marked as a highlighted alarm - interpolation generation in the system log, thereby improving the robustness of large models in noisy engineering environments.

[0055] further, Figure 4 This is a schematic diagram of the semantic segmentation result to graph data flow according to a specific embodiment of the present invention, such as... Figure 4 As shown, the semantic segmentation image of the structural plan is first processed into a beam skeleton, and then the beam endpoints and intersections are identified, and the column center points are identified. Next, the beam endpoints are snapped to the column center and the nodes are merged, and then the graph edges are generated. The coordinate transformation is completed to obtain the single-layer structural diagram data. At the same time, the original drawing is parsed by a multimodal large language model to identify the component names and synchronously supplement them into the single-layer structural diagram data.

[0056] Furthermore, the elevations (e.g., ±0.000, 3.600, 7.200) of each structural planar diagram are extracted using a large language model, and the data structures of each single-layer two-dimensional diagram are then processed. G Placed in the corresponding three-dimensional space Z At the axial height, vertical column nodes with the same number across floors are used as 3D anchor points for spatial topology alignment, and fine-tuning and registration are performed on each floor. Finally, the assembled 3D spatial diagram is output as IFC format or other 3D software formats via API, completing end-to-end automated structural reconstruction and obtaining a computable structural model of the target building.

[0057] To enable those skilled in the art to better understand the computable structural model reconstruction method of the present invention, the following explanation will be provided in conjunction with specific embodiments.

[0058] Figure 5 This is a schematic diagram of a multi-drawing collaborative BIM automatic reconstruction method based on structural construction drawing raster images according to a specific embodiment of the present invention, as shown below. Figure 5As shown, firstly, multiple raster images of structural construction drawings are used as input. After preprocessing, the drawings are classified using a multimodal large language model, distinguishing between general specification drawings, reinforcement details of various components, and standard construction drawings. Subsequently, global material information such as concrete strength is extracted from the general specification drawings, and component parameters are extracted from the reinforcement details. At the same time, a two-stage semantic segmentation model is used to process the standard construction drawings to obtain component segmentation results, which are then constructed into image data using morphological methods. Finally, the component parameters, material information, and image data are fused together, and a complete building structure model is generated after floor-by-floor stitching.

[0059] The computable structural model reconstruction method based on structural construction drawing raster images provided by the present invention takes the structural construction drawing raster image data of the target building as input, and generates three types of drawing data through classification. Then, it generates single-layer structural drawing data from the structural plan drawing, extracts and corrects component parameter information from the component reinforcement detail drawing, and extracts global material information from the structural general description drawing. Finally, it integrates the three types of data to generate a computable structural model of the target building. This method solves the problems of low efficiency in building structure modeling and poor applicability of model structure calculation and analysis in related technologies, improves modeling efficiency, and enhances the applicability of computable structural models.

[0060] Next, the computable structural model reconstruction apparatus according to an embodiment of the present invention is described with reference to the accompanying drawings.

[0061] Figure 6 This is a block diagram of a computable structural model reconstruction device provided according to an embodiment of the present invention.

[0062] like Figure 6 As shown, the computable structural model reconstruction device 10 includes: an acquisition module 100, an analysis module 200, and a construction module 300.

[0063] The acquisition module 100 is used to acquire raster image data of the structural construction drawings of the target building, and generate raster image data of the general structural description drawing, the structural plan drawing, and the component reinforcement detail drawing based on the raster image data of the structural construction drawing. The analysis module 200 is used to obtain single-story structural drawing data based on the structural plan drawing raster image data, extract component parameter information through the component reinforcement detail drawing raster image data, perform cross-sectional geometric verification and position correction on the relative coordinates of the steel bars in the component parameter information to obtain the corrected component parameter information, and extract global material information from the general structural description drawing raster image data. The construction module 300 is used to generate a computable structural model of the target building based on the global material information, the single-story structural drawing data, and the corrected component parameter information.

[0064] Furthermore, in some embodiments, the acquisition module 100 is specifically used for: preprocessing the structural construction drawing raster image data to obtain a preprocessed structural construction drawing raster image; performing type recognition on the preprocessed structural construction drawing raster image according to preset structured prompt words to obtain a structural construction drawing raster image classification result; and outputting the structural construction drawing raster image classification result into structural general description raster image data, structural plan raster image data, and component reinforcement detail raster image data according to a preset structured format.

[0065] Furthermore, in some embodiments, the analysis module 200 is specifically used for: performing semantic segmentation on the structural planar raster image data to obtain the semantic segmentation result of the structural planar image; and generating single-layer structural image data based on the semantic segmentation result of the structural planar image.

[0066] Furthermore, in some embodiments, the analysis module 200 is also used to: locate and mask non-structural primitives in the structural planar raster image data, and combine a preset image inpainting algorithm to smooth the background and complete the connectivity of the broken structural lines caused by the masking area to obtain a purified structural base map; perform pixel-level feature extraction and classification on the purified structural base map to obtain a structural semantic segmentation classification result, and map the structural semantic segmentation classification result to a preset RGB pixel value to obtain a semantic segmentation result of the structural planar image.

[0067] Furthermore, in some embodiments, the analysis module 200 is also used to: obtain the column component connected domain, beam component, and wall component skeleton / axis from the semantic segmentation result of the structural plan view; take the geometric centroid of the column component connected domain, the endpoints and intersections of the beam component and wall component skeleton / axis as graph nodes, and construct an initial graph node set; merge the coordinates of all beam endpoints whose distance from the center of the column component is less than a preset threshold to the corresponding column component center coordinates, take the merged column component center coordinates as a single graph node, and update the initial graph node set according to the single graph node to obtain a target graph node set; split the beam component skeleton based on the target graph node set to obtain split beam segments and wall segments, take the two ends of the split beam segments as graph nodes, construct a target graph edge set according to the split beam segments, and generate single-layer structural diagram data.

[0068] Furthermore, in some embodiments, the analysis module 200 is also used to: extract component information from the raster image data of the component reinforcement detail drawing to obtain parametric information of the component reinforcement detail drawing, wherein the parametric information includes component number, component cross-sectional dimensions and component reinforcement list; calculate the component cross-sectional boundary and geometric center position based on the component cross-sectional dimensions, and extract the rebar diameter, protective layer thickness and stirrup diameter from the component reinforcement list; determine whether the distance from the rebar center coordinates in the component reinforcement list to the edge of the component cross-section is less than the sum of the protective layer thickness, stirrup diameter and longitudinal rebar radius; if it is less, then translate the rebar center coordinates inward along the component cross-section normal until the boundary construction requirements are met, and obtain the corrected component parameter information.

[0069] Furthermore, in some embodiments, the analysis module 200 is also used to: extract component information from the raster image data of the component reinforcement detail drawing to obtain parametric information of the component reinforcement detail drawing, wherein the parametric information includes component number, component cross-sectional dimensions, and component reinforcement list; determine the component cross-sectional boundary and the geometric center position of the component cross-section based on the component cross-sectional dimensions, and extract the steel bar diameter, protective layer thickness, and stirrup diameter recorded in the component reinforcement list; determine whether the distance from the steel bar center coordinate in the component reinforcement list to the edge of the component cross-section is less than the sum of the protective layer thickness, stirrup diameter, and longitudinal steel bar radius; if it is less, then translate the steel bar center coordinate inward along the normal of the component cross-section until the geometric center position coordinate of the component cross-section meets the preset boundary construction requirements, and output the corrected steel bar design position coordinate and parametric information together as the corrected component parameter information.

[0070] Furthermore, in some embodiments, the analysis module 200 is also used to: identify concrete strength, design standards and general construction requirements in the raster image data of the general structural description diagram, and output global material information in a preset structured format.

[0071] Furthermore, in some embodiments, the construction module 300 is specifically used to: embed global material information and corrected component parameter information into single-layer structural diagram data according to a preset component number matching method to obtain single-layer structural diagram data with embedded information; perform topological verification and completion on the single-layer structural diagram data with embedded information to obtain single-layer structural diagram data with complete attributes; and perform spatial alignment and combination on the single-layer structural diagram data with complete attributes to generate a computable structural model of the target building.

[0072] It should be noted that the foregoing explanation of the embodiment of the computable structural model reconstruction method based on structural construction drawing raster images also applies to the computable structural model reconstruction device based on structural construction drawing raster images of this embodiment, and will not be repeated here.

[0073] The computable structural model reconstruction device based on structural construction drawing raster images provided in this embodiment of the invention takes the structural construction drawing raster image data of the target building as input, and generates three types of drawing data through classification. Then, it generates single-layer structural drawing data from the structural plan drawing, extracts and corrects component parameter information from the component reinforcement detail drawing, and extracts global material information from the structural general description drawing. Finally, it integrates the three types of data to generate a computable structural model of the target building. This solves the problems of low efficiency in building structure modeling and poor applicability of model structure calculation and analysis in related technologies, improves modeling efficiency, and enhances the applicability of computable structural models.

[0074] Figure 7 This is a schematic diagram of an electronic device provided according to an embodiment of the present invention. The electronic device may include: The memory 701, the processor 702, and the computer program stored on the memory 701 and executable on the processor 702.

[0075] When the processor 702 executes the program, it implements the computable structural model reconstruction method based on the structural construction drawing raster image provided in the above embodiments.

[0076] Furthermore, electronic devices also include: Communication interface 703 is used for communication between memory 701 and processor 702.

[0077] The memory 701 is used to store computer programs that can run on the processor 702.

[0078] The memory 701 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0079] If the memory 701, processor 702, and communication interface 703 are implemented independently, then the communication interface 703, memory 701, and processor 702 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 7 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0080] Optionally, in a specific implementation, if the memory 701, processor 702, and communication interface 703 are integrated on a single chip, then the memory 701, processor 702, and communication interface 703 can communicate with each other through an internal interface.

[0081] The processor 702 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement embodiments of the present invention.

[0082] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0083] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0084] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for reconstructing a computable structural model based on raster images of structural construction drawings, characterized in that, Includes the following steps: Obtain raster image data of structural construction drawings of the target building, and generate raster image data of general structural description drawing, structural plan drawing, and component reinforcement detail drawing based on the raster image data of structural construction drawings; Single-layer structural diagram data is obtained from the raster image data of the structural plan view. Component parameter information is extracted from the raster image data of the component reinforcement detail drawing. The relative coordinates of the steel bars in the component parameter information are checked for cross-sectional geometry and corrected for position to obtain the corrected component parameter information. Global material information is also extracted from the raster image data of the structural general description drawing. A computable structural model of the target building is generated based on the global material information, the single-layer structural diagram data, and the corrected component parameter information.

2. The method according to claim 1, characterized in that, The process of generating raster image data of the general structural description drawing, structural plan drawing, and component reinforcement detail drawing based on the raster image data of the structural construction drawing includes: The structural construction drawing raster image data is preprocessed to obtain a preprocessed structural construction drawing raster image; The preprocessed structural construction drawing raster image is classified according to the preset structured prompt words to obtain the classification result of the structural construction drawing raster image; According to the preset structured format, the classification results of the structural construction drawing raster images are output as raster image data of the general structural description drawing, raster image data of the structural plan drawing, and raster image data of the component reinforcement detail drawing.

3. The method according to claim 1, characterized in that, The step of obtaining single-layer structure map data based on the structure planar raster image data includes: Semantic segmentation is performed on the raster image data of the structural planar diagram to obtain the semantic segmentation result of the structural planar diagram; The single-layer structure graph data is generated based on the semantic segmentation results of the structure planar graph.

4. The method according to claim 3, characterized in that, The semantic segmentation of the structural planar raster image data to obtain the semantic segmentation result of the structural planar image includes: The unstructure primitives in the structural planar raster image data are located and masked, and the broken structural lines caused by the masked area are smoothed and connected by a preset image restoration algorithm to obtain a purified structural base image. Pixel-level feature extraction and classification are performed on the purified structural base image to obtain structural semantic segmentation classification results. The structural semantic segmentation classification results are mapped to preset RGB pixel values ​​to obtain the semantic segmentation results of the structural planar image.

5. The method according to claim 3, characterized in that, The step of generating the single-layer structure graph data based on the semantic segmentation result of the structure plane graph includes: Obtain the connected domains of column members, and the skeleton / axis of beam members and wall members from the semantic segmentation results of the structural plan view; The geometric centroid of the connected domain of the column member, the endpoints and intersections of the skeleton / axis of the beam member and the wall member are used as graph nodes, and an initial set of graph nodes is constructed. The coordinates of all beam endpoints whose distance from the center of the column member is less than a preset threshold are merged into the corresponding center coordinates of the column member. The merged center coordinates of the column member are used as a single graph node, and the initial graph node set is updated according to the single graph node to obtain the target graph node set. The beam member skeleton is split based on the target graph node set to obtain split beam segments and wall segments. The two ends of the split beam segments are used as graph nodes. A target graph edge set is constructed based on the split beam segments to generate the single-layer structure graph data.

6. The method according to claim 1, characterized in that, The step of performing cross-sectional geometric verification and position correction on the relative coordinates of the reinforcing bars in the component parameter information to obtain the corrected component parameter information includes: The component information is extracted from the raster image data of the component reinforcement detail drawing to obtain the parameterized information of the component reinforcement detail drawing, wherein the parameterized information includes the component number, component cross-sectional dimensions and component reinforcement list; The component's cross-sectional dimensions are used to calculate the component's cross-sectional boundary and geometric center location, and the steel bar diameter, protective layer thickness, and stirrup diameter are extracted from the component's reinforcement list. Determine whether the distance from the center coordinate of the steel bar in the reinforcement list to the edge of the component section is less than the sum of the protective layer thickness, the stirrup diameter, and the longitudinal steel bar radius; if it is less, then translate the center coordinate of the steel bar inward along the normal of the component section until the boundary construction requirements are met, and obtain the corrected component parameter information.

7. The method according to claim 1, characterized in that, The step of extracting global material information from the raster image data of the overall structural description diagram includes: identifying the concrete strength, design standards, and general construction requirements in the raster image data of the overall structural description diagram, and outputting the global material information in a preset structured format.

8. The method according to claim 1, characterized in that, The process of generating a computable structural model of the target building based on the global material information, the single-layer structural diagram data, and the corrected component parameter information includes: According to the preset component number matching method, the global material information and the corrected component parameter information are embedded into the single-layer structure diagram data to obtain the single-layer structure diagram data after embedding information. Perform topology verification and completion on the single-layer structure graph data after embedding the information to obtain single-layer structure graph data with complete attributes; Spatially align and combine the complete single-layer structural diagram data of the aforementioned attributes to generate a computable structural model reconstruction result of the target building.

9. A computationally readable structural model reconstruction device based on raster images of structural construction drawings, characterized in that, include: The acquisition module is used to acquire structural construction drawing raster image data of the target building, and generate structural general description raster image data, structural plan raster image data, and component reinforcement detail raster image data based on the structural construction drawing raster image data; The analysis module is used to obtain single-layer structural diagram data based on the structural plan raster image data, extract component parameter information from the component reinforcement detail raster image data, perform cross-sectional geometry verification and position correction on the relative coordinates of the steel bars in the component parameter information to obtain corrected component parameter information, and extract global material information from the structural general description raster image data. A construction module is used to generate a computable structural model of the target building based on the global material information, the single-layer structural diagram data, and the corrected component parameter information.

10. An electronic device, characterized in that, include: The method includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the program to implement the method for reconstructing a computable structural model based on a structural construction drawing raster image as described in any one of claims 1-8.

Images