Building construction drawing intelligent examination method and system

Abstract

The invention discloses an intelligent examination method and system for a building construction drawing, and belongs to the technical field of building construction. The technical problems that a traditional manual review mode is low in efficiency and prone to making mistakes, and an existing automatic system is difficult to process heterogeneous drawing data and understand complex building specifications are solved. According to the technical scheme, the method comprises the steps of receiving original drawing files generated by different drawing software through a data interface and analyzing the original drawing files into standardized structured data objects; fusing two-dimensional geometric and text information based on the data object, and reconstructing a lightweight three-dimensional building information model with semantic attributes; converting the building standard provisions into structured computational logic rule nodes through natural language processing to form a standard knowledge graph; and finally, carrying out dynamic matching reasoning on the three-dimensional model and the knowledge graph, automatically identifying violation through parameterization calculation and logic judgment, and generating a visual review report. The system is mainly used for realizing automatic, intelligent and precise examination of the building construction drawing.

Description

Technical Field

[0001] This invention relates to the field of building construction technology. More specifically, this invention relates to an intelligent review method and system for building construction drawings. Background Technology

[0002] In the field of architectural construction drawing review, existing technologies mainly rely on manual review or automated review based on two-dimensional computer-aided design drawings. These methods face difficulties when processing heterogeneous drawing data generated by different design software because the differences in data format and structure lead to complex information parsing processes and make it difficult to achieve automated data extraction and integration.

[0003] Furthermore, the national and industry building codes upon which the reviews are based are mostly in the form of natural language text. These code provisions lack structured semantics that can be directly understood and executed in computer systems, making it difficult for automated review systems to accurately understand and apply the complete logic and constraints of the codes. In such cases, reviewers usually still need to manually interpret and convert them into actionable instructions.

[0004] Furthermore, existing automated review methods mostly operate at the two-dimensional drawing level, making it difficult to accurately calculate geometric relationships and parameters in three-dimensional space, such as the actual spatial distance between components or the volume of complex areas. At the same time, these methods typically perform global batch reviews of drawings, failing to flexibly focus on specific components or local areas according to actual needs. When facing large projects or partial design iterations, the review process consumes significant computational resources and has low response efficiency. How to effectively parse heterogeneous data, computationally understand natural language specifications, and achieve precise and efficient spatial and scope control of the review process are urgent technical problems to be solved in this field. Summary of the Invention

[0005] One object of the present invention is to solve at least the above-mentioned problems and to provide at least the advantages that will be described later.

[0006] Another objective of this invention is to provide an intelligent method and system for reviewing architectural construction drawings, which solves the technical problems of low efficiency and error-proneness in traditional manual review methods, as well as the difficulty of existing automated systems in handling heterogeneous drawing data and understanding complex building codes. This achieves automated, intelligent, and accurate review of architectural construction drawings.

[0007] To achieve these objectives and other advantages according to the present invention, a smart review method for architectural construction drawings is provided, comprising the following steps: S1. Standardized Input and Parsing of Heterogeneous Drawing Data: The system receives original electronic architectural construction drawings generated by different computer-aided design software via a data interface. It then calls the format parsing module to uniformly parse and convert these original electronic architectural construction drawings into standardized structured data objects containing geometric vector information, layer attribute information, and text annotation information. Geometric vector information includes the coordinates and topological relationships of points, lines, surfaces, and entities; layer attribute information includes the layer name, line type, and color of the component; and text annotation information includes dimension annotations, text descriptions, and their associated geometric objects. S2. Multi-source information fusion and 3D building information model reconstruction: Based on standardized structured data objects, two-dimensional geometric primitives and text annotation information are fused through a component recognition rule base to generate three-dimensional building components with semantic attributes, which are then aggregated into a lightweight three-dimensional building information model. S3. Knowledge graph processing of computer-readable standard provisions: Pre-store national and industry mandatory standard provisions related to architectural construction drawings, and convert them into structured, computer-executable computational logic rule nodes through natural language processing technology to form a standard knowledge graph; Step S4: Intelligent review and conflict detection based on dynamic rule matching: The lightweight 3D building information model is dynamically matched and reasoned with the standard knowledge graph. The lightweight 3D building information model is parametrically calculated and logically judged according to the calculation logic rule nodes. When a violation is detected, a review conflict record is generated and visualized and a report is generated.

[0008] Preferably, in step S2, the process of generating a three-dimensional building component with semantic attributes by fusing two-dimensional geometric primitives and text annotation information through a component recognition rule base specifically includes: S211, Component Type Recognition: Based on the layer name in the layer attribute information, combined with the primitive type and spatial distribution in the geometric vector information, the candidate building components and their types are initially identified by matching through a predefined component recognition rule base. The types include walls, beams, columns, slabs, doors, and windows. S212, 3D geometry generation: For the identified candidate building components, the corresponding 3D generation algorithm is called according to their type: For wall and column components, the 2D closed contour is extruded, and the extruded height is determined according to the height value in the bound text annotation information or the default specification value; For beam and slab components, the 3D solid is generated by lofting or filling operations based on their 2D contour lines and the marked elevation information. S213, Topology Relationship Reconstruction: Based on the spatial positions of all generated 3D building components, automatically calculate and record the spatial topology relationships between components, including connection relationships, support relationships, and spatial containment relationships.

[0009] Preferably, in step S2, the aggregation into a lightweight 3D building information model specifically includes: S221, Semantic Information Injection: The dimensions, materials, and fire rating attributes related to the components in the text annotation information are associated and bound one-to-one or one-to-many with the three-dimensional building components generated in step S212 to complete the injection of semantic information. S222, Lightweight Model Encapsulation: An incremental lightweight algorithm is adopted to remove intermediate data and historical operation records in the 3D modeling process, while retaining the final geometric data, semantic attribute data and topological relationship data between building components, and encapsulating them into a lightweight 3D building information model. The format of the lightweight 3D building information model is independent of any specific computer-aided design software environment.

[0010] Preferably, step S4 specifically includes: S410, Review Rule Trigger: Traverse each building component entity and its semantic attributes in the lightweight 3D building information model, and automatically match and activate the associated computational logic rule nodes from the specification knowledge graph according to the component type; S420, Parametric Calculation and Logical Judgment: Based on the activated calculation logic rule nodes, the required geometric and semantic parameters are extracted from the lightweight 3D building information model in real time, and parametric mathematical calculations and spatial relationship logical judgments are performed. S430. Review Result Generation and Visual Annotation: The results of parametric calculation and logical judgment are compared with the thresholds or conditions in the rule nodes. If a rule is violated, a review conflict record is generated, which includes the identifier of the violating component, the specific code clause violated, the calculated value of the violating parameter, and the standard requirement value. All review conflict records are visualized and highlighted in three-dimensional space in the lightweight three-dimensional building information model, and a structured review report is generated.

[0011] Preferably, in step S4, before dynamically matching and reasoning the lightweight 3D building information model with the standard knowledge graph, the method further includes: dynamically filtering the target component set or target area from the lightweight 3D building information model based on the user-input review scope selection instruction, and automatically expanding the target review domain based on the component topology relationship; the dynamic matching and reasoning is only performed on the target component set or target area.

[0012] Preferably, step S4, after generating the review conflict record, further includes a review conflict intelligent grading and classification step, specifically: Based on a predefined conflict severity level rule base, each review conflict record is automatically classified, with conflict severity levels including at least a major risk level, a general violation level, and a recommended optimization level; At the same time, the violations of the normative provisions in the review conflict record are automatically classified according to their professional fields, which include at least fire safety, structural safety, energy conservation and environmental protection, and ergonomics. In the generated structured review report and 3D visualization annotations, the review conflict records are sorted, filtered, and displayed in a differentiated manner according to the severity level of the conflict and the professional field classification.

[0013] Preferably, the intelligent review method for architectural construction drawings also includes a model optimization feedback loop independent of the core review process. This model optimization feedback loop is executed periodically or triggered periodically, specifically including: Construction of historical conflict sample library: After the user confirms or corrects the review conflict record generated in step S4, the finally confirmed conflict case, including its corresponding model component data, triggered calculation logic rule node and final judgment result, will be stored as annotated sample in the historical conflict sample library. Rule confidence adaptive adjustment: Based on the historical conflict sample library, the trigger frequency and false alarm rate of each calculation logic rule node are statistically analyzed. When the false alarm rate of a rule is consistently higher than the threshold of 10%, the system automatically reduces the confidence weight of the rule when it is triggered, or gives the user an uncertainty prompt when it is triggered. New rule pattern mining: When the case in the historical conflict sample library is found to have frequent coexistence of two or more types of component violations in the sample, and their combination pattern is not covered by the existing normative knowledge graph, the potential new rule pattern is submitted to the administrator for review as an optimization suggestion, so as to expand or correct the normative knowledge graph.

[0014] Preferably, the intelligent review method for architectural construction drawings also includes an incremental review mode, which is triggered when an update to the specification knowledge graph or a partial revision to the lightweight 3D building information model is detected. Specifically, this includes: Change impact domain analysis: When the specification knowledge graph is updated, the newly added or modified calculation logic rule nodes are analyzed to determine the set of building component types that are affected by them; when the 3D model is partially revised, the set of building components that have been added, deleted, or modified is located. Dynamic construction of the target review domain: The set of affected component types obtained from the above analysis is combined with the set of components that have been revised to generate a target review domain; Targeted rule matching and review: For components within the target review domain, perform dynamic rule matching and reasoning, parameterized calculation and logical judgment, and generate review conflict records.

[0015] Preferably, in step S4, while generating each review conflict record, a corresponding review logic traceability description is automatically generated; the process of generating the review logic traceability description includes: Automatically record the original specification clause content of the computational logic rule node that triggered the conflict; The key model parameters and their values ​​used in this determination are automatically extracted. The key model parameters include at least the geometric parameters and semantic parameters extracted in real time from the lightweight 3D building information model. Based on rule nodes and key model parameters, a natural language description text is automatically generated, which includes the identifier of the non-compliant component, the violated normative clause, the specific parameter values ​​used, the logical judgment process, and the final conclusion. The review logic traceability description and the corresponding review conflict record are associated and stored, and then output to the structured review report and the 3D visualization interface.

[0016] This invention also provides a system for an intelligent review method of architectural construction drawings, comprising: The data parsing module is configured to receive original electronic construction drawings generated by different computer-aided design software through a data interface, and parse and convert them into standardized structured data objects. The 3D model reconstruction module is configured to build a lightweight 3D building information model that integrates geometric and semantic information based on standardized structured data objects and component recognition and 3D generation algorithms. The standardized knowledge graph module is configured to store and manage structured computational logic rule nodes transformed through natural language processing technology, and to form a standardized knowledge graph. The intelligent review engine module is configured to dynamically match and reason with the lightweight 3D building information model and the standard knowledge graph, perform parametric calculations and logical judgments, and generate review conflict records. The review scope selection module is configured to receive user instructions and dynamically filter target component sets or target areas from the lightweight 3D building information model for targeted review by the intelligent review engine module. The conflict handling module is configured to automatically classify the severity level and professional field of the review conflict records generated by the intelligent review engine module. The model optimization feedback module is configured to build a historical conflict sample library and adaptively adjust the rule confidence and mine potential new rule patterns based on it in order to optimize the canonical knowledge graph module. The incremental review control module is configured to automatically analyze the impact domain of the change and construct the target review domain when a model revision or specification update is detected, thereby triggering the incremental review mode. The logic traceability generation module is configured to automatically generate a natural language description for each review conflict record, including the original specification clauses, key parameters, and the judgment process.

[0017] The present invention has at least the following beneficial effects: 1. The intelligent review method for architectural construction drawings of this invention automates and intelligentizes the review process by constructing a complete technical solution from heterogeneous data parsing, 3D model reconstruction, standard knowledge graphization to intelligent review. It integrates the traditionally manual and discrete review processes into a coherent computer-executed process, fundamentally improving the overall efficiency, standardization, and consistency of the review process, and reducing omissions and deviations caused by human intervention.

[0018] 2. The intelligent review method for architectural construction drawings of the present invention ensures the accurate and automatic conversion from two-dimensional drawing information to three-dimensional model elements, and reconstructs the spatial topological relationship between components, laying a solid foundation for subsequent accurate calculation and relationship determination based on three-dimensional space, and improving the quality and usability of model data.

[0019] 3. The intelligent review method for architectural construction drawings of the present invention, on the one hand, enables the model to not only have geometric shape but also contain key engineering semantics through attribute binding, making it an intelligent model that can be "understood" by computers; on the other hand, it reduces data redundancy of the model through lightweight processing, and improves the efficiency of model transmission, loading and calculation in subsequent review processes.

[0020] 4. The intelligent review method for architectural construction drawings of the present invention realizes dynamic and accurate matching between review rules and model components, and performs objective parametric calculations and logical judgments based on the three-dimensional model, making the review conclusions more accurate and traceable. At the same time, through three-dimensional visualization annotation, it greatly facilitates the user's location and understanding of the review results.

[0021] 5. The intelligent review method for architectural construction drawings of the present invention allows users to review specific areas or component types in a targeted manner, avoiding a full scan of the entire model each time, thereby significantly saving computing resources and shortening the review response time. It is especially suitable for local design iteration and rapid verification of large projects.

[0022] 6. The intelligent review method for building construction drawings of the present invention can automatically distinguish the severity of problems and their professional affiliation, guide users to prioritize the handling of major risk items, and help engineers of different professions quickly focus on the problems within their scope of responsibility, thereby optimizing the problem rectification path and improving decision-making and collaboration efficiency.

[0023] 7. The intelligent review method for building construction drawings of the present invention can automatically optimize the accuracy of rule application by utilizing historical review data (such as reducing the weight of false alarm rules) and can proactively discover potential new violation patterns that are not clearly defined, thereby continuously improving the review accuracy and adaptability of the entire system as the usage time increases.

[0024] 8. The intelligent review method for architectural construction drawings of the present invention achieves precise and efficient review by accurately analyzing the impact domain of changes, thus avoiding the need for a full model re-review triggered by minor modifications. This greatly improves the review efficiency and resource utilization in a continuous iterative design environment.

[0025] 9. The intelligent review method for architectural construction drawings of the present invention transforms the system's "judgment conclusion" into a clear "reasoning process," enabling designers to quickly understand the root cause of violations without guessing or reverse engineering, thereby accelerating the process of locating and correcting design errors and improving the smoothness of human-machine collaboration.

[0026] 10. The intelligent review method for building construction drawings of the present invention implements the entire intelligent review method on a specific software architecture, clarifies the responsibilities and collaborative relationships of each functional module, ensures the stable and efficient execution of the method process, and facilitates the integration, deployment and maintenance of the system.

[0027] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Detailed Implementation

[0028] The present invention will now be described in further detail so that those skilled in the art can implement it based on the description.

[0029] It should be understood that terms such as "having," "comprising," and "including" used in the embodiments of this application do not exclude the presence or addition of one or more other elements or combinations thereof. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of components in a specific posture. If the specific posture changes, the directional indication will also change accordingly. When an element is referred to as "fixed to" or "set on" another element, it can be directly on the other element or may have an intervening element present. When an element is referred to as "connected to" another element, it can be directly connected to the other element or indirectly connected to the other element through an intervening element. Descriptions involving "first," "second," etc., in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features.

[0030] It should be noted that, unless otherwise specified, the experimental methods described in the following implementation plan are all conventional methods, and the reagents and materials described are all commercially available unless otherwise specified.

[0031] It should be noted that the technical solutions of the various embodiments of this application can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this application.

[0032] This invention provides an intelligent review method for architectural construction drawings, which includes the following steps: S1. Standardized Input and Parsing of Heterogeneous Drawing Data: The system receives original electronic architectural construction drawings generated by different computer-aided design software via a data interface. It then calls the format parsing module to uniformly parse and convert these original electronic architectural construction drawings into standardized structured data objects containing geometric vector information, layer attribute information, and text annotation information. Geometric vector information includes the coordinates and topological relationships of points, lines, surfaces, and entities; layer attribute information includes the layer name, line type, and color of the component; and text annotation information includes dimension annotations, text descriptions, and their associated geometric objects. S2. Multi-source information fusion and 3D building information model reconstruction: Based on standardized structured data objects, two-dimensional geometric primitives and text annotation information are fused through a component recognition rule base to generate three-dimensional building components with semantic attributes, which are then aggregated into a lightweight three-dimensional building information model. S3. Knowledge graph processing of computer-readable standard provisions: Pre-store national and industry mandatory standard provisions related to architectural construction drawings, and convert them into structured, computer-executable computational logic rule nodes through natural language processing technology to form a standard knowledge graph; Step S4: Intelligent review and conflict detection based on dynamic rule matching: The lightweight 3D building information model is dynamically matched and reasoned with the standard knowledge graph. The lightweight 3D building information model is parametrically calculated and logically judged according to the calculation logic rule nodes. When a violation is detected, a review conflict record is generated and visualized and a report is generated.

[0033] In the above technical solution, the method runs on a computer workstation equipped with a high-performance central processing unit, large-capacity memory and professional graphics card, and realizes intelligent review of building construction drawings by executing a series of automated steps.

[0034] First, the system receives original electronic architectural construction drawings uploaded by users through a standardized data interface. These files may be generated by different computer-aided design software, such as AutoCAD, Revit, or Archicad, and therefore their underlying data formats vary. The system calls its built-in format parsing engine to deconstruct and analyze these heterogeneous drawing files. The parsing process accurately extracts geometric vector information from the drawings, including points (coordinates X / Y / Z, accuracy ±0.01 mm, unique point ID), lines (start / end ID, line width, line type code, line topology association ID), surfaces (comprising a set of line IDs, surface area), and the coordinates (comprising a set of surface IDs, solid volume) and connection relationships of polylines; it also extracts layer attribute information (including the layer name where the component is located, layer classification code (according to GB / T 50104-2010 "Architectural Drawing Standard", such as "A-WALL" corresponding to classification code "01"), line type (code and corresponding line width value, such as "CONTINUOUS" corresponding to 0.3 mm), and color (RGB). The system identifies and associates various text annotation information, including dimensions (annotation value, unit, precision, e.g., "3600mm±1mm"), text descriptions (text content, font size, e.g., "C30 concrete", "Class A fire door"), and geometric object IDs associated with annotations (relationships are determined by "spatial distance ≤ 5 mm and consistent layer affiliation," supporting one-to-one / one-to-many, e.g., annotation "3600mm" associated with wall component "W005")). For example, a dimension of 3600 mm in length, or text indicating the concrete strength grade as C30. All this information is uniformly integrated and converted into a standardized, structured data object defined within the system, thus eliminating barriers caused by differences in data sources.

[0035] Next, based on the standardized data objects mentioned above, the system initiates the 3D model reconstruction process. This process automatically identifies various building components according to a pre-built component recognition rule library. For example, the system will identify a closed polyline located in the "Wall" layer as a wall component based on its specific geometric shape. For identified vertical components such as walls and columns, the system generates 3D solids through extrusion based on their 2D contours and the height values ​​(e.g., 3000 mm) in the bound text annotation information. For horizontal components such as beams and slabs, 3D solids are generated through a lofting algorithm based on their contour lines and the elevation information of the annotations. Finally, all generated 3D components are aggregated, and their geometric information, semantic attributes extracted from the annotations such as material and dimensions, and automatically calculated topological relationships such as connections and supports between components are encapsulated into a lightweight 3D building information model. This model is independent of any specific design software, has a small data volume, and is easy to load and process quickly.

[0036] In the standardization process, the system pre-stores mandatory national and industry standards such as the Code for Fire Protection Design of Buildings and the Unified Standard for Civil Building Design (including 28 current standards, such as GB 50016-2014 "Code for Fire Protection Design of Buildings" and GB 50352-2019 "Unified Standard for Civil Building Design", totaling over 3200 valid clauses). Using Natural Language Processing (NLP) technology, these clauses, written in human language, are converted into structured, computer-executable computational logic rule nodes. For example, the text "The safe evacuation distance should not exceed forty meters" is converted into a rule node: when this node is triggered, the actual three-dimensional path distance from the room door to the safe exit is calculated and compared with the forty-meter threshold. All similar rule nodes are interconnected, forming a vast, reasonable standard knowledge graph. Pre-processing of the clause: Remove punctuation and footnotes (such as "Note: This clause applies to buildings with a height ≤ 100 m"), and extract the core wording (e.g., the original clause "Fire compartments of a building shall be separated by firewalls, and the fire resistance rating of the firewalls shall not be less than 3.00h" is pre-processed to "Fire compartments shall be separated by firewalls, and the fire resistance rating of the firewalls shall be ≥ 3.00h"). Entity recognition: A BERT pre-trained model finely tuned based on building code corpus (accuracy 92.5%) is used to identify "component entities" (such as "firewall"), "parameter entities" (such as "fire resistance limit"), and "threshold entities" (such as "3.00h"). Relationship extraction: The rule engine transforms logical relationships (e.g., "should not be lower than" → "≥", "should be adopted" → "must be used") to form a "component-parameter-relationship-threshold" quadruple (e.g., <firewall, fire resistance limit, ≥, 3.00h>). Ambiguity resolution: A five-person expert group from the construction industry (all of whom are intermediate or higher-level engineers) will be formed to calibrate ambiguous clauses (such as "evacuation distance should not exceed 40m" without a clear start / end point). Clauses with a consistency rate of ≥80% will be deemed valid, and conditions will be supplemented to clarify them (e.g., "evacuation distance refers to the straight-line distance from the room door to the nearest safe exit, and should not exceed 40m"). Rule node generation – Convert the quadruple into a computer-executable logical expression (e.g., “IF component type = firewall THEN fire resistance rating ≥ 3.00 h”), and assign a unique ID (e.g., “R-GB50016-2014-3.2.9”, including the specification number and clause number). Graph construction: The Neo4j graph database is used to establish associations according to the hierarchy of "standard number → clause node → component type → parameter type" (e.g., "GB 50016-2014" → "R-GB50016-2014-3.2.9" → "firewall" → "fire resistance limit"). Dynamic addition and deletion of nodes and associations are supported.

[0037] Finally, intelligent review and conflict detection are performed. The system dynamically matches the generated lightweight 3D building information model with the standard knowledge graph. It traverses every component in the model and automatically activates relevant computational logic rule nodes from the knowledge graph based on the component type. Subsequently, according to the rule requirements, it extracts the necessary geometric and semantic parameters from the 3D model in real time for calculation and logical judgment. For example, when reviewing fire separation distances, the system accurately calculates the shortest Euclidean distance between two fire compartment walls in 3D space. If the calculation result is 5.5 meters, while the standard requires 6 meters, it is judged as a violation. For all detected violations, the system generates a review conflict record containing the identifier of the violating component, the specific standard clause violated, the calculated value of 5.5 meters, and the standard value of 6 meters, and highlights the violation location in red in the 3D model, while also generating a detailed structured review report.

[0038] The existing technology is an automated rule checking system based on two-dimensional computer-aided design drawings. This system can read two-dimensional drawing files of a specific format and perform batch checks on elements such as layers, line types, and annotations through pre-written scripts.

[0039] The main differences between this invention and the prior art are as follows: Data Dimensions and Review Accuracy: Existing technologies are limited to processing two-dimensional planar information and cannot accurately express and calculate three-dimensional spatial relationships. For example, for the review of fire separation distances, it can only measure projected distances on a two-dimensional planar drawing, which may yield inaccurate results. This invention, by reconstructing a three-dimensional model from two-dimensional drawings, enables precise Euclidean distance calculations in real three-dimensional space, resulting in more accurate and reliable review results.

[0040] Machine understandability of specifications: Existing review rules are typically hard-coded, fixed scripts, lacking flexibility and struggling to handle updates to specification provisions and complex logic. This invention, through natural language processing and knowledge graph technology, transforms textual specifications into structured, flexibly manageable, and reasoning-oriented computational logic nodes, enabling computers to better understand the specifications' meaning and enhancing the system's adaptability and intelligence.

[0041] Information Fusion and Model Value: Existing technologies typically process geometric, layer, and annotation information from drawings in isolation. This invention constructs a lightweight 3D Building Information Model (BIM) containing complete geometric, semantic attributes, and topological relationships through multi-source information fusion. This model is not only used for review but is itself a rich digital asset, unlike the reusable model value found in inspection results generated by existing technologies.

[0042] This invention provides technological improvements that differ from existing technologies in terms of data processing dimensions, depth of standard understanding, and construction and utilization of information models, achieving an advancement from two-dimensional surface inspection to three-dimensional depth understanding and computation.

[0043] In another technical solution, step S2, which involves fusing two-dimensional geometric primitives and text annotation information through a component recognition rule base to generate three-dimensional architectural components with semantic attributes, specifically includes: S211, Component Type Recognition: Based on the layer name in the layer attribute information, combined with the primitive type and spatial distribution in the geometric vector information, the candidate building components and their types are initially identified by matching through a predefined component recognition rule base. The types include walls, beams, columns, slabs, doors, and windows. S212, 3D geometry generation: For the identified candidate building components, the corresponding 3D generation algorithm is called according to their type: For wall and column components, the 2D closed contour is extruded, and the extruded height is determined according to the height value in the bound text annotation information or the default specification value; For beam and slab components, the 3D solid is generated by lofting or filling operations based on their 2D contour lines and the marked elevation information. S213, Topology Relationship Reconstruction: Based on the spatial positions of all generated 3D building components, automatically calculate and record the spatial topology relationships between components, including connection relationships, support relationships, and spatial containment relationships.

[0044] In the above technical solution, after completing the standardized parsing of drawing data, the next stage is multi-source information fusion and 3D building information model reconstruction. This stage aims to solve the technical challenge of automatically fusing discrete two-dimensional geometric primitives with text annotations to generate 3D building components with semantic attributes.

[0045] First, component type identification is performed. The system is based on layer attribute information from standardized data objects, especially layer names, combined with primitive types and spatial distribution characteristics from geometric vector information. It uses a predefined component identification rule base (the rule base is trained and optimized using drawings from 100+ typical architectural projects; core rule examples: Wall: Layer name characteristics: contains WALL / wall body, primitive type: closed polyline, spatial distribution characteristics: length ≥ 2000 mm, width 100-300 mm, additional judgment condition: line type "CONTINUOUS", color RGB (128,128,128); Column: Layer name characteristics: contains "COLUMN" / "column", primitive type: closed polygon, spatial distribution characteristics: side length / diameter 300-800 mm, discrete distribution, additional judgment condition: surface area ≥ 0.09 m²). 2 No continuous extension features; Door: Layer name features: contains "DOOR" / "door", element type: block reference / line segment combination, spatial distribution features: width 600-1200 mm, attached to wall components, additional judgment condition: distance from wall edge ≤10mm, label contains "M" prefix; Window: Layer name features: contains "WINDOW" / "window", element type: block reference / line segment combination, spatial distribution features: width 800-2400mm, attached to wall components, additional judgment condition: distance from wall edge ≥300 mm, label contains "C" prefix; Beam: Layer name features: contains "BEAM" / "beam", element type: polyline, spatial distribution features: length ≥3000 mm, width 200-500 mm, additional judgment condition: layer classification code "02", elevation contains "+"; Slab: Layer name features: contains "SLAB" / "slab", element type: closed polygon, spatial distribution features: area ≥1.0 m² 2 The system uses a planar distribution with additional criteria: elevation includes "-" / "±", and a default thickness of 100 mm for unmarked thicknesses. For example, if the system detects a closed polyline on a layer named "A-WALL" or "Wall" with a continuous, narrow rectangle shape, it is initially identified as a candidate "Wall" component. Similarly, elements on the "A-DOOR" layer composed of specific blocks or line segments are identified as "Door" components. In this way, the system can initially filter and classify major building components such as walls, beams, columns, slabs, doors, and windows from a chaotic set of two-dimensional primitives.

[0046] Next, 3D geometry generation is performed. For identified 2D components, the system calls the corresponding 3D generation algorithm based on their type. For wall and column components, the system obtains their 2D closed contours, such as a wall contour line with a thickness of 200 mm, and then extrudes it according to the height value in the attached text annotation information. If the annotation explicitly states a height of 3600 mm, it is extruded according to this value; if not annotated, the default specification value, such as 2800 mm, is used to generate a 3D solid. For beam and slab components, the system operates based on their 2D contour lines and the elevation information of the annotations. For example, the contour line of a beam combined with its labeled elevation "++2.500" generates a 3D beam solid at the corresponding height through a lofting algorithm; the contour line of a closed room combined with the floor slab elevation generates a 3D floor slab solid through a filling operation.

[0047] Finally, topological relationship reconstruction is performed. After all 3D building components are generated, the system automatically analyzes their spatial locations, calculates and records the spatial topological relationships between components. For example, the system records which wall is connected to which column in space, which beam supports which floor slab, and which door or window is contained within which wall. These connection relationships, support relationships, and spatial inclusion relationships are explicitly recorded in the model, forming an interconnected network between components, providing an important spatial relationship data foundation for subsequent intelligent review.

[0048] Existing technology is layer management and block referencing technology in a two-dimensional computer-aided design environment. This technology manages different types of components by placing specific elements on named layers or creating block definitions, but its essence is still two-dimensional representation.

[0049] The main differences between this invention and the prior art are as follows: The conversion from two-dimensional symbols to three-dimensional entities: Existing technologies remain at the two-dimensional level, with layers and block references merely serving as visual classification symbols and lacking three-dimensional geometric attributes. This invention achieves automated and parametric generation from two-dimensional primitives to three-dimensional entities, creating realistic spatial volumes and enabling three-dimensional spatial analysis.

[0050] Deep binding of semantic information: In existing technologies, annotation text and geometric primitives are usually separate and have weak correlation. In the 3D generation process, this invention strongly binds text annotation information such as dimensions, height, and elevation to the generated geometric entities, so that the 3D components not only have shape, but also carry key engineering semantic information.

[0051] Automated Construction of Spatial Topological Relationships: Existing technologies cannot automatically identify and record the spatial logical relationships between two-dimensional elements. This invention, after generating three-dimensional components, actively calculates and reconstructs rich topological relationships between components, such as connections, supports, and inclusions, constructing an interconnected intelligent model, rather than simply a collection of geometric shapes.

[0052] The above-mentioned technical solution of the present invention effectively solves the core technical problem of transforming abstract two-dimensional drawings into concrete and computable three-dimensional information models, laying a solid data foundation for subsequent automated intelligent review.

[0053] In another technical solution, step S2, the aggregation into a lightweight 3D building information model specifically includes: S221, Semantic Information Injection: The dimensions, materials, and fire rating attributes related to the components in the text annotation information are associated and bound one-to-one or one-to-many with the three-dimensional building components generated in step S212 to complete the injection of semantic information. S222, Lightweight Model Encapsulation: An incremental lightweight algorithm is adopted to remove intermediate data and historical operation records in the 3D modeling process, while retaining the final geometric data, semantic attribute data and topological relationship data between building components, and encapsulating them into a lightweight 3D building information model. The format of the lightweight 3D building information model is independent of any specific computer-aided design software environment.

[0054] In the above technical solution, after completing the geometric generation and topological relationship reconstruction of the 3D building components, the final stage of model aggregation and optimization is reached. This stage aims to solve technical problems such as the bloated nature of the generated 3D model data, dependence on specific software environments, and lack of machine-readable semantic information.

[0055] First, semantic information is injected. The system precisely associates the previously parsed text annotations with the generated 3D building components. This is a crucial data fusion step. For example, the system binds a dimension labeled "300 millimeters" to the corresponding wall thickness attribute; assigns a description labeled "Class A fire door" to the corresponding door component as its fire rating attribute; and associates a material description labeled "C30 concrete" with a specified beam or column component. These associations can be one-to-one, such as a dimension label belonging to only one wall; or they can be one-to-many, such as a material label applying to multiple columns simultaneously. Through this process, the originally geometric 3D components are endowed with rich, computer-recognizable, and processable engineering semantics, becoming true information carriers.

[0056] The model is then lightweighted and packaged. The system performs final optimization on the complete model data, which integrates geometric, topological, and semantic information. It employs an incremental lightweighting algorithm to systematically remove all intermediate data, historical construction records, and redundant geometric details generated during the 3D modeling process. Ultimately, the model retains only the core elements of each building component: the final geometric data, the injected semantic attribute data, and the topological relationship data between components. After this processing, a detailed model that might have occupied hundreds of megabytes of storage space is packaged into a lightweight 3D Building Information Model that occupies only tens of megabytes of storage space. This model format is designed to be independent of any specific computer-aided design software environment, becoming a neutral and efficient data packet that facilitates rapid transmission in network environments, seamless loading across different system platforms, and direct use by the subsequent intelligent review engine.

[0057] Existing technology involves creating architectural models using general-purpose 3D modeling software such as Autodesk Revit and SketchUp. These models are typically used directly for design, presentation, and drawing.

[0058] The main differences between this invention and the prior art are as follows: Data Composition and Purpose Orientation: Models created by general-purpose 3D modeling software contain a large amount of redundant data used for visualization, editing, and historical review, resulting in large model files. This invention, through lightweight encapsulation, proactively eliminates all non-core data, generating a dedicated model highly optimized for "computation and analysis," with a small data volume and fast processing speed.

[0059] Explicit Semantic Expression and Integration: In general software, semantic information (such as materials and types) may be stored in independent parameter tables, with complex logic relating to the geometric model, making it difficult for external programs to directly parse. This invention, through a semantic information injection step, strongly binds semantic attributes as directly accessible data fields to geometric components, forming a highly integrated, self-interpreting data structure, which greatly facilitates the direct extraction and utilization of information by review procedures.

[0060] Platform Dependency and Interoperability: Models generated by general-purpose software are typically deeply bound to their native software environment, requiring complex conversions for use in other systems or web environments and prone to information loss. The lightweight model generated by this invention is essentially platform-neutral; its data structure does not depend on any specific commercial software, ensuring seamless integration and smooth operation in subsequent intelligent review processes, achieving better interoperability and system independence.

[0061] The above-mentioned technical solution of the present invention effectively solves the data gap between 3D models and "design visualization" and "intelligent analysis application", generating an ideal data carrier that combines information integrity, processing efficiency and platform independence.

[0062] In another technical solution, step S4 specifically includes: S410, Review Rule Trigger: Traverse each building component entity and its semantic attributes in the lightweight 3D building information model, and automatically match and activate the associated computational logic rule nodes from the specification knowledge graph according to the component type; S420, Parametric Calculation and Logical Judgment: Based on the activated calculation logic rule nodes, the required geometric and semantic parameters are extracted from the lightweight 3D building information model in real time, and parametric mathematical calculations and spatial relationship logical judgments are performed. S430. Review Result Generation and Visual Annotation: The results of parametric calculation and logical judgment are compared with the thresholds or conditions in the rule nodes. If a rule is violated, a review conflict record is generated, which includes the identifier of the violating component, the specific code clause violated, the calculated value of the violating parameter, and the standard requirement value. All review conflict records are visualized and highlighted in three-dimensional space in the lightweight three-dimensional building information model, and a structured review report is generated.

[0063] In the aforementioned technical solution, after completing the 3D model reconstruction and the construction of the specification knowledge graph, the process enters the intelligent review and conflict detection stage based on dynamic rule matching. This stage aims to solve the technical challenge of how to automatically and accurately apply abstract specification provisions to specific building models and generate actionable review results.

[0064] First, the review rules are triggered. The system traverses every building component entity and its semantic attributes in the lightweight 3D Building Information Model. For each component, the system automatically matches and activates associated computational logic rule nodes from the specification knowledge graph based on its type. For example, when the system traverses a component identified as a "room," it automatically activates rule nodes related to "daylighting calculation" and "area index"; when it traverses a component identified as a "fire door," it automatically activates rule nodes related to "fire resistance rating" and "opening direction." This process achieves a dynamic and precise association between review rules and model components.

[0065] Next, parametric calculations and logical judgments are performed. Based on the activated calculation logic rule nodes, the system extracts the necessary geometric and semantic parameters from the lightweight 3D building information model in real time, and performs precise mathematical calculations and spatial relationship logical judgments. For example, for the rule "the clear height of the stair platform should not be less than two meters," the system automatically extracts the bottom elevation of the relevant beams and the platform elevation from the stair components in the model, and calculates the vertical clear height. If the calculated clear height is 1.9 meters, it is judged as a violation. As another example, for the rule "the fire resistance rating of the firewall between two fire compartments should not be less than three hours," the system extracts the "fire resistance rating" semantic attribute value of the wall components. If this value is two hours, it is judged as not meeting the requirement.

[0066] Finally, the review results are generated and visualized. The system automatically compares the results of parametric calculations and logical judgments with the preset thresholds or conditions in the rule nodes. For all rule violations, the system generates a structured review conflict record. This record clearly includes the unique identifier of the violating component, the specific code clause number violated, the calculated actual parameter value (e.g., 1.9 meters), and the standard value required by the rule (e.g., 2 meters). Simultaneously, the system highlights the 3D spatial location of all violating components in the lightweight 3D building information model, for example, rendering stair beams with insufficient clearance in red. Finally, the system summarizes all conflict records and generates a structured review report containing text, data, and screenshots, allowing designers to locate and rectify problems.

[0067] Existing technology is a keyword search and simple rule-based batch processing inspection system based on two-dimensional drawings. This system typically runs in a two-dimensional CAD environment, identifying problems by searching specific layers or annotation text, or by performing simple distance and area measurements.

[0068] The main differences between this invention and the prior art are as follows: Review depth and context awareness: Existing technologies mostly involve stateless batch processing, lacking awareness of component types and context. This invention achieves targeted dynamic rule matching based on component semantics through rule-triggered steps, giving the review process intelligent context awareness capabilities.

[0069] Accuracy of Calculation and Three-Dimensional Spatial Representation: Existing technologies measure on two-dimensional drawings, which cannot accurately handle three-dimensional spatial problems (such as net height). This invention performs spatial calculations and parameter extraction directly on the three-dimensional model. The calculation results accurately reflect the real spatial conditions, fundamentally avoiding the errors caused by two-dimensional projection.

[0070] The operability and intuitiveness of the results: Existing technologies typically output text logs or tables, requiring users to locate the problem on the drawings themselves. This invention directly links the review results (conflict records) with their three-dimensional spatial location in the model and presents them through visual highlighting and structured reports, making the problem location clear at a glance and greatly improving the operability and communication efficiency of the review results.

[0071] In another technical solution, before dynamically matching and reasoning the lightweight 3D building information model with the standard knowledge graph in step S4, the method further includes: dynamically filtering the target component set or target area from the lightweight 3D building information model based on the user-input review scope selection instruction, and automatically expanding the target review domain based on the component topology relationship; the dynamic matching and reasoning is only performed on the target component set or target area.

[0072] In the above technical solution, an optimized mode is provided to improve review efficiency before initiating a comprehensive intelligent review. This mode aims to solve the problem of resource waste and response delay caused by the system still needing to perform a full model scan when the user only cares about a specific part of the building model.

[0073] After the system generates a lightweight 3D building information model, users can interact with the model through a graphical interface. For example, when designing a 20-story office building, if the designer has just modified the floor plan of the fifth floor, they can select only the "fifth floor" option in the 3D model browser. The system then receives this user-input instruction to select the scope of review.

[0074] Based on this instruction, the system first dynamically filters all components located on the fifth floor from the complete building model, forming a preliminary set of target components. Then, the system initiates an intelligent analysis process, automatically expanding the target review domain based on the previously established component topological relationships. Specifically, it analyzes the connection relationships between the fifth-floor components and components on other floors. For example, the fifth-floor staircase must be connected to the stair landings on the fourth and sixth floors; a pipe shaft running through multiple floors has an access door on the fifth floor; and the load-bearing columns on the fifth floor may extend upwards to the sixth floor. These key components on adjacent floors that have a direct "connection" or "support" relationship with the user-selected fifth-floor components are automatically included in the final target review domain.

[0075] Once the target review domain is constructed, the system's subsequent intelligent review engine will only work on this precisely defined set of target components. It will traverse every component in this set, performing rule matching, parameter calculation, and logical judgment, while completely skipping the tens of thousands of components in the other fifteen unselected floors. For a large project, this optimization can significantly reduce review time from thirty minutes for the entire model to less than two minutes for the target area.

[0076] The existing technology is the "layout" or "viewport" isolation function in the review of two-dimensional drawings. This function allows users to manually turn the display of different layers on and off, but the review process still requires a global or manual selection check on the displayed two-dimensional elements.

[0077] The main differences between this invention and the prior art are as follows: Selection Logic Dimensions and Intelligence: Existing technologies are based on two-dimensional layer switches, resulting in a coarse selection range and an inability to automatically identify spatially related components across layers. This invention allows selection in three-dimensional space and can intelligently expand the review scope based on the actual topological relationships between components (such as connections and supports), ensuring the integrity and accuracy of local reviews and preventing the oversight of related issues due to human selection errors.

[0078] A fundamental improvement in review efficiency: Existing technologies, even with certain layers disabled, may still require the review process to process all data in the background, or necessitate users manually defining a review area—a cumbersome process. This invention generates a precise, computer-recognizable target set through "dynamic filtering" and "topology expansion," enabling the review engine to fundamentally avoid any calculations on irrelevant data, achieving a leap in efficiency.

[0079] Naturalness of operation and understanding of intent: Existing technologies require users to have a high level of knowledge of drawings to accurately switch layers on and off. This invention allows users to directly select physical spaces (such as floors or areas) on an intuitive 3D model. The system automatically understands the user's review intent and converts it into a precise data set, lowering the operational threshold and improving the user-friendliness of human-computer interaction.

[0080] In another technical solution, step S4, after generating the review conflict record, also includes a step of intelligent classification and grading of review conflicts, specifically: Based on a predefined conflict severity level rule base, each review conflict record is automatically classified, with conflict severity levels including at least a major risk level, a general violation level, and a recommended optimization level; At the same time, the violations of the normative provisions in the review conflict record are automatically classified according to their professional fields, which include at least fire safety, structural safety, energy conservation and environmental protection, and ergonomics. In the generated structured review report and 3D visualization annotations, the review conflict records are sorted, filtered, and displayed in a differentiated manner according to the severity level of the conflict and the professional field classification.

[0081] In the above technical solution, after the system generates review conflict records, an intelligent classification and categorization step for review conflicts is added. This step aims to solve the technical challenge of users finding it difficult to quickly identify key issues and efficiently allocate processing priorities when reviews generate a large number of results.

[0082] The system has a built-in predefined rule base for classifying conflict severity levels. After generating a conflict review record, the system immediately classifies each record according to this rule base. For example, any issue involving insufficient cross-sectional dimensions of load-bearing structural members, affecting safe evacuation width or distance, will be automatically marked as a "major risk level." Such issues may directly endanger personal safety and must be addressed with priority. Violations that do not comply with mandatory provisions but do not immediately pose a safety risk, such as a room area slightly exceeding the allowable tolerance, are marked as a "general violation level." Suggestions that only involve recommended standards or best practices, such as further optimization of lighting in certain areas, are categorized as "suggested optimization level."

[0083] Meanwhile, the system automatically categorizes conflicts into different professional fields based on the specific regulations violated in the conflict records. For example, conflicts violating the "Code for Fire Protection Design of Buildings" are classified as "Fire Safety"; issues involving component load-bearing capacity and reinforcement ratio fall under "Structural Safety"; those related to thermal insulation and heat insulation performance are classified as "Energy Conservation and Environmental Protection"; and issues concerning stair tread height and bathroom facility layout fall under "Ergonomics".

[0084] After completing the grading and classification, the system optimizes the presentation of the results in the final structured review report and 3D visualization interface. In the review report, all conflict records are first arranged in the order of "Major Risk," "General Violation," and "Recommended Optimization," allowing users to see the most pressing issues first. Simultaneously, the report provides a function to filter by professional field, allowing structural engineers to focus on "structural safety" issues, while equipment engineers can quickly filter for "energy conservation and environmental protection" issues. In the 3D model, different levels of conflicts are marked with different colors; for example, "Major Risk" is highlighted in red, "General Violation" is indicated in yellow, and "Recommended Optimization" is indicated in blue, thus achieving differentiated display.

[0085] Existing technology is an inspection system that outputs a simple list or table, which indiscriminately lists all detected problems.

[0086] The main differences between this invention and the prior art are as follows: Value density and decision support capability of the results: Existing technologies provide an unprocessed "raw problem list," requiring users to spend a significant amount of time manually reading and judging the importance of each problem. This invention, through automatic hierarchical classification, transforms the problem list into a "processing order" with clear priorities, directly guiding the user's subsequent actions and significantly improving the value density and decision support capability of the results.

[0087] Automation and specialization of problem classification: In existing technologies, users need to manually interpret the professional affiliation of a problem based on regulatory clauses. This invention, through automatic classification, accurately categorizes problems into different professional fields, saving users classification time and, more importantly, facilitating collaboration among different professions in large projects, achieving intelligent matching of problems with professional roles.

[0088] Multidimensional and intuitive information presentation: Existing technologies present results in a single way. This invention combines multiple methods such as list sorting, professional filtering, and three-dimensional visual color coding to present review results from multiple dimensions, making the structure and key points of massive amounts of information clear at a glance, greatly reducing the user's cognitive load, and improving overall review efficiency and user experience.

[0089] In another technical solution, the intelligent review method for architectural construction drawings also includes a model optimization feedback loop independent of the core review process. This model optimization feedback loop is executed periodically or triggered by specific events, including: Construction of historical conflict sample library: After the user confirms or corrects the review conflict record generated in step S4, the finally confirmed conflict case, including its corresponding model component data, triggered calculation logic rule node and final judgment result, will be stored as annotated sample in the historical conflict sample library. Rule confidence adaptive adjustment: Based on the historical conflict sample library, the trigger frequency and false alarm rate of each calculation logic rule node are statistically analyzed. When the false alarm rate of a rule is consistently higher than the threshold of 10%, the system automatically reduces the confidence weight of the rule when it is triggered, or gives the user an uncertainty prompt when it is triggered. New rule pattern mining: When the case in the historical conflict sample library is found to have frequent coexistence of two or more types of component violations in the sample, and their combination pattern is not covered by the existing normative knowledge graph, the potential new rule pattern is submitted to the administrator for review as an optimization suggestion, so as to expand or correct the normative knowledge graph.

[0090] In the above technical solution, an independently operating model optimization feedback loop is established outside the core review process. This loop aims to address the technical challenge that the system, due to its reliance on an initial static rule base, may gradually become rigid and unable to learn from practice and adapt to dynamic changes in regulations.

[0091] This feedback loop can be set to execute automatically once a week, or manually triggered after the canonical knowledge graph is manually updated.

[0092] First, the system builds a historical conflict sample library. After users process the conflict records generated by the intelligent review—for example, if a user confirms that a record stating "fire door width is less than 800 mm" is true, or corrects a false alarm where the system misclassified a pipe shaft as a room—the system will store these manually verified final results, along with their corresponding model component data and the triggered calculation logic rule node information, as high-quality labeled samples in the historical conflict sample library. For example, a sample may contain the geometric data of the confirmed non-compliant fire door component, the triggered "evacuation door clear width" rule node ID, and the final judgment result marked as "genuine non-compliance" by the user.

[0093] Subsequently, the system adaptively adjusts the rule confidence level. It periodically analyzes the sample database and statistically analyzes the performance of each calculation logic rule node. The system calculates the false alarm rate for each rule, which is the proportion of times the rule is ultimately judged as a false alarm by the user after being triggered, out of the total number of triggers. For example, suppose a rule about "net height of the pipe layer" is triggered 50 times in the past month, and eight of them are confirmed as false alarms by the user (such as mistakenly including the equipment mezzanine in the living space), with a false alarm rate of 16%. Since this value is consistently higher than the system's set threshold of 10%, the system will automatically lower the confidence weight of this rule when it is triggered in a match from the initial 0.9 to 0.6. When this rule is triggered again in the next review, the system may give the user an uncertainty prompt in the report: "This result may be a false alarm; please review it carefully."

[0094] Finally, the system performs new rule pattern mining. It uses association rule mining algorithms to analyze the inherent relationships among a large number of cases in the sample database. For example, the analysis found that in over 200 samples, when a project simultaneously used "a single-layer, 12mm thick Type A glass curtain wall" and "a Type B natural ventilation system," up to 95% of the cases ultimately violated a mandatory clause regarding "overall building energy efficiency," even though the initial rule base did not explicitly define a direct link between this combination and energy-saving regulations. The system will submit this potential new rule pattern—"When Type A glass curtain walls and Type B ventilation systems are used in combination, a mandatory special energy-saving calculation is required"—as an optimization suggestion to the system administrator for review. After administrator confirmation, this new rule pattern will be used to expand and revise the regulatory knowledge graph.

[0095] Existing technology consists of review systems with a fixed rule base that allow administrators to manually add or modify rules. System optimization relies entirely on external human input.

[0096] The main differences between this invention and the prior art are as follows: The driving force for optimization: Optimization in existing technologies is passive, relying on administrators to discover problems externally and manually update rules. The optimization in this invention is proactive and data-driven; the system can automatically discover problematic rules and uncover potential patterns based on internally accumulated historical data, achieving self-evolution.

[0097] The fineness of rule application: Existing technologies typically only have two states for rules: "enabled" or "disabled". This invention introduces the concept of "confidence weight", which allows for fine-grained and quantitative adjustment of rule application. This enables effective management of the risks (such as high false positives) associated with rules without completely disabling them, making the system behavior more intelligent and robust.

[0098] Knowledge discovery capability: Existing technologies cannot automatically discover new knowledge or rule patterns. This invention, through association rule mining, can automatically identify undefined and complex combinations of violation patterns from past conflict cases, possessing the ability to assist and even guide the expansion of the rule base, enabling the system to cope with more complex and hidden design problems.

[0099] In another technical solution, the intelligent review method for architectural construction drawings also includes an incremental review mode, which is triggered when an update to the specification knowledge graph or a partial revision to the lightweight 3D building information model is detected. Specifically, this includes: Impact domain analysis of changes: When the specification knowledge graph is updated, the newly added or modified calculation logic rule nodes are parsed to determine the set of building component types that are affected by them; when the 3D model is partially revised, the set of building components that have been added, deleted, or modified is located; among them, when the 3D model is partially revised, the impact domain is located and expanded through the following steps: Component Revision Location: A hash value comparison algorithm is used to generate a unique hash value for the geometric parameters (coordinates, dimensions) and semantic attributes (material, fire rating) of each component. The hash values ​​of the components in the model before and after revision are compared. Components with different hash values ​​or those that are added / missing are identified as "revision components" and their unique ID and revision type (addition / deletion / modification) are recorded. Related component expansion: Based on the pre-built "component topology relationship table" (which stores the "connection / support / containment" relationships between components, e.g., the partition wall "W058" and beam "B12" have a "connection" relationship, and the partition wall "W058" and door "D023" have a "containment" relationship), the directly related components of the revised component are automatically queried and included in the influence domain; Influence domain verification: For the expanded influence domain, a second screening is performed using "distance judgment" (the spatial distance between the associated component and the revised component is ≤100mm) to remove incorrectly associated components (e.g., when revising the 5th floor partition wall, exclude the 1st floor irrelevant beam component). Example: If the revised component is "W058" (5th floor partition wall, change position), then the set of the revised components is {"W058"}, the set of related components is {"B12" (beam), "D023" (door)}, and the final influence domain is {"W058" "B12" "D023"}.

[0100] Dynamic construction of the target review domain: The set of affected component types obtained from the above analysis is combined with the set of components that have been revised to generate a target review domain; Targeted rule matching and review: For components within the target review domain, perform dynamic rule matching and reasoning, parameterized calculation and logical judgment, and generate review conflict records.

[0101] The above technical solution provides an incremental review mode, which aims to solve the problem of wasted computing resources and review delay caused by the system still needing to perform a full model review when building codes are updated or the 3D model is partially revised.

[0102] This mode is automatically triggered in two situations: first, when a version update of the standard knowledge graph is detected; and second, when a partial revision of the lightweight 3D building information model is detected and saved.

[0103] The specific implementation process is as follows: First, an impact domain analysis is performed. When the specification knowledge graph is updated, the system parses all newly added or modified calculation logic rule nodes to determine the set of building component types that are affected by them. For example, if the new fire protection code increases the requirement for the width of evacuation staircases in densely populated areas from 1.2 meters to 1.4 meters, the system will parse that this rule node is associated with all "evacuation staircase" component types. When a partial revision occurs to the 3D model, the system uses version comparison to accurately locate the set of building components that have been added, deleted, or modified. For example, the designer may have only changed the positions of three partition walls on the fifth floor plan and updated the model.

[0104] Next, the target review domain is dynamically constructed. The system performs a union operation on the set of affected component types obtained from the above analysis and the set of components that have undergone revision, generating the final target review domain. Continuing with the example above: the specification update affects all "evacuation staircases" components; the model revision involves "three partition walls" on the fifth floor. The target review domain obtained by the system through the union operation is "all evacuation staircases in the entire building" plus "the three partition walls on the fifth floor." This step ensures that the review scope covers both all components that may be affected by the specification change and all actually modified components.

[0105] Finally, targeted rule matching and review are performed. The system only initiates the review process for components within the target review domain. In this example, the system will traverse all evacuation staircases in the building, checking whether their width meets the new 1.4-meter requirement; it will also check whether the modified partition wall on the fifth floor complies with relevant specifications. For the vast majority of unchanged components in the model that are not affected by specification updates, the system completely skips the review process. In a model containing 3,000 components, if the target review domain contains only 50 components, the incremental review workload is only one-sixtieth of the full review workload, reducing the review time from 30 minutes to approximately 30 seconds.

[0106] Existing technology is a review system with version control functionality. However, even after detecting changes to the model or specification, this system still needs to initiate a complete review process to ensure overall compliance.

[0107] The main differences between this invention and the prior art are as follows: Precision of review granularity: Existing technologies, after detecting changes, know that "there has been a change," but cannot precisely analyze "what has changed" and "what the change has affected," thus resorting to a conservative full-scale review strategy. This invention, through change impact domain analysis, can accurately identify the specific content and scope of the change, achieving a shift from "full review of all changes" to "precise positioning and targeted review."

[0108] The essence of resource utilization efficiency: Existing version control technologies mainly address data storage and rollback issues, without optimizing the resource consumption of the review process itself. This invention, by constructing a precise target review domain, enables the review engine to fundamentally avoid redundant calculations on massive amounts of unaffected components, achieving a positive correlation between computational resource consumption and the scale of changes, rather than a positive correlation with the total model size.

[0109] Response speed and practicality: In the current full-model review mode, any minor modification requires a lengthy wait comparable to the initial review, making it difficult to support rapid design iteration. This invention shortens the response time to a level commensurate with the scale of the change through incremental review, enabling a rapid design cycle of "modification-review-re-modification," greatly improving the system's practicality in real-world environments.

[0110] In another technical solution, in step S4, while generating each review conflict record, a corresponding review logic traceability description is automatically generated; the process of generating the review logic traceability description includes: Automatically record the original specification clause content of the computational logic rule node that triggered the conflict; The key model parameters and their values ​​used in this determination are automatically extracted. The key model parameters include at least the geometric parameters and semantic parameters extracted in real time from the lightweight 3D building information model. Based on rule nodes and key model parameters, a natural language description text is automatically generated, which includes the identifier of the non-compliant component, the violated normative clause, the specific parameter values ​​used, the logical judgment process, and the final conclusion. The review logic traceability description and the corresponding review conflict record are associated and stored, and then output to the structured review report and the 3D visualization interface.

[0111] In the above technical solution, an automatic generation process for review logic tracing explanations is executed simultaneously with the generation of each review conflict record. This step aims to address the technical challenge of system review results being too conclusive and lacking a demonstration of the reasoning process, making it difficult for designers to quickly understand the root cause of violations.

[0112] The specific implementation process is as follows: When the system determines that a component violates regulations and generates a conflict record, it simultaneously activates the description generation mechanism. This mechanism first automatically records the original code clause content corresponding to the calculation logic rule node that triggered the current conflict. For example, it records the complete text of a certain article in the "Code for Fire Protection Design of Buildings" stating that "Evacuation corridors shall be equipped with Class A fire doors at fire compartments".

[0113] Next, the system extracts key calculation parameters and their specific values ​​from the model data upon which this judgment is based. These parameters include at least the geometric and semantic parameters obtained in real time from the 3D model. For example, for a conflict judged as "fire door rating mismatch", the system will extract the semantic attribute of the door component "fire rating: Class B" and the semantic attribute of its location "fire compartment: firewall".

[0114] Then, based on the aforementioned rule node content and key model parameters, the system automatically generates a complete and clearly descriptive natural language text. This text explicitly includes the unique identifier of the non-compliant component, the violated code clause, the specific parameter values ​​used in the judgment, the logical reasoning process, and the final conclusion. Taking the aforementioned fire door as an example, the generated explanatory text might be: "The component identified as 'Fire Door-108' has a semantic attribute 'Fire Rating' of 'Class B,' while according to a certain article of the 'Code for Fire Protection Design of Buildings,' doors located at firewalls should be 'Class A fire doors.' Upon comparison, 'Class B' does not meet the requirements of 'Class A,' therefore it is determined to be non-compliant." Finally, the system associates and stores this review logic traceability description with the corresponding review conflict record, and outputs them together to the final structured review report and 3D visualization interface. When designers view the review report or click on the highlighted non-compliant components in the 3D model, this description will be presented directly as detailed information, clearly demonstrating the system's entire judgment logic.

[0115] Existing technology is a review system that only provides error codes, simple prompts, or simply lists the numbers of the violated specification clauses.

[0116] The main differences between this invention and the prior art are as follows: Information Content and Operability: Existing technologies provide limited information in error codes or simple conclusions (such as "level mismatch"), requiring users to manually consult specifications and component parameters to understand the problem. The traceability instructions provided by this invention integrate all core information—the violating object, specification content, key data, and judgment logic—forming a self-contained, independently understandable report unit. This allows users to fully understand the problem details without additional searching, greatly improving the operability and rectification guidance value of the information.

[0117] Transparency and Credibility of the Process: Existing review processes resemble a "black box," where users can only see the inputs and final conclusions, but cannot know the basis for the judgments, thus reducing the credibility and acceptability of the results. This invention significantly enhances the transparency of the system's operation and the persuasiveness of the review conclusions by "white-boxing" the review reasoning process using natural language, helping to build user trust in the intelligent review system.

[0118] Efficiency of Problem Identification and Communication: In existing technologies, designers need to spend a significant amount of time analyzing the root causes of problems themselves. This invention directly provides a complete "diagnostic report," enabling designers to instantly understand the crux of the problem. This not only speeds up the identification of individual problems but also facilitates clear and unambiguous communication and review of findings within the project team, thereby improving the efficiency of collaborative rectification.

[0119] This invention also provides a system for an intelligent review method of architectural construction drawings, comprising: The data parsing module is configured to receive original electronic construction drawings generated by different computer-aided design software through a data interface, and parse and convert them into standardized structured data objects. The 3D model reconstruction module is configured to build a lightweight 3D building information model that integrates geometric and semantic information based on standardized structured data objects and component recognition and 3D generation algorithms. The standardized knowledge graph module is configured to store and manage structured computational logic rule nodes transformed through natural language processing technology, and to form a standardized knowledge graph. The intelligent review engine module is configured to dynamically match and reason with the lightweight 3D building information model and the standard knowledge graph, perform parametric calculations and logical judgments, and generate review conflict records. The review scope selection module is configured to receive user instructions and dynamically filter target component sets or target areas from the lightweight 3D building information model for targeted review by the intelligent review engine module. The conflict handling module is configured to automatically classify the severity level and professional field of the review conflict records generated by the intelligent review engine module. The model optimization feedback module is configured to build a historical conflict sample library and adaptively adjust the rule confidence and mine potential new rule patterns based on it in order to optimize the canonical knowledge graph module. The incremental review control module is configured to automatically analyze the impact domain of the change and construct the target review domain when a model revision or specification update is detected, thereby triggering the incremental review mode. The logic traceability generation module is configured to automatically generate a natural language description for each review conflict record, including the original specification clauses, key parameters, and the judgment process.

[0120] In the above technical solution, the system is deployed on a server equipped with at least one eight-core processor, 32 gigabytes of memory and a professional-grade graphics workstation graphics card, and achieves intelligent review through the collaborative work of multiple functional modules.

[0121] The data parsing module receives drawing files generated by various design software such as AutoCAD and Revit through a standard application programming interface (API). This module has built-in format parsers that can uniformly convert files in different formats, such as DWG and RVT, into standardized JSON data objects containing geometric information, layer information, and text annotation information. A 100-megabyte original drawing file is parsed to generate a structured data object of approximately 50 megabytes.

[0122] After receiving standardized data objects, the 3D model reconstruction module activates the component recognition engine. Based on two hundred predefined component recognition rules, this engine automatically identifies component types such as walls, beams, and columns, and converts the 2D data into a 3D model using a geometric generation algorithm. For a building project containing five thousand components, the reconstruction process takes approximately three minutes, and the generated lightweight 3D model data size is controlled to within eighty megabytes.

[0123] The specification knowledge graph module stores over a thousand building code provisions, which are then converted into executable computational logic rules using natural language processing technology. For example, fire separation distance requirements are converted into rule nodes that can calculate spatial distances, forming a specification knowledge graph containing approximately two thousand nodes.

[0124] The intelligent review engine module uses parallel computing technology to match 3D models with a specification knowledge graph. This engine can process rule matching and parameter calculations for approximately one hundred components per second; in a project with five thousand components, a full review takes about fifty seconds.

[0125] The review scope selection module provides a graphical interface, allowing users to specify the review scope by checking floors or selecting areas. For example, after selecting the fifth to eighth floors of a building, the system automatically filters out approximately eight hundred target components for targeted review.

[0126] The conflict resolution module automatically classifies and categorizes the review results, marking conflicts into three levels: major risks, general violations, and recommendations for optimization, and classifying and archiving them according to six professional areas, including fire safety and structural safety.

[0127] The model optimization feedback module automatically analyzes historical review data weekly. When the false positive rate of a rule exceeds 10% for four consecutive weeks, it automatically reduces its confidence weight and sends optimization suggestions to the administrator.

[0128] The incremental review control module monitors changes to the model and specifications in real time. When a local modification is detected, it automatically analyzes the scope of the change and converts the full model review into an incremental review targeting approximately 5% of the changed area, reducing the review time from 50 minutes to 3 minutes.

[0129] The logic traceability generation module automatically generates a natural language description for each conflict record, which includes the standard clauses, parameter data, and decision logic. On average, each description contains about 150 Chinese characters, fully explaining the review and reasoning process.

[0130] Existing technologies are based on single-function software review toolsets, which typically operate independently and lack systematic integration.

[0131] The main differences between this invention and the prior art are as follows: System integration and collaboration: Existing technologies consist of multiple independent tools, requiring manual import and export of data between different software, creating information silos. This invention achieves seamless connection between functional components through a modular architecture, allowing data to flow automatically between modules, forming a complete automated processing pipeline, eliminating efficiency bottlenecks and error risks caused by manual intervention.

[0132] The level and depth of intelligence: Most existing technologies automate the checking of basic rules but lack advanced intelligent functions. This invention adds deep intelligent modules such as intelligent hierarchical classification, adaptive optimization, and incremental review on top of basic review, enabling the system to have the ability to self-optimize and execute accurately, evolving from a tool-level application to an intelligent system-level application.

[0133] System architecture scalability and maintainability: Existing technologies are mostly monolithic applications, making functional expansion difficult. The modular design of this invention allows each function to be upgraded and maintained independently. For example, updating the canonical knowledge graph module does not require modifications to other modules, greatly improving the system's maintainability and the flexibility of technological iteration, laying a solid foundation for subsequent functional expansion.

[0134] Although the technical solutions of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and embodiments shown and described herein.

Claims

1. A method for intelligent review of architectural construction drawings, characterized in that, Includes the following steps: S1. Standardized Input and Parsing of Heterogeneous Drawing Data: The system receives original electronic architectural construction drawings generated by different computer-aided design software via a data interface. It then calls the format parsing module to uniformly parse and convert these original electronic architectural construction drawings into standardized structured data objects containing geometric vector information, layer attribute information, and text annotation information. Geometric vector information includes the coordinates and topological relationships of points, lines, surfaces, and entities; layer attribute information includes the layer name, line type, and color of the component; and text annotation information includes dimension annotations, text descriptions, and their associated geometric objects. S2. Multi-source information fusion and 3D building information model reconstruction: Based on standardized structured data objects, two-dimensional geometric primitives and text annotation information are fused through a component recognition rule base to generate three-dimensional building components with semantic attributes, which are then aggregated into a lightweight three-dimensional building information model. S3. Knowledge graph processing of computer-readable standard provisions: Pre-store national and industry mandatory standard provisions related to architectural construction drawings, and convert them into structured, computer-executable computational logic rule nodes through natural language processing technology to form a standard knowledge graph; Step S4: Intelligent review and conflict detection based on dynamic rule matching: The lightweight 3D building information model is dynamically matched and reasoned with the standard knowledge graph. The lightweight 3D building information model is parametrically calculated and logically judged according to the calculation logic rule nodes. When a violation is detected, a review conflict record is generated and visualized and a report is generated.

2. The intelligent review method for architectural construction drawings as described in claim 1, characterized in that, In step S2, the process of generating 3D architectural components with semantic attributes by fusing 2D geometric primitives and text annotation information through a component recognition rule base specifically includes: S211, Component Type Recognition: Based on the layer name in the layer attribute information, combined with the primitive type and spatial distribution in the geometric vector information, the candidate building components and their types are initially identified by matching through a predefined component recognition rule base. The types include walls, beams, columns, slabs, doors, and windows. S212, 3D geometry generation: For the identified candidate building components, the corresponding 3D generation algorithm is called according to their type: For wall and column components, the 2D closed contour is extruded, and the extruded height is determined according to the height value in the bound text annotation information or the default specification value; For beam and slab components, the 3D solid is generated by lofting or filling operations based on their 2D contour lines and the marked elevation information. S213, Topology Relationship Reconstruction: Based on the spatial positions of all generated 3D building components, automatically calculate and record the spatial topology relationships between components, including connection relationships, support relationships, and spatial containment relationships.

3. The intelligent review method for architectural construction drawings as described in claim 2, characterized in that, In step S2, the aggregation into a lightweight 3D building information model specifically includes: S221, Semantic Information Injection: The dimensions, materials, and fire rating attributes related to the components in the text annotation information are associated and bound one-to-one or one-to-many with the three-dimensional building components generated in step S212 to complete the injection of semantic information. S222, Lightweight Model Encapsulation: An incremental lightweight algorithm is adopted to remove intermediate data and historical operation records in the 3D modeling process, while retaining the final geometric data, semantic attribute data and topological relationship data between building components, and encapsulating them into a lightweight 3D building information model. The format of the lightweight 3D building information model is independent of any specific computer-aided design software environment.

4. The intelligent review method for architectural construction drawings as described in claim 1, characterized in that, Step S4 specifically includes: S410, Review Rule Trigger: Traverse each building component entity and its semantic attributes in the lightweight 3D building information model, and automatically match and activate the associated computational logic rule nodes from the specification knowledge graph according to the component type; S420, Parametric Calculation and Logical Judgment: Based on the activated calculation logic rule nodes, the required geometric and semantic parameters are extracted from the lightweight 3D building information model in real time, and parametric mathematical calculations and spatial relationship logical judgments are performed. S430. Review Result Generation and Visual Annotation: The results of parametric calculation and logical judgment are compared with the thresholds or conditions in the rule nodes. If a rule is violated, a review conflict record is generated, which includes the identifier of the violating component, the specific code clause violated, the calculated value of the violating parameter, and the standard requirement value. All review conflict records are visualized and highlighted in three-dimensional space in the lightweight three-dimensional building information model, and a structured review report is generated.

5. The intelligent review method for architectural construction drawings as described in claim 1, characterized in that, In step S4, before dynamically matching and reasoning the lightweight 3D building information model with the standard knowledge graph, the following steps are also included: dynamically filtering the target component set or target area from the lightweight 3D building information model based on the user-input review scope selection instruction, and automatically expanding the target review domain based on the component topology relationship; dynamic matching and reasoning are only performed on the target component set or target area.

6. The intelligent review method for architectural construction drawings as described in claim 1, characterized in that, Step S4, after generating the review conflict record, also includes a review conflict intelligent grading and classification step, specifically: Based on a predefined conflict severity level rule base, each review conflict record is automatically classified, with conflict severity levels including at least a major risk level, a general violation level, and a recommended optimization level; At the same time, the violations of the normative provisions in the review conflict record are automatically classified according to their professional fields, which include at least fire safety, structural safety, energy conservation and environmental protection, and ergonomics. In the generated structured review report and 3D visualization annotations, the review conflict records are sorted, filtered, and displayed in a differentiated manner according to the severity level of the conflict and the professional field classification.

7. The intelligent review method for architectural construction drawings as described in claim 1, characterized in that, It also includes a model optimization feedback loop independent of the core review process, which is executed periodically or triggered periodically, specifically including: Construction of historical conflict sample library: After the user confirms or corrects the review conflict record generated in step S4, the finally confirmed conflict case, including its corresponding model component data, triggered calculation logic rule node and final judgment result, will be stored as annotated sample in the historical conflict sample library. Rule confidence adaptive adjustment: Based on the historical conflict sample library, the trigger frequency and false alarm rate of each calculation logic rule node are statistically analyzed. When the false alarm rate of a rule is consistently higher than the threshold of 10%, the system automatically reduces the confidence weight of the rule when it is triggered, or gives the user an uncertainty prompt when it is triggered. New rule pattern mining: When the case in the historical conflict sample library is found to have frequent coexistence of two or more types of component violations in the sample, and their combination pattern is not covered by the existing normative knowledge graph, the potential new rule pattern is submitted to the administrator for review as an optimization suggestion, so as to expand or correct the normative knowledge graph.

8. The intelligent review method for architectural construction drawings as described in claim 7, characterized in that, It also includes an incremental review mode, which is triggered when an update to the specification knowledge graph or a partial revision to the lightweight 3D building information model is detected. Specifically, it includes: Change impact domain analysis: When the specification knowledge graph is updated, the newly added or modified calculation logic rule nodes are analyzed to determine the set of building component types that are affected by them; when the 3D model is partially revised, the set of building components that have been added, deleted, or modified is located. Dynamic construction of the target review domain: The set of affected component types obtained from the above analysis is combined with the set of components that have been revised to generate a target review domain; Targeted rule matching and review: For components within the target review domain, perform dynamic rule matching and reasoning, parameterized calculation and logical judgment, and generate review conflict records.

9. The intelligent review method for architectural construction drawings as described in claim 1, characterized in that, In step S4, while generating each review conflict record, a corresponding review logic traceability description is automatically generated. The process of generating the review logic traceability description includes: Automatically record the original specification clause content of the computational logic rule node that triggered the conflict; The key model parameters and their values ​​used in this determination are automatically extracted. The key model parameters include at least the geometric parameters and semantic parameters extracted in real time from the lightweight 3D building information model. Based on rule nodes and key model parameters, a natural language description text is automatically generated, which includes the identifier of the non-compliant component, the violated normative clause, the specific parameter values ​​used, the logical judgment process, and the final conclusion. The review logic traceability description and the corresponding review conflict record are associated and stored, and then output to the structured review report and the 3D visualization interface.

10. The system of the intelligent review method for architectural construction drawings as described in claims 1 to 9, characterized in that, include: The data parsing module is configured to receive original electronic construction drawings generated by different computer-aided design software through a data interface, and parse and convert them into standardized structured data objects. The 3D model reconstruction module is configured to build a lightweight 3D building information model that integrates geometric and semantic information based on standardized structured data objects and component recognition and 3D generation algorithms. The standardized knowledge graph module is configured to store and manage structured computational logic rule nodes transformed through natural language processing technology, and to form a standardized knowledge graph. The intelligent review engine module is configured to dynamically match and reason with the lightweight 3D building information model and the standard knowledge graph, perform parametric calculations and logical judgments, and generate review conflict records. The review scope selection module is configured to receive user instructions and dynamically filter target component sets or target areas from the lightweight 3D building information model for targeted review by the intelligent review engine module. The conflict handling module is configured to automatically classify the severity level and professional field of the review conflict records generated by the intelligent review engine module. The model optimization feedback module is configured to build a historical conflict sample library and adaptively adjust the rule confidence and mine potential new rule patterns based on it in order to optimize the canonical knowledge graph module. The incremental review control module is configured to automatically analyze the impact domain of the change and construct the target review domain when a model revision or specification update is detected, thereby triggering the incremental review mode. The logic traceability generation module is configured to automatically generate a natural language description for each review conflict record, including the original specification clauses, key parameters, and the judgment process.