Timber building knowledge graph construction and construction process reasoning method and system

Abstract

The application discloses a kind of timber construction knowledge graph construction and construction process reasoning method and system, the method includes the following steps: S1: constructing the timber construction field ontology architecture for the whole process of construction;S2: timber construction multi-source heterogeneous data acquisition and directional pretreatment;S3: multi-modal knowledge extraction and knowledge graph construction of priori knowledge fusion field;S4: three-layer progressive fusion construction process reasoning;S5: adaptive optimal construction / rehabilitation process scheme landing verification and knowledge graph self-iterative optimization.The application relates to the technical field of timber construction intelligent construction and cultural heritage digitization protection, and can solve the problems that timber knowledge graph and construction practice are disconnected in the prior art, timber multi-source heterogeneous knowledge extraction precision is low, implicit experience cannot be reused, general AI model has poor adaptability in timber field, timber construction process planning relies on artificial experience, non-standard adaptability is poor, compliance cannot be guaranteed, and knowledge construction and practice landing are disconnected.

Description

Technical Field

[0001] This invention relates to the fields of intelligent construction of wooden buildings and digital protection of cultural heritage, and in particular to a method and system for constructing a knowledge graph of wooden buildings and reasoning about construction techniques. Background Technology

[0002] Timber-structured architecture is a core cultural carrier of Chinese civilization, and its construction techniques have been listed as a national intangible cultural heritage. Simultaneously, under the "dual carbon" goal (referring to carbon emissions), modern timber-structured architecture, as a core form of green and low-carbon building, has ushered in opportunities for large-scale development. With the development of digital technology, BIM technology, artificial intelligence, and knowledge graphs have been gradually applied to the field of timber-structured architecture. However, existing technologies still have the following significant core shortcomings, severely restricting the digital transformation and industrial development of timber-structured architecture:

[0003] (1) Existing solutions are mostly geared towards document retrieval, digital archiving of cultural relics, and semantic query. They only realize the knowledge structure storage at the text semantic level. The ontology architecture does not embed the mandatory compliance constraint rules of wooden construction, and cannot express the inherent construction logic of "component-module-process-material-compliance-quality". The constructed map can only realize static information query and does not have the dynamic reasoning ability for construction implementation. It is completely disconnected from production and construction practice.

[0004] (2) Knowledge of timber construction is scattered in official books, industry standards, oral experience of craftsmen, and historical process documents, covering multiple modal forms such as text, drawings, audio and video, and three-dimensional models. It also has industry characteristics such as vertical layout, obscure professional terms, deep binding of text and graphics, and a high proportion of implicit experience. General knowledge graph construction methods cannot adapt to the special needs of the timber construction field, the knowledge extraction accuracy is low, and it is impossible to transform the implicit experience passed down by craftsmen into structured knowledge that can be reasoned and reused.

[0005] (3) Existing computer vision and natural language processing models can only process general documents and ordinary drawings. For complex texts in wooden structure classics that are vertically arranged and have no modern punctuation, contain a large number of variant characters and phonetic loan characters, and have deep interweaving of text and graphics (such as hand-drawn illustrations and text sharing the same page), as well as non-standard annotations and mixed text and graphics scenarios in wooden structure drawings, the extraction accuracy is difficult to meet the requirements of engineering applications. General machine learning models do not combine knowledge of the wooden structure field and can only process standardized components. They fail when encountering non-standard forms and special repair scenarios, and have extremely poor generalization ability.

[0006] (4) The entire process of traditional timber construction planning relies heavily on the personal experience of craftsmen. The planning of non-standard components and special-shaped buildings requires multiple trials and errors, resulting in low efficiency, poor consistency, and high quality risks. Existing intelligent reasoning solutions either use a fixed process library for matching, which can only handle standardized components, or use a purely data-driven machine learning model, which lacks the mandatory constraints of knowledge in the field of timber construction. The reasoning results do not meet the construction specifications and structural safety requirements. Especially in the scenario of cultural relic restoration, it cannot meet the legal requirement of "minimal intervention and no change to the original state of cultural relics" and cannot take into account the adaptability and compliance of non-standard projects.

[0007] (5) In the existing technology, the construction of knowledge graphs and the reasoning and implementation of the construction process are completely separated. The quality data, defect data and process optimization experience after the construction is completed cannot be fed back to the knowledge graph, the reasoning rules cannot be self-iterated, the reasoning accuracy and adaptability cannot be continuously improved, and it is always impossible to get rid of human intervention and it is difficult to achieve full-process intelligence.

[0008] Therefore, there is a need to provide a method and system for constructing a knowledge graph of timber structures and reasoning about construction techniques, which can solve the problems in existing technologies such as the disconnect between timber structure knowledge graphs and construction practices, low accuracy of multi-source heterogeneous knowledge extraction from timber structures and the inability to reuse implicit experience, poor adaptability of general AI models in the field of timber structures, reliance on human experience for timber structure construction process planning, poor adaptability to non-standard structures and inability to guarantee compliance, and the separation between knowledge construction and practical implementation. Summary of the Invention

[0009] The purpose of this invention is to provide a method and system for constructing a knowledge graph of timber structures and reasoning about construction processes. This method and system can solve the problems in the prior art, such as the disconnect between timber structure knowledge graphs and construction practices, low accuracy of multi-source heterogeneous knowledge extraction from timber structures and the inability to reuse implicit experience, poor adaptability of general AI models in the field of timber structures, reliance on human experience for timber structure construction process planning, poor adaptability to non-standard structures and inability to guarantee compliance, and the separation between knowledge construction and practical implementation.

[0010] This invention is implemented as follows:

[0011] A method for constructing a knowledge graph of wooden architecture and reasoning about its construction techniques includes the following steps:

[0012] Step S1: Construct an ontology architecture for the entire construction process of timber-framed buildings;

[0013] Step S2: Acquisition and targeted preprocessing of multi-source heterogeneous data for timber structure construction;

[0014] Step S3: Multimodal knowledge extraction and knowledge graph construction integrating domain prior knowledge;

[0015] Step S4: Reasoning for the three-layer progressive integration construction process;

[0016] Step S5: Verify the implementation of the optimal construction / repair process and iteratively optimize the knowledge graph.

[0017] Step S1 includes the following sub-steps:

[0018] Step S11: Construct a three-dimensional hierarchical domain ontology architecture that integrates value perception, entity construction, and process characteristics;

[0019] Step S12: Predefine the creation of guided semantic relationships between entities;

[0020] Step S13: Embed the underlying mandatory constraint rules of the ontology.

[0021] In step S11, the three-dimensional integrated hierarchical domain ontology architecture includes:

[0022] Dimension 1, Value Perception Dimension: Predefined core entity layer and standardized attribute layer; Among them, the core entity layer predefined three categories of core entities: protection level entity, architectural era entity, and architectural regulation entity; the standardized attribute layer is used to standardize the description of the cultural attributes, protection level attributes, historical evolution attributes, form level attributes, and repair intervention restriction attributes of wooden buildings, providing semantic support for the principle of minimum intervention in cultural relic repair scenarios.

[0023] Dimension Two: Entity Construction Dimension: Predefined hierarchical entity system and spatial association rule layer; the hierarchical entity system is based on a hierarchical line classification method of "building part → core component → component", predefined with all timber structural component entities under four major building parts: platform part entity, body part entity, roof part entity, and decorative part entity, establishing a hierarchical entity system from the overall building to the smallest functional unit; the spatial association rule layer predefined spatial subordination rules, assembly and overlap relationship rules, and modular association constraint rules;

[0024] Dimension Three: Technological Characteristics Dimension: Predefined full-scale standardized attribute set and standardized specification layer; the full-scale standardized attribute set includes six core attributes: material attributes, structural attributes, processing technology attributes, construction procedure attributes, defect characteristic attributes, and spatial positioning attributes. Each attribute has a unified enumeration value specification, which fully describes the physical characteristics, technological requirements, and status information of the timber components; the standardized specification layer predefined attribute value range enumeration specification, process parameter threshold specification, quality acceptance standard specification, and defect level coding specification.

[0025] In step S12, based on the inherent logic of timber construction, construction-oriented semantic relationships between entities are predefined. These construction-oriented semantic relationships include: composition relationship, modular relationship, process adaptation relationship, process sequence relationship, causal relationship, constraint relationship, compliance correspondence relationship, and spatial subordination relationship, providing a semantic basis for the relationship construction and process reasoning of the knowledge graph.

[0026] In step S13, the underlying mandatory constraint rules of the ontology include:

[0027] ① Terminology unification and disambiguation rules: Establish a standardized mapping library of wooden structure terminology, clarify the mapping of synonyms for Song and Qing dynasty terminology, the mapping of aliases for official and local school terminology, and the disambiguation rules for synonyms / different objects with the same name.

[0028] ② Multi-source knowledge priority rules: Establish a priority system of "National mandatory standards > Recommended national standards > Industry standards > Official construction records > Local practices and regulations > Historical verification of process data > Experience of intangible cultural heritage inheritors > Documentary materials"; For cultural relic protection projects, an additional prerequisite rule is added: "The principle of preserving the original state of cultural relics" has the highest priority;

[0029] ③ Create mandatory compliance rules: predefined modular mandatory constraint rules, structural safety mandatory constraint rules, process sequence constraint rules, and minimum intervention rules for cultural relic restoration.

[0030] Step S2 includes the following sub-steps:

[0031] Step S21: Collect multi-source raw data for the entire timber construction chain;

[0032] Step S22: Perform targeted preprocessing on the multi-source raw data of the entire wooden structure construction chain.

[0033] In step S21, the scope of data collection for multi-source raw data across the entire timber construction chain includes:

[0034] ① Standards and specifications: National / industry standards related to timber structure design, construction, and cultural relic protection, including the *Yingzao Fashi* (Building Standards), the *Qing Gongbu Gongzao Zuofa Zeli* (Regulations and Examples of Engineering Practices of the Qing Dynasty), and local regulations and examples;

[0035] ② Craftsmanship knowledge: historical construction process documents, component processing SOPs, audio and video recordings of intangible cultural heritage inheritors' oral accounts, and craftsman's operation manuals;

[0036] ③ Production measurement data: wood property test data, equipment operation data, component BIM model / design drawings, quality defects and rework records, acceptance reports;

[0037] ④ Project Case Studies: Complete documentation of implemented ancient building restoration and antique-style wooden structure projects;

[0038] In step S22, the targeted preprocessing process for documents / drawings is as follows: Considering the characteristics of vertical layout, deep text-image binding, dense annotation of obscure terminology, and blurred lines in hand-drawn drawings, PDF scans and drawing scans undergo single-page splitting, noise reduction, grayscale conversion, and tilt correction preprocessing. Then, a layout analysis model finely tuned using a dataset specific to the timber construction field is employed to divide the drawings / documents into three core areas: text area, drawing area, and table area. Each partition is bound to a unique traceability identifier in the format "data source number - chapter number - page number - partition number" for knowledge tracing, conflict location, and version management, recording data source, chapter, and page number information.

[0039] The targeted preprocessing process for audio and video is as follows: ASR speech transcription and endpoint detection are performed on the audio and video of interviews with intangible cultural heritage inheritors and artisans' operations. The speech recognition model is optimized for professional terminology, implicit knowledge is extracted, structured text is generated, and source, time period, and speaker information are bound.

[0040] The directional preprocessing flow for BIM model types is as follows: IFC data parsing, component parameter extraction, and assembly relationship identification.

[0041] The targeted preprocessing process for structured production data is as follows: deduplication, missing value filling, outlier removal, and normalization are performed on material property testing, equipment operation, and quality acceptance data, and the data is then archived according to the entity type of the ontology architecture.

[0042] Step S3 includes the following sub-steps:

[0043] Step S31: Multimodal knowledge extraction;

[0044] Step S32: Cross-validation and conflict resolution;

[0045] Step S33: Standardized entity alignment and knowledge graph construction.

[0046] In step S31, the multimodal knowledge extraction includes:

[0047] ① Text knowledge extraction: A pre-constructed dictionary for the timber construction field based on the ontology terminology system is built. This dictionary includes exclusive terms for timber construction, mappings of Song and Qing dynasty terminology, and alternative names of local schools. An entity and relation extraction model that integrates prior knowledge of the domain is constructed. The entity and relation extraction model is optimized into a token-level attention mechanism guided by domain terminology. Through attention weights, it automatically focuses on timber construction terms and finally outputs entities, attributes, and relations that conform to the ontology definition, generating standardized knowledge triples <subject, predicate, object>.

[0048] ② Drawing / BIM Model Knowledge Extraction: The model is segmented using instances to separate text and graphics. A contour extraction algorithm is used to vectorize the drawing area, generating vector drawings and extracting contour features. An object detection model is used to identify dimensions, mortise and tenon joints, and component numbers within the drawings. OCR is used to extract annotation parameter values ​​and optimize technical terminology. A parametric model of the components is generated, and the BIM model IFC data is parsed to extract construction parameters, components, component assembly, and attribute mapping, generating standardized drawing / BIM knowledge triples.

[0049] ③ Audio and video technical knowledge extraction: The speech recognition is optimized for wood structure professional terminology in the audio and video. Implicit knowledge is extracted from the spoken text after speech recognition correction. Implicit knowledge is structured and mapped to the ontology. The credibility of craftsmen's experience knowledge is graded and standardized technical knowledge triples are generated.

[0050] In step S32, the same attribute parameters of the same entity in the triples from the three sources in step S31 are compared. If the deviation of the size parameter exceeds the preset first threshold and the deviation of the process parameter exceeds the preset second threshold, it is marked as a parameter conflict. Automatic resolution is performed in strict accordance with the multi-source knowledge priority rules of step S1. If automatic resolution is not possible, expert manual review and calibration are triggered. After the review is passed, it is included in the primary knowledge triple.

[0051] In step S33, a dual entity alignment mechanism of "ontology standard encoding benchmark alignment + semantic similarity supplementary alignment" is adopted: First, the entity definition in the three-dimensional integrated hierarchical domain ontology architecture is used as the benchmark to align different name expressions of the same entity; then, through the semantic similarity model, local gender names and colloquial names are supplemented and aligned, and a unique global ID is assigned to each entity; the standardized primary knowledge triples are stored in the graph database to construct a production-level wooden building knowledge graph, and the visualization engine of the production-level wooden building knowledge graph provides visualization query and editing functions.

[0052] Step S4 includes the following sub-steps:

[0053] Step S41: Perform structured analysis and standardized vector transformation on all dimensions of the requirements of the project to be processed;

[0054] Step S41 includes the following sub-steps:

[0055] Step S411: Obtain the full-dimensional requirement information of the project to be processed, and complete the standardized parsing based on the three-dimensional integrated hierarchical domain ontology architecture of step S1 to generate a standardized inference input vector set; the standardized inference input vector set includes: project basic feature vector, component design feature vector, resource constraint feature vector, compliance and status feature vector;

[0056] Among them, the basic feature vector of the project is generated by extracting the building form, number of bays, number of depths, roof type, modular system, protection level, and building era.

[0057] Component design feature vector: Extract the component list, mortise and tenon joint details, geometric dimensions, machining accuracy requirements, and assembly constraints from the BIM model / design drawings to generate the component feature vector X2;

[0058] Resource constraint feature vector: Extract the tree species, material properties, available processing equipment, and construction period requirements of the timber used to generate a resource feature vector X3;

[0059] Compliance and Status Feature Vector: Extract the standards and specifications that the project must follow, the requirements for cultural relic restoration, the current status of component defects, and the quality acceptance standards to generate a constraint feature vector X4;

[0060] The above standardized inference input vectors are concatenated to generate a standardized inference input vector set X=[X1,X2,X3,X4];

[0061] Step S42: Coarse screening of candidate process solutions based on semantic similarity;

[0062] Load the pre-trained domain semantic matching model, calculate the semantic similarity between the standardized inference input vector set X of the project to be processed and the component-level process templates and substructure process modules in the knowledge graph, filter out the top N candidate process schemes with the highest similarity, and generate a set of candidate process schemes.

[0063] Step S43: Compliance screening based on the underlying mandatory constraint rules of ontology;

[0064] Based on the underlying mandatory constraint rules embedded in the ontology in step S1, an expert rule base is constructed to perform progressive compliance verification on the candidate process scheme set: ① Form compliance verification, filtering out schemes that do not meet the requirements of the project era and regulations; ② Structural safety verification, filtering out schemes that do not meet the mandatory constraints of module, component size, and tenon and mortise joint; ③ Process logic verification, filtering out schemes that do not meet the process sequence constraints; ④ Constraint condition verification, filtering out schemes that do not meet the requirements of equipment, materials, construction period, and cultural relic protection; After verification, a pre-selected process scheme set is obtained.

[0065] Step S44: Multi-objective optimization reasoning that incorporates domain constraints;

[0066] With multiple optimization objectives of "maximizing compliance, processing efficiency, construction cost, quality defect rate, and minimizing repair intervention," a graph attention network reasoning model integrating mandatory constraints in the timber structure field is constructed. The graph attention network reasoning model takes the subgraph structure of the knowledge graph as input, transforms the underlying compliance rules embedded in the ontology into reasoning constraints, and automatically learns the influence weights of different entities and constraints on the process scheme through the attention mechanism. It iteratively optimizes the process sequence, process parameters, and acceptance standards of the pre-selected process scheme, and finally outputs the optimal construction / repair process scheme, including component processing process cards, CNC machining parameter sets, assembly process flow, on-site construction plan, and quality acceptance standards.

[0067] Step S5 includes the following sub-steps:

[0068] Step S51: Simulation verification of the optimal construction / repair process scheme;

[0069] Input the processing parameters from the optimal construction / repair process scheme into the CNC machining simulation software to conduct tool path collision simulation and wood cutting stress simulation. If the simulation fails, feed the defect information back to the graph attention network inference model in step S44 and re-execute the process inference until the simulation passes. Input the construction procedure scheme into the virtual construction simulation platform to conduct construction procedure pre-play and collision detection. If the simulation passes, output the final executable process scheme.

[0070] Step S52: Implementation and Data Collection of the Final Executable Process Scheme: The verified final executable process scheme is distributed to the production workshop and construction project department to drive CNC equipment to complete component processing and guide on-site assembly / repair operations; collect component size inspection data, processing quality and defect data, construction progress and process data, completion acceptance data, material and equipment operation data, defect and rework data, and IoT sensing data after processing, and bind them to the unique global ID of the corresponding component entity;

[0071] Step S53: Knowledge graph self-iterative optimization;

[0072] Based on the collected data, the implementation effect of the final executable process solution is evaluated from two dimensions. If the compliance assessment is passed and the performance indicators meet the standards, the final executable process solution is updated to the knowledge graph as a new knowledge instance, and a corresponding triple is added. If the compliance assessment fails, or the compliance assessment is passed but the performance indicators do not meet the standards, i.e. there is a quality defect or room for optimization, the underlying mandatory constraint rules, process adaptation relationships and weights of the graph attention network inference model in the ontology are updated after analyzing the causes, and the self-iterative optimization of the knowledge graph and graph attention network inference model is completed.

[0073] A system for implementing the aforementioned method for constructing a knowledge graph of wooden structures and reasoning about construction techniques includes:

[0074] Domain ontology construction and management module: used to execute step S1, construct a three-dimensional integrated layered domain ontology architecture for the entire creation process, and predefine the core entity layer, standardized attribute layer, hierarchical entity system, spatial association rule layer, full standardized attribute set, standardized specification layer and underlying mandatory constraint rules;

[0075] Multi-source data preprocessing module: used to execute step S2, complete the collection of multi-source raw data of the entire chain of timber construction, targeted preprocessing in timber construction scenarios and binding of full-chain traceability identifiers, and generate standardized preprocessed datasets;

[0076] Multimodal knowledge extraction and graph construction module: used to execute step S3, based on a three-dimensional integrated hierarchical domain ontology architecture to complete multimodal knowledge extraction, cross-validation, conflict resolution and entity alignment that integrates domain prior knowledge, and construct a production-grade wooden building knowledge graph;

[0077] Project Requirements Analysis and Process Reasoning Module: Used to execute step S4, complete the standardized analysis of all dimensions of the requirements of the project to be processed, and automatically generate the best construction / repair process solution through a three-layer progressive reasoning framework.

[0078] Simulation verification and iterative optimization module: used to execute step S5, complete the simulation verification of the final executable process scheme, collect data for implementation, and perform self-iterative optimization of the knowledge graph of construction-level wooden structures.

[0079] Compared with the prior art, the present invention has the following advantages:

[0080] 1. This invention constructs a three-dimensional hierarchical domain ontology architecture that integrates "value cognition, entity construction, and process characteristics," achieving deep integration of knowledge and practice. By creating mandatory compliance rules and terminology disambiguation mechanisms through pre-embedded modules, safety, and processes at the bottom layer, it solves the pain point of the disconnect between knowledge graphs and practical construction, enabling the knowledge base to have inherent logical constraints and reasoning capabilities, and can directly guide production implementation.

[0081] 2. This invention employs a multimodal domain knowledge extraction method to solve the problem of acquiring ancient books and implicit experience. Targeting vertically formatted classics, non-standard drawings, and craftsmen's oral experience, it integrates a domain prior dictionary and attention optimization mechanism to significantly improve the extraction accuracy of obscure terms and mixed text and image data. It realizes the structured transformation and reuse of craftsmen's "oral and mental" implicit experience, greatly improves the domain knowledge coverage, and solves the problems of low extraction accuracy of multi-source heterogeneous knowledge in wooden structures and the inability to reuse implicit experience.

[0082] 3. This invention establishes a three-layer progressive reasoning framework, which eliminates the reliance on human experience. Through progressive reasoning of "semantic coarse screening - rule fine screening - multi-constraint fusion optimization", it automatically generates the optimal process solution that balances efficiency and cost under the premise of strictly ensuring structural safety and repair compliance. It effectively solves the problem of personalized adaptation of non-standard components to cultural relic repair scenarios and the problem of non-compliance. The reasoning results are more engineering feasible.

[0083] 4. This invention constructs a closed-loop data system for the entire process, enabling self-iterative optimization of knowledge. By connecting the links of "graph construction - process reasoning - implementation - feedback optimization", construction quality data is fed back to the knowledge graph of timber construction, realizing the continuous iterative evolution of the underlying mandatory constraint rules and graph attention network reasoning model. The system's intelligence level is continuously improved with the accumulation of applications, forming a virtuous cycle of self-optimization.

[0084] 5. The invention has significant application value. It can directly drive CNC machining and on-site assembly, solving the problem of the disconnect between knowledge construction and practical implementation. It can shorten the construction period, reduce reliance on scarce craftsmen, and promote the industrialization and upgrading of wooden buildings. At the same time, it provides scientific and compliant decision support for the restoration of cultural relics, which is of great significance for the protection of cultural heritage and the inheritance of intangible cultural heritage skills. Attached Figure Description

[0085] Figure 1 This is a flowchart of the method for constructing a knowledge graph of wooden buildings and reasoning about construction techniques according to the present invention;

[0086] Figure 2 This is a three-dimensional, hierarchical domain ontology architecture diagram in the method for constructing a knowledge graph of wooden buildings and reasoning about construction techniques in this invention;

[0087] Figure 3 This is a flowchart of steps S2 and S3 in the method for constructing a knowledge graph of wooden buildings and reasoning about construction techniques of the present invention;

[0088] Figure 4 This is a flowchart of step S4 in the method for constructing a knowledge graph of wooden buildings and reasoning about construction techniques in this invention;

[0089] Figure 5 This is a flowchart of step S5 in the method for constructing a knowledge graph of wooden buildings and reasoning about construction techniques in this invention;

[0090] Figure 6 This is a module diagram of the knowledge graph construction and construction process reasoning system for wooden buildings of the present invention.

[0091] Figure 7 This invention is a closed-loop architecture diagram of knowledge graph and process reasoning for wooden structure construction based on three-dimensional ontology. Detailed Implementation

[0092] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0093] Please see the appendix Figure 1 and attached Figure 7 A method for constructing a knowledge graph of wooden architecture and reasoning about its construction techniques, comprising the following steps:

[0094] Step S1: Construct an ontology architecture for the entire construction process of timber-framed buildings.

[0095] In step S1, taking the entire life cycle construction process of wooden buildings as the core, a three-dimensional hierarchical domain ontology architecture integrating value cognition, physical construction, and craft features is constructed. The core entities, entity attributes, semantic relationships, and underlying mandatory constraint rules of the value cognition dimension are predefined to provide a standardized semantic framework for knowledge graph construction and craft reasoning.

[0096] The inputs to step S1 are the specifications for timber construction, official classics, and industry standards; the outputs of step S1 are the standardized ontology architecture and constraint rule library for the field of timber architecture.

[0097] Step S1 includes the following sub-steps:

[0098] Please see the appendix Figure 2 Step S11: Construct a three-dimensional hierarchical domain ontology architecture that integrates value perception, entity construction, and process characteristics.

[0099] This three-dimensional, hierarchical domain ontology architecture includes:

[0100] Dimension 1: Value Perception Dimension: Predefined Core Entity Layer and Standardized Attribute Layer. The core entity layer predefines three main categories of core entities: protection level entities (e.g., national, provincial, municipal, none), architectural era entities (e.g., Song, Qing, modern imitations), and architectural style entities (e.g., grand style, small style, palace, hall). The standardized attribute layer is used to standardize the descriptions of the cultural attributes, protection level attributes, historical evolution attributes, form level attributes, and restoration intervention restriction attributes of wooden structures, providing semantic support for the principle of minimum intervention in cultural relic restoration scenarios.

[0101] Dimension Two: Physical Structure Dimension. This dimension predefines a hierarchical physical system and a spatial association rule layer. The hierarchical physical system is based on a hierarchical classification method of "building parts → core components → components," predefining all timber structural components under four major building parts: platform, body, roof, and decoration. This establishes a hierarchical physical system from the overall building to the smallest functional unit, accurately expressing the physical composition and spatial logic of the timber structure. The spatial association rule layer predefines spatial subordination rules, assembly and overlap rules, and modular association constraint rules.

[0102] Dimension Three: Technological Characteristics Dimension: This dimension predefines a complete set of standardized attributes and a standardized specification layer. The complete set of standardized attributes includes six core categories: material attributes (e.g., tree species, moisture content, material properties), structural attributes (e.g., tenon and mortise type, dimensional parameters), processing technology attributes (e.g., procedures, tools, cutting parameters), construction procedure attributes (e.g., installation process, timing requirements), defect characteristic attributes (e.g., decay, cracking, deformation type), and spatial positioning attributes (e.g., coordinates, assembly position). Each attribute has a unified enumeration value specification, comprehensively describing the physical characteristics, technological requirements, and status information of the timber components. The standardized specification layer predefines attribute value range enumeration specifications, technological parameter threshold specifications, quality acceptance standard specifications, and defect level coding specifications.

[0103] Please see the appendix Figure 2 Step S12: Predefine the creation of directional semantic relationships between entities.

[0104] Based on the inherent logic of timber construction, a construction-oriented semantic relationship is predefined between entities. This construction-oriented semantic relationship includes: composition relationship, modular relationship, process adaptation relationship, process sequence relationship, causal relationship, constraint relationship, compliance correspondence relationship, and spatial subordination relationship, providing a semantic foundation for the relationship construction and process reasoning of the knowledge graph.

[0105] Step S13: Embed the underlying mandatory constraint rules of the ontology.

[0106] Establishing a robust constraint system for timber construction from the ground up provides compliance assurance for knowledge integration and process reasoning.

[0107] In step S13, the underlying mandatory constraint rules of the ontology include:

[0108] ① Terminology unification and disambiguation rules: Establish a standardized mapping library for timber structure terminology, clarify the mapping of synonyms for Song and Qing dynasty terminology, the mapping of alternative names for official and local school terminology, and disambiguation rules for synonyms / different objects with the same name. To address the differences in terminology across different eras and regions, a disambiguation mechanism of "unifying abstract functional categories and differentiating specific forms through attributes" is adopted to eliminate semantic ambiguity.

[0109] ② Multi-source knowledge priority rules: In conventional scenarios, a priority system is established as follows: "National mandatory standards > Recommended national standards > Industry standards > Official construction records > Local practices and regulations > Historical verification of process data > Experience of intangible cultural heritage inheritors > Documentary materials." For cultural relic protection projects, an additional prerequisite rule is added: the "principle of preserving the original state of cultural relics" has the highest priority. Under this principle, the priority of historical process data, local practices and regulations, and experience of intangible cultural heritage inheritors can be elevated to the same or higher level as the current standards, and the final decision is made through an expert review mechanism.

[0110] ③ Create mandatory compliance rules: predefine mandatory modular constraint rules, mandatory structural safety constraint rules, process sequence constraint rules, and minimum intervention rules for cultural relic restoration to ensure the compliance of knowledge application and process reasoning from the bottom up.

[0111] To address the disconnect between existing timber structure knowledge graphs and construction practices, this invention constructs a three-dimensional, hierarchical domain ontology architecture that integrates the entire construction process. It embeds construction compliance constraints from the bottom layer to create a construction-level knowledge graph with reasoning and decision-making capabilities, achieving a core leap from "static archiving" to "dynamic application".

[0112] Step S2: Acquisition and targeted preprocessing of multi-source heterogeneous data for timber construction.

[0113] In step S2, multi-source data from the entire timber construction chain are collected, and targeted preprocessing is performed on the specific characteristics of the data in the timber construction field to provide a standardized data foundation for knowledge extraction.

[0114] The input to step S2 is multi-source raw data of the entire wooden structure construction chain, and the output of step S2 is a standardized preprocessed dataset and a full-chain traceability identification system.

[0115] Step S2 includes the following sub-steps:

[0116] Step S21: Collect multi-source raw data for the entire wooden structure construction chain.

[0117] The scope of multi-source raw data collection for the entire timber construction chain includes:

[0118] ① Standards and specifications: National / industry standards related to timber structure design, construction, and cultural relic protection; official classics such as the "Yingzao Fashi" and "Qing Gongbu Gongcheng Zuofa Zeli"; and local regulations and practices.

[0119] ② Craftsmanship knowledge: historical construction process documents, component processing SOPs, audio and video recordings of intangible cultural heritage inheritors' oral accounts, and craftsman's operation manuals.

[0120] ③ Production measurement data: wood property test data, equipment operation data, component BIM model / design drawings, quality defects and rework records, and acceptance reports.

[0121] ④ Project Case Studies: Complete documentation of implemented ancient building restoration and antique-style wooden structure projects.

[0122] Step S22: Perform targeted preprocessing on the multi-source raw data of the entire wooden structure construction chain.

[0123] The targeted preprocessing workflow for documents / drawings is as follows: Considering the characteristics of vertical layout, deep text-image binding, dense annotation of obscure terminology, and blurred lines in hand-drawn drawings, PDF scans and drawing scans undergo single-page splitting, noise reduction, grayscale conversion, and slant correction preprocessing. Then, a layout analysis model finely tuned using a dataset specific to the timber construction field is employed to divide the drawings / documents into three core areas: text area, drawing area, and table area. Each area is bound to a unique traceability identifier in the format "Data Source Number - Chapter Number - Page Number - Partition Number" for knowledge tracing, conflict localization, and version management, recording data source, chapter, and page number information.

[0124] The targeted preprocessing process for audio and video is as follows: ASR speech transcription and endpoint detection are performed on the audio and video of interviews with intangible cultural heritage inheritors and artisans' operations. The speech recognition model is optimized for construction professional terminology, and implicit knowledge such as construction procedures, parameter thresholds, and defect avoidance is extracted to generate structured text and bind source, time period, and speaker traceability information.

[0125] The orientation preprocessing workflow for BIM model classes is as follows:

[0126] IFC data parsing: First, IFC file version verification and format validity checks are performed to filter out entity types related to non-timber buildings (such as mechanical and electrical, pipeline entities); a standardized mapping table is established between IFC entity attributes and attributes of the three-dimensional integrated layered domain ontology architecture, automatically mapping the native attributes of IFC to the full set of standardized attribute fields such as "geometric dimensions" and "material attributes" defined by the ontology.

[0127] Component parameter extraction: Based on the hierarchical entity system of the ontology, it automatically identifies all wooden components of the four major building parts: the platform part entity, the body part entity, the roof part entity, and the decorative part entity, and extracts the geometric parameters, material parameters, processing accuracy requirements, and spatial positioning parameters of the components; for complex components such as mortise and tenon joints, it extracts the core structural parameters such as the size, angle, and overlap depth of the tenon and mortise.

[0128] Intelligent assembly relationship recognition: Automatically parses spatial subordination and physical connection relationships between components through relation entities in IFC files; verifies modular matching relationships between components by combining predefined modular association constraint rules of the ontology; generates standardized relation triples; assigns a temporary unique identifier to each extracted component, providing a basis for subsequent global ID alignment.

[0129] The targeted preprocessing process for structured production data is as follows: deduplication, missing value filling, (3σ criterion) outlier removal, and normalization are performed on material property testing, equipment operation, and quality acceptance data, and the data is archived according to the entity type of the ontology architecture.

[0130] To address the issues of low accuracy in extracting multi-source heterogeneous knowledge from timber structures and the inability to reuse implicit experience, this invention employs a multimodal knowledge extraction method that integrates prior knowledge in the timber structure field. This method achieves high-precision structured transformation of classical texts, drawings, and oral experiences from craftsmen, significantly improving the reuse rate of domain knowledge.

[0131] Step S3: Multimodal knowledge extraction and knowledge graph construction that integrates prior domain knowledge.

[0132] Based on the three-dimensional hierarchical domain ontology architecture constructed in step S1, a multimodal knowledge extraction method that integrates prior knowledge of the timber structure domain is used to complete knowledge extraction. After cross-validation, conflict resolution, and entity alignment, a timber structure construction-level knowledge graph is constructed.

[0133] The inputs to step S3 are: standardized preprocessed datasets (including Song and Qing dynasty classics, process documents, acceptance reports, preprocessed drawings, BIM model dataset design drawings, as-built drawings, BIM models, preprocessed audio and video transcribed text, craftsmen's oral accounts, operation demonstrations, intangible cultural heritage interviews, etc.), multi-source knowledge priority rule base, and a three-dimensional hierarchical domain ontology architecture; the outputs of step S3 are: production-grade wooden building knowledge graph and standardized knowledge triplet library.

[0134] Please see the appendix Figure 3 Step S3 includes the following sub-steps:

[0135] Step S31: Multimodal knowledge extraction.

[0136] For different data types, extraction methods adapted to the timber construction field are adopted to address industry pain points such as low accuracy in extracting obscure terms and difficulty in extracting textual and graphical information. Specific extraction methods include:

[0137] ① Text Knowledge Extraction: A pre-constructed dictionary for the timber construction field is built based on an ontology terminology system. This dictionary includes terminology specific to timber construction, mappings of Song and Qing dynasty terminology, and alternative names for local schools of thought. An entity and relation extraction model integrating prior domain knowledge is constructed. The core optimization of the entity and relation extraction model is a token-level attention mechanism guided by domain terminology. This mechanism automatically focuses on timber construction terminology through attention weights, addressing the issues of insufficient attention to obscure terms and low extraction accuracy in general models. The final output consists of entities, attributes, and relations conforming to the ontology definition, generating standardized knowledge triples <subject, predicate, object>. The core of the entity and relation extraction model is the token-level attention mechanism guided by domain terminology, implemented as follows:

[0138] A pre-constructed terminology dictionary for the timber construction field is built, and a corresponding pre-trained embedding vector is generated for each term. When the entity and relation extraction model processes the input text sequence, the following operations are performed on each text segment: First, it is determined whether the text segment matches a domain term in the timber construction terminology dictionary—if it matches, the pre-trained embedding vector of that domain term is extracted and incorporated into the BERT embedding layer; if it does not match, the zero vector is used as the domain feature representation of that text segment. Subsequently, the entity and relation extraction model uses a learnable attention scoring function to calculate the relevance score between the hidden state of each text segment and its corresponding domain feature vector, and performs Softmax normalization on the scores of the entire sequence to obtain the attention weight for each domain term. The magnitude of this attention weight directly reflects the degree of attention the entity and relation extraction model pays to that text segment: domain terms receive higher attention weights due to higher relevance scores, while the weights of non-domain terms are suppressed.

[0139] The aforementioned attention weights are applied to the original hidden layer vectors of domain terms to generate weighted semantic embedding vectors that incorporate domain prior knowledge. These vectors are then sequentially input into the BiLSTM layer and the CRF layer. The entity label sequence is then constrained and decoded in the CRF layer, ultimately achieving high-precision extraction of entities and relationships in the timber structure field.

[0140] ② Drawing / BIM Model Knowledge Extraction: A finely tuned instance segmentation model is used to separate text and images. An improved contour extraction algorithm is employed to vectorize the drawing area, generating vector drawings and extracting contour features. An object detection model is used to identify dimensions, mortise and tenon joints, and component numbers within the drawings. OCR is used to extract annotation parameter values ​​and optimize technical terminology. Parametric models of components are generated, and BIM model IFC data is parsed to extract core construction parameters, components, component assembly, and attribute mappings, generating standardized drawing / BIM knowledge triples.

[0141] ③ Audio and video technical knowledge extraction: The audio and video are optimized with professional terminology for wood construction through speech recognition correction. Implicit knowledge such as process flow, experience parameters, defect avoidance, and segment extraction measures are extracted from the spoken text after speech recognition correction. Implicit knowledge is structured and mapped to ontology. The credibility of craftsmen's experience knowledge is graded and standardized technical knowledge triplets are generated.

[0142] Step S32: Cross-validation and conflict resolution.

[0143] In step S31, the same attribute parameters of the same entity in the three types of sources (text knowledge, drawing / BIM model knowledge, and audio-visual technology knowledge) are compared. If the deviation of the dimensional parameter exceeds the preset first threshold (e.g., the first threshold is 3%), or the deviation of the process parameter exceeds the preset second threshold (e.g., the second threshold is 5%), it is marked as a parameter conflict. Automatic resolution is carried out in strict accordance with the multi-source knowledge priority rules of step S1. If automatic resolution is not possible, expert manual review and calibration are triggered. After the review is passed, it is included in the initial knowledge triplet.

[0144] Step S33: Standardized entity alignment and knowledge graph construction.

[0145] A dual entity alignment mechanism of "ontology standard encoding benchmark alignment + semantic similarity supplementary alignment" is adopted: First, the entity definition in the three-dimensional hierarchical domain ontology architecture is used as the benchmark to align different name expressions of the same entity; then, the semantic similarity model is used to supplement the alignment of local gender names and colloquial names, and a unique global ID is assigned to each entity; the standardized primary knowledge triples are stored in the graph database to construct a production-level wooden building knowledge graph, and the visualization engine of the production-level wooden building knowledge graph provides visualization query and editing functions for entities, relations and attributes.

[0146] Step S4: Three-layer progressive integration construction process reasoning.

[0147] In step S4, the full-dimensional input information of the project to be built / repaired is obtained. After the structured parsing is completed, it is input into the knowledge graph constructed in step S3. Through a three-layer progressive reasoning framework of semantic coarse screening (first layer, i.e. step S42), rule fine screening (second layer, i.e. step S43), and multi-constraint fusion optimization (third layer, i.e. step S44), the optimal construction / repair process scheme is automatically generated.

[0148] The input for step S4 is: the full-dimensional requirements of the project to be processed (to be built / repaired, etc.) and the production-grade wooden building knowledge graph; the output for step S4 is: the standardized optimal construction / repair process scheme.

[0149] Please see the appendix Figure 4 Step S4 includes the following sub-steps:

[0150] Step S41: Perform structured analysis and standardized vector transformation on all dimensions of the requirements of the project to be processed.

[0151] Step S41 includes the following sub-steps:

[0152] Step S411: Obtain the full-dimensional requirement information of the project to be processed, complete the standardized parsing based on the three-dimensional integrated hierarchical domain ontology architecture of step S1, and generate a standardized inference input vector set.

[0153] The standardized inference input vector set includes: project basic feature vector, component design feature vector, resource constraint feature vector, and compliance and status feature vector.

[0154] Among them, the basic feature vector of the project is generated by extracting the building form, number of bays, number of depths, roof type, modular system, protection level, and building era.

[0155] Component design feature vector: Extract the component list, mortise and tenon joint details, geometric dimensions, processing accuracy requirements, and assembly constraints from the BIM model / design drawings to generate the component feature vector X2.

[0156] Resource constraint feature vector: Extract the tree species, material properties, available processing equipment, and construction period requirements of the timber used to generate a resource feature vector X3.

[0157] Compliance and Status Feature Vector: Extract the standards and specifications that the project must follow, the requirements for cultural relic restoration, the current status of component defects, and the quality acceptance standards to generate a constraint feature vector X4.

[0158] The above standardized inference input vectors are concatenated to generate a standardized inference input vector set X=[X1,X2,X3,X4].

[0159] Step S42: Coarse screening of candidate process schemes based on semantic similarity.

[0160] Specifically, a pre-trained domain semantic matching model is loaded, and the standardized inference input vector set X of the project to be processed is used to calculate the semantic similarity with the component-level process templates and substructure process modules in the knowledge graph. The top N (Top-N, N is preferably 10) candidate process schemes (or module combinations) with the highest similarity are selected to generate a candidate process scheme set.

[0161] For entirely new non-standard projects where no similar historical cases exist in the knowledge graph, this layer automatically breaks down the project into independent components or substructures (such as brackets, beams, and roof corners), and matches the process templates of each component separately, thereby avoiding the problem of failure due to reliance on the similarity of the overall case. If a component has no matching template, the process transitions to the zero-shot inference branch—generating the process plan for that component from scratch entirely based on the underlying mandatory constraint rules and entity attribute relationships in step S1.

[0162] The zero-sample reasoning branch is based on the underlying mandatory constraint rules and entity attribute relationships in the ontology. It uses a forward chain rule engine or graph search algorithm to combine the process sequence and process parameters that meet the compliance requirements layer by layer to generate a process solution that conforms to the ontology definition.

[0163] Step S43: Compliance screening based on the underlying mandatory constraint rules of ontology (rule screening - progressive compliance verification).

[0164] Specifically, an expert rule base is constructed based on the underlying mandatory constraint rules embedded in the ontology in step S1, and progressive compliance verification is performed on the candidate process scheme set: ① Form compliance verification, filtering out schemes that do not meet the requirements of the project era and regulations; ② Structural safety verification, filtering out schemes that do not meet the mandatory constraints of module, component size, and tenon and mortise joint; ③ Process logic verification, filtering out schemes that do not meet the process sequence constraints; ④ Constraint condition verification, filtering out schemes that do not meet the requirements of equipment, materials, construction period, and cultural relic protection; after verification, a pre-selected process scheme set is obtained.

[0165] Step S44: Multi-objective optimization reasoning that incorporates domain constraints.

[0166] With multiple optimization objectives of "maximizing compliance, processing efficiency, construction cost, quality defect rate, and minimizing repair intervention," a graph attention network reasoning model integrating mandatory constraints in the timber structure field is constructed. The graph attention network reasoning model takes the subgraph structure of the knowledge graph as input, transforms the underlying compliance rules embedded in the ontology into reasoning constraints, and automatically learns the influence weights of different entities and constraints on the process scheme through the attention mechanism. It iteratively optimizes the process sequence, process parameters, and acceptance standards of the pre-selected process scheme, and finally outputs the optimal construction / repair process scheme, including component processing process cards, CNC machining parameter sets, assembly process flow, on-site construction plan, and quality acceptance standards.

[0167] To address the issue of poor adaptability of general AI models in the timber construction field, this invention specifically addresses industry pain points in timber construction scenarios by performing domain-specific adaptation and optimization on knowledge extraction and graph attention network inference models. This solves the problems of low accuracy and inability to meet engineering requirements of general models.

[0168] To address the problems of existing timber construction process planning relying on manual experience, poor adaptability to non-standard projects, and lack of compliance assurance, this invention adopts a three-layer progressive fusion process reasoning framework. It deeply integrates the mandatory constraints of knowledge in the timber construction field with data-driven intelligent optimization, taking into account both the compliance of construction standards and the personalized adaptability of non-standard projects, and completely eliminating the reliance on craftsmen's experience.

[0169] Step S5: Verify the implementation of the optimal construction / repair process and iteratively optimize the knowledge graph.

[0170] In step S5, the optimal construction / repair process scheme is verified and the knowledge graph is iterated to form a closed loop of intelligent operation throughout the entire process.

[0171] The input for step S5 is the optimal construction / repair process scheme, and the output for step S5 is the implementable scheme, full life cycle traceability data, and the iterated knowledge graph.

[0172] Please see the appendix Figure 5 Step S5 includes the following sub-steps:

[0173] Step S51: Simulation verification of the optimal construction / repair process scheme.

[0174] Specifically, the processing parameters in the optimal construction / repair process scheme are input into the CNC machining simulation software to conduct tool path collision simulation and wood cutting stress simulation. If the simulation fails, the defect information is fed back to the graph attention network inference model in step S44, and the process inference is re-executed until the simulation passes. The construction procedure scheme is input into the virtual construction simulation platform to conduct construction procedure pre-playing and collision detection. If the simulation passes, the final executable process scheme is output, thereby verifying the feasibility of the optimal construction / repair process scheme.

[0175] Step S52: Implementation and Data Collection of the Final Executable Process Scheme: The verified final executable process scheme is distributed to the production workshop and construction project department to drive CNC equipment to complete component processing and guide on-site assembly / repair operations; collect component size inspection data, processing quality and defect data, construction progress and process data, completion acceptance data, material and equipment operation data, defect and rework data, and IoT sensing data after processing, and bind them to the unique global ID of the corresponding component entity.

[0176] Step S53: Knowledge graph self-iterative optimization.

[0177] Specifically, based on the collected data, the implementation effect of the final executable process solution is evaluated in two dimensions (compliance and performance indicators). If the compliance assessment is passed and the performance indicator assessment meets the standards, the final executable process solution is updated to the knowledge graph as a new knowledge instance, and a corresponding triple is added, which is a positive knowledge update, and the qualified case is added to the database. If the compliance assessment fails, or the compliance assessment is passed but the performance indicator assessment does not meet the standards, that is, there is a quality defect or room for optimization, the causes are analyzed and reverse rule optimization is performed, that is, the underlying mandatory constraint rules, process adaptation relationship and graph attention network inference model weights in the ontology are updated, and the self-iterative optimization of knowledge graph and graph attention network inference model is completed to continuously improve the subsequent inference accuracy.

[0178] To address the disconnect between knowledge construction and practical application, this invention establishes a fully intelligent closed loop encompassing "graph construction – process reasoning – implementation – feedback optimization," enabling self-iterative optimization of the production-grade wooden building knowledge graph and graph attention network reasoning model, thereby continuously improving construction quality and intelligence levels.

[0179] Please see the appendix Figure 6 A knowledge graph construction and construction process reasoning system for wooden architecture, comprising:

[0180] Domain ontology construction and management module: used to execute step S1, construct a three-dimensional integrated layered domain ontology architecture for the entire creation process, and predefine the core entity layer, standardized attribute layer, hierarchical entity system, spatial association rule layer, full standardized attribute set, standardized specification layer and underlying mandatory constraint rules.

[0181] Multi-source data preprocessing module: Used to execute step S2, complete the collection of multi-source raw data of the entire wooden structure construction chain, directional preprocessing in the wooden structure scenario and binding of full-link traceability identifiers, and generate a standardized preprocessed dataset; at the same time, it is responsible for the storage, backup and version management of the entire process data, and supports the unified management of structured data, semi-structured data and unstructured data.

[0182] Multimodal knowledge extraction and graph construction module: used to execute step S3, based on a three-dimensional integrated hierarchical domain ontology architecture to complete multimodal knowledge extraction, cross-validation, conflict resolution and entity alignment that integrates domain prior knowledge, and construct a production-grade wooden building knowledge graph.

[0183] Project Requirements Analysis and Process Reasoning Module: This module is used to execute step S4, complete the standardized analysis of all dimensions of the requirements of the project to be processed, and automatically generate the optimal construction / repair process solution through a three-layer progressive reasoning framework.

[0184] Simulation verification and iterative optimization module: used to execute step S5, complete the simulation verification of the final executable process scheme, collect data for implementation, and perform self-iterative optimization of the knowledge graph of construction-level wooden structures.

[0185] Preferably, depending on actual usage needs, other functional modules such as system management and security modules may also be included.

[0186] This invention first constructs a three-dimensional, hierarchical domain ontology architecture encompassing "value perception – physical structure – technological features" for the entire timber construction process, embedding mandatory compliance constraints rules for the entire construction process from the bottom layer. Addressing the industry pain point of multi-source heterogeneous data in timber construction, it employs a multimodal knowledge extraction method that integrates prior knowledge in the timber construction domain to complete knowledge extraction, cross-validation, and standardized fusion, constructing a construction-level timber building knowledge graph. It proposes a three-layer progressive reasoning framework of "semantic coarse screening – rule fine screening – multi-constraint fusion optimization," automatically generating compliant and adaptable final executable process solutions based on the full-dimensional needs of the project to be constructed / renovated. Finally, through implementation and result feedback, it achieves self-iterative optimization of the construction-level timber building knowledge graph and the graph attention network reasoning model.

[0187] This invention solves the core problems of existing knowledge graphs of timber construction being disconnected from construction practice, process planning being highly dependent on craftsmen's experience, poor adaptability of non-standard components, and inability to guarantee compliance. It realizes the digital inheritance and intelligent reuse of knowledge of timber construction, improves the efficiency of timber construction and the yield rate of component processing, and can be widely applied to the whole process planning, intelligent decision-making and implementation of ancient building restoration, antique timber construction, and modern industrialized production of timber buildings.

[0188] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the invention. Therefore, any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing a knowledge graph of wooden architecture and reasoning about its construction techniques, characterized by: Includes the following steps: Step S1: Construct an ontology architecture for the entire construction process of timber-framed buildings; Step S2: Acquisition and targeted preprocessing of multi-source heterogeneous data for timber structure construction; Step S3: Multimodal knowledge extraction and knowledge graph construction integrating domain prior knowledge; Step S4: Reasoning for the three-layer progressive integration construction process; Step S4 includes the following sub-steps: Step S41: Perform structured analysis and standardized vector transformation on all dimensions of the requirements of the project to be processed; Step S41 includes the following sub-steps: Step S411: Obtain the full-dimensional requirement information of the project to be processed, and complete the standardized parsing based on the three-dimensional integrated hierarchical domain ontology architecture of step S1 to generate a standardized inference input vector set; the standardized inference input vector set includes: project basic feature vector, component design feature vector, resource constraint feature vector, compliance and status feature vector; Among them, the basic feature vector of the project is generated by extracting the building form, number of bays, number of depths, roof type, modular system, protection level, and building era. Component design feature vector: Extract the component list, mortise and tenon joint details, geometric dimensions, machining accuracy requirements, and assembly constraints from the BIM model / design drawings to generate the component feature vector X2; Resource constraint feature vector: Extract the tree species, material properties, available processing equipment, and construction period requirements of the timber used to generate a resource feature vector X3; Compliance and Status Feature Vector: Extract the standards and specifications that the project must follow, the requirements for cultural relic restoration, the current status of component defects, and the quality acceptance standards to generate a constraint feature vector X4; The above standardized inference input vectors are concatenated to generate a standardized inference input vector set X=[X1,X2,X3,X4]; Step S42: Coarse screening of candidate process solutions based on semantic similarity; Load the pre-trained domain semantic matching model, calculate the semantic similarity between the standardized inference input vector set X of the project to be processed and the component-level process templates and substructure process modules in the knowledge graph, filter out the top N candidate process schemes with the highest similarity, and generate a set of candidate process schemes. Step S43: Compliance screening based on the underlying mandatory constraint rules of ontology; Based on the underlying mandatory constraint rules embedded in the ontology in step S1, an expert rule base is constructed to perform progressive compliance verification on the candidate process scheme set: ① Form compliance verification, filtering out schemes that do not meet the requirements of the project era and regulations; ② Structural safety verification, filtering out schemes that do not meet the mandatory constraints of module, component size, and tenon and mortise joint; ③ Process logic verification, filtering out schemes that do not meet the process sequence constraints; ④ Constraint condition verification, filtering out schemes that do not meet the requirements of equipment, materials, construction period, and cultural relic protection; After verification, a pre-selected process scheme set is obtained. Step S44: Multi-objective optimization reasoning that incorporates domain constraints; With multiple optimization objectives of "maximizing compliance, processing efficiency, construction cost, quality defect rate, and minimizing repair intervention," a graph attention network reasoning model integrating mandatory constraints from the timber structure field is constructed. The graph attention network reasoning model takes the subgraph structure of the knowledge graph as input, transforms the underlying compliance rules embedded in the ontology into reasoning constraints, and automatically learns the influence weights of different entities and constraints on the process scheme through the attention mechanism. It iteratively optimizes the process sequence, process parameters, and acceptance standards of the pre-selected process scheme, and finally outputs the optimal construction / repair process scheme, including component processing process cards, CNC machining parameter sets, assembly process flow, on-site construction scheme, and quality acceptance standards. Step S5: Verify the implementation of the optimal construction / repair process and iteratively optimize the knowledge graph. 2.The method according to claim 1, wherein the method is characterized in that: Step S1 includes the following sub-steps: Step S11: Construct a three-dimensional hierarchical domain ontology architecture that integrates value perception, entity construction, and process characteristics; Step S12: Predefine the creation of guided semantic relationships between entities; Step S13: Embed the underlying mandatory constraint rules of the ontology.

3. The method for constructing a knowledge graph of wooden structures and reasoning about construction techniques according to claim 2, characterized in that: In step S11, the three-dimensional integrated hierarchical domain ontology architecture includes: Dimension 1, Value Perception Dimension: Predefined core entity layer and standardized attribute layer; Among them, the core entity layer predefined three categories of core entities: protection level entity, architectural era entity, and architectural regulation entity; the standardized attribute layer is used to standardize the description of the cultural attributes, protection level attributes, historical evolution attributes, form level attributes, and repair intervention restriction attributes of wooden buildings, providing semantic support for the principle of minimum intervention in cultural relic repair scenarios. Dimension Two: Entity Construction Dimension: Predefined hierarchical entity system and spatial association rule layer; the hierarchical entity system is based on a hierarchical line classification method of "building part → core component → component", predefined with all timber component entities under the four major building parts: platform part entity, body part entity, roof part entity, and decorative part entity, establishing a hierarchical entity system from the overall building to the smallest functional unit; the spatial association rule layer predefined spatial subordination rules, assembly and overlap relationship rules, and modular association constraint rules; Dimension Three: Technological Characteristics Dimension: Predefined full-scale standardized attribute set and standardized specification layer; the full-scale standardized attribute set includes six core attributes: material attributes, structural attributes, processing technology attributes, construction procedure attributes, defect characteristic attributes, and spatial positioning attributes. Each attribute has a unified enumeration value specification, which fully describes the physical characteristics, technological requirements, and status information of the timber components; the standardized specification layer predefined attribute value range enumeration specification, process parameter threshold specification, quality acceptance standard specification, and defect level coding specification. In step S12, based on the inherent logic of timber construction, construction-oriented semantic relationships between entities are predefined. These construction-oriented semantic relationships include: composition relationship, modular relationship, process adaptation relationship, process sequence relationship, causal relationship, constraint relationship, compliance correspondence relationship, and spatial subordination relationship, providing a semantic basis for the relationship construction and process reasoning of the knowledge graph. In step S13, the underlying mandatory constraint rules of the ontology include: ① Terminology unification and disambiguation rules: Establish a standardized mapping library of wooden structure terminology, clarify the mapping of synonyms for Song and Qing dynasty terminology, the mapping of aliases for official and local school terminology, and the disambiguation rules for synonyms / different objects with the same name. ② Multi-source knowledge priority rules: Establish a priority system of "National mandatory standards > Recommended national standards > Industry standards > Official construction records > Local practices and regulations > Historical verification of process data > Experience of intangible cultural heritage inheritors > Documentary materials"; For cultural relic protection projects, an additional prerequisite rule is added: "The principle of preserving the original state of cultural relics" has the highest priority; ③ Create mandatory compliance rules: predefined modular mandatory constraint rules, structural safety mandatory constraint rules, process sequence constraint rules, and minimum intervention rules for cultural relic restoration.

4. The method for constructing a knowledge graph of wooden structures and reasoning about construction techniques according to claim 1, characterized in that: Step S2 includes the following sub-steps: Step S21: Collect multi-source raw data for the entire timber construction chain; Step S22: Perform targeted preprocessing on the multi-source raw data of the entire wooden structure construction chain.

5. The method for constructing a knowledge graph of wooden structures and reasoning about construction techniques according to claim 4, characterized in that: In step S21, the scope of data collection for multi-source raw data across the entire timber construction chain includes: ① Standards and specifications: National / industry standards related to timber structure design, construction, and cultural relic protection, including the *Yingzao Fashi* (Building Standards), the *Qing Gongbu Gongzao Zuofa Zeli* (Regulations and Examples of Engineering Practices of the Qing Dynasty), and local regulations and examples; ② Craftsmanship knowledge: historical construction process documents, component processing SOPs, audio and video recordings of intangible cultural heritage inheritors' oral accounts, and craftsman's operation manuals; ③ Production measurement data: wood property test data, equipment operation data, component BIM model / design drawings, quality defects and rework records, acceptance reports; ④ Project Case Studies: Complete documentation of implemented ancient building restoration and antique-style wooden structure projects; In step S22, the targeted preprocessing process for documents / drawings is as follows: Considering the characteristics of vertical layout, deep text-image binding, dense annotation of obscure terminology, and blurred lines in hand-drawn drawings, PDF scans and drawing scans undergo single-page splitting, noise reduction, grayscale conversion, and skew correction preprocessing. Then, a layout analysis model finely tuned using a dataset specific to the timber construction field is employed to divide the drawings / documents into three core areas: text area, drawing area, and table area. Each partition is bound to a unique traceability identifier in the format "data source number - chapter number - page number - partition number" for knowledge tracing, conflict location, and version management, recording data source, chapter, and page number information. The targeted preprocessing process for audio and video is as follows: ASR speech transcription and endpoint detection are performed on the audio and video of interviews with intangible cultural heritage inheritors and artisans' operations. The speech recognition model is optimized for professional terminology, implicit knowledge is extracted, structured text is generated, and source, time period, and speaker information are bound. The directional preprocessing flow for BIM model types is as follows: IFC data parsing, component parameter extraction, and assembly relationship identification. The targeted preprocessing process for structured production data is as follows: deduplication, missing value filling, outlier removal, and normalization are performed on material property testing, equipment operation, and quality acceptance data, and the data is then archived according to the entity type of the ontology architecture.

6. The method for constructing a knowledge graph of wooden structures and reasoning about construction techniques according to claim 1, characterized in that: Step S3 includes the following sub-steps: Step S31: Multimodal knowledge extraction; Step S32: Cross-validation and conflict resolution; Step S33: Standardized entity alignment and knowledge graph construction.

7. The method for constructing a knowledge graph of wooden structures and reasoning about construction techniques according to claim 6, characterized in that: In step S31, the multimodal knowledge extraction includes: ① Text knowledge extraction: A pre-constructed dictionary for the timber construction field based on the ontology terminology system is built. This dictionary includes exclusive terms for timber construction, mappings of Song and Qing dynasty terminology, and alternative names of local schools. An entity and relation extraction model that integrates prior knowledge of the domain is constructed. The entity and relation extraction model is optimized into a token-level attention mechanism guided by domain terminology. Through attention weights, it automatically focuses on timber construction terms and finally outputs entities, attributes, and relations that conform to the ontology definition, generating standardized knowledge triples <subject, predicate, object>. ② Drawing / BIM Model Knowledge Extraction: The model is segmented using instances to separate text and graphics. A contour extraction algorithm is used to vectorize the drawing area, generating vector drawings and extracting contour features. An object detection model is used to identify dimensions, mortise and tenon joints, and component numbers within the drawings. OCR is used to extract annotation parameter values ​​and optimize technical terminology. A parametric model of the components is generated, and the BIM model IFC data is parsed to extract construction parameters, components, component assembly, and attribute mapping, generating standardized drawing / BIM knowledge triples. ③ Audio and video technical knowledge extraction: The speech recognition is optimized for wood structure professional terminology in the audio and video. Implicit knowledge is extracted from the spoken text after speech recognition correction. Implicit knowledge is structured and mapped to the ontology. The credibility of craftsmen's experience knowledge is graded and standardized technical knowledge triples are generated. In step S32, the same attribute parameters of the same entity in the triples from the three sources in step S31 are compared. If the deviation of the size parameter exceeds the preset first threshold and the deviation of the process parameter exceeds the preset second threshold, it is marked as a parameter conflict. Automatic resolution is performed in strict accordance with the multi-source knowledge priority rules of step S1. If automatic resolution is not possible, expert manual review and calibration are triggered. After the review is passed, it is included in the primary knowledge triple. In step S33, a dual entity alignment mechanism of "ontology standard encoding benchmark alignment + semantic similarity supplementary alignment" is adopted: First, the entity definition in the three-dimensional integrated hierarchical domain ontology architecture is used as the benchmark to align different name expressions of the same entity; then, through the semantic similarity model, local gender names and colloquial names are supplemented and aligned, and a unique global ID is assigned to each entity; the standardized primary knowledge triples are stored in the graph database to construct a production-level wooden building knowledge graph, and the visualization engine of the production-level wooden building knowledge graph provides visualization query and editing functions.

8. The method for constructing a knowledge graph of wooden structures and reasoning about construction techniques according to claim 1, characterized in that: Step S5 includes the following sub-steps: Step S51: Simulation verification of the optimal construction / repair process scheme; Input the processing parameters from the optimal construction / repair process scheme into the CNC machining simulation software to conduct tool path collision simulation and wood cutting stress simulation. If the simulation fails, feed the defect information back to the graph attention network inference model in step S44 and re-execute the process inference until the simulation passes. Input the construction procedure scheme into the virtual construction simulation platform to conduct construction procedure pre-play and collision detection. If the simulation passes, output the final executable process scheme. Step S52: Implementation and Data Collection of the Final Executable Process Scheme: The verified final executable process scheme is distributed to the production workshop and construction project department to drive CNC equipment to complete component processing and guide on-site assembly / repair operations; collect component size inspection data, processing quality and defect data, construction progress and process data, completion acceptance data, material and equipment operation data, defect and rework data, and IoT sensing data after processing, and bind them to the unique global ID of the corresponding component entity; Step S53: Knowledge graph self-iterative optimization; Based on the collected data, the implementation effect of the final executable process solution is evaluated from two dimensions. If the compliance assessment is passed and the performance indicators meet the standards, the final executable process solution is updated to the knowledge graph as a new knowledge instance, and a corresponding triple is added. If the compliance assessment fails, or the compliance assessment is passed but the performance indicators do not meet the standards, i.e. there is a quality defect or room for optimization, the underlying mandatory constraint rules, process adaptation relationships and weights of the graph attention network inference model in the ontology are updated after analyzing the causes, and the self-iterative optimization of the knowledge graph and graph attention network inference model is completed.

9. A system for implementing the method for constructing a knowledge graph of wooden structures and reasoning about construction techniques as described in any one of claims 1-8, characterized in that: include: Domain ontology construction and management module: used to execute step S1, construct a three-dimensional integrated layered domain ontology architecture for the entire creation process, and predefine the core entity layer, standardized attribute layer, hierarchical entity system, spatial association rule layer, full standardized attribute set, standardized specification layer and underlying mandatory constraint rules; Multi-source data preprocessing module: used to execute step S2, complete the collection of multi-source raw data of the entire chain of timber construction, targeted preprocessing in timber construction scenarios and binding of full-chain traceability identifiers, and generate standardized preprocessed datasets; Multimodal knowledge extraction and graph construction module: used to execute step S3, based on a three-dimensional integrated hierarchical domain ontology architecture to complete multimodal knowledge extraction, cross-validation, conflict resolution and entity alignment that integrates domain prior knowledge, and construct a production-grade wooden building knowledge graph; Project Requirements Analysis and Process Reasoning Module: Used to execute step S4, complete the standardized analysis of all dimensions of the requirements of the project to be processed, and automatically generate the best construction / repair process solution through a three-layer progressive reasoning framework. Simulation verification and iterative optimization module: used to execute step S5, complete the simulation verification of the final executable process scheme, collect data for implementation, and perform self-iterative optimization of the knowledge graph of construction-level wooden structures.

Images