Method for constructing intelligent knowledge graph of building engineering review

Abstract

The application discloses a building engineering examination intelligent knowledge graph construction method, relates to the technical field of building engineering examination knowledge graph construction, and comprises the following steps: S1, acquiring geometric feature data and attribute feature data of building components in a building information model, acquiring building engineering examination specification articles, and parsing the specification articles into a rule set, wherein the rule set comprises a space constraint condition; S2, determining a space topological relation between each building component based on the geometric feature data, and generating a first relation pair set; the knowledge graph of the application takes building components as nodes, takes a space topological relation and a specification semantic relation as edges, so that the space correlation between each pair of components and the corresponding examination conclusion are stored in the graph in the form of an explicit relation, and when it is necessary to trace the basis of a certain rule violation conclusion, the original rule article generating the conclusion can be directly located through graph traversal.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph construction technology for building engineering review, specifically to an intelligent knowledge graph construction method for building engineering review. Background Technology

[0002] Building project review is a crucial step in ensuring that design deliverables comply with national regulations, industry standards, and safety requirements. It directly impacts project quality, construction safety, and the reliability of subsequent operation and maintenance. Traditional building project reviews primarily rely on manual methods, where reviewers use their professional knowledge to check the component layout and dimensional relationships in the building information model or 2D drawings against paper or electronic versions of the regulations. However, with increasingly complex building scales, a surge in the number of regulations, and frequent updates, manual review methods have gradually revealed shortcomings such as inefficiency, inconsistent standard implementation, and the potential to overlook hidden problems.

[0003] To improve review efficiency, automated review technologies based on Building Information Modeling (BIM) have emerged in recent years. These technologies first extract component attribute data from the BIM, then convert the code provisions into computer-processable logical expressions, and finally iterate and match the model data by writing specific query scripts or rule engines to output a list of components that violate the code. However, the conversion process of code provisions in this type of method is highly dependent on manual coding. Professionals need to manually write the code or structured query statements from the natural language descriptions of the code, resulting in low conversion efficiency and easy introduction of human error. In addition, each code is usually processed independently, and the review results are output in the form of discrete violation lists, failing to establish a network of relationships between components and between components and codes, making it difficult to trace the review results and conduct comprehensive analysis. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for constructing an intelligent knowledge graph for building engineering review.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for constructing an intelligent knowledge graph for architectural engineering review, specifically including the following steps:

[0006] S1. Obtain the geometric feature data and attribute feature data of building components in the building information model, obtain the building engineering review standard clauses, and parse the standard clauses into a rule set, the rule set containing spatial constraints;

[0007] S2. Based on the geometric feature data, determine the spatial topological relationship between each building component and generate a first set of relation pairs;

[0008] S3. Match the spatial constraints in the rule set with the geometric feature data, identify the component combinations that satisfy the spatial constraints, and generate a second set of relation pairs based on the matching results;

[0009] S4. Using the building components as nodes and the relationship pairs in the first set of relationship pairs and the second set of relationship pairs as edges, construct a knowledge graph for building engineering review.

[0010] Preferably, in step S1, the component type, spatial location coordinates, and external dimensions of each building component are extracted from the building information model. Based on the spatial location coordinates and external dimensions, a spatial range representation of each building component is generated. The standard provisions are parsed into a set of rules. Specifically, this includes: performing semantic parsing on the standard provisions, identifying spatial relationship descriptions in the standard provisions, mapping the spatial relationship descriptions to preset spatial relationship types, constructing rule entries based on spatial relationship types, and the rule entries constituting a set of rules.

[0011] Semantic analysis of the normative provisions, identifying spatial relationship descriptions within them, specifically includes the following steps:

[0012] The text data of the standard provisions is obtained, and word segmentation is performed on the text data to divide the continuous natural language text into a word sequence composed of independent words. Part-of-speech tagging is performed on each word in the word sequence to determine the part-of-speech category corresponding to each word. The part-of-speech categories include nouns, verbs, adjectives, prepositions and conjunctions.

[0013] Obtain a preset spatial relationship keyword library, which stores multiple keywords representing spatial location or direction. Traverse the word sequence, compare each word with the keywords in the spatial relationship keyword library, and mark the successfully matched words as candidate spatial relationship words.

[0014] For each candidate spatial relation word, a segment containing the candidate spatial relation word and a predetermined number of adjacent words is extracted from the word sequence as the context segment corresponding to the candidate spatial relation word. Dependency parsing is performed on the context segment to generate a syntactic dependency tree of the context segment. The syntactic dependency tree is marked with the subject-predicate relationship, verb-object relationship, prepositional object relationship and modification relationship between words.

[0015] Based on the syntactic dependency tree, the first and second noun words that have a direct syntactic association with the candidate spatial relation words are identified. The first and second noun words describe the two architectural components involved in the spatial relation, respectively.

[0016] The candidate spatial relation words, the first noun words, and the second noun words are combined to generate a spatial relation description triplet, which is the identified spatial relation description.

[0017] Preferably, in step S2, a first building component is selected from the set of building components, and a second building component is selected from the remaining components to form a pair of components to be judged, thereby obtaining a first spatial range representation corresponding to the first building component and a second spatial range representation corresponding to the second building component;

[0018] Based on the first spatial range representation and the second spatial range representation, calculate the minimum Euclidean distance between them. Based on the first spatial range representation and the second spatial range representation, perform spatial intersection detection to determine whether there is an intersection region between them. Based on the first spatial range representation and the second spatial range representation, perform inclusion detection to determine whether one spatial range representation is completely inside the other spatial range representation.

[0019] If the minimum Euclidean distance is zero and there is no intersection region, the component pair is determined to satisfy the adjacency relationship. If the minimum Euclidean distance is greater than zero and less than the preset proximity distance threshold and there is no intersection region, the component pair is determined to satisfy the adjacency relationship. If the intersection region is a non-empty set, the component pair is determined to satisfy the intersection relationship.

[0020] If all points in the first spatial range are inside the second spatial range and the volume of the first spatial range is smaller than the volume of the second spatial range, then the component pair is determined to satisfy the inclusion relationship, and the first building component is included by the second building component.

[0021] If all points in the second spatial range are inside the first spatial range and the volume of the second spatial range is smaller than the volume of the first spatial range, then the component pair is determined to satisfy the inclusion relationship, and the second building component is included by the first building component.

[0022] For a pair of components that satisfy at least one of the spatial relationship types of adjacent relationship, intersection relationship, or inclusion relationship, a relationship record is generated. The relationship record contains the first component identifier of the first building component, the second component identifier of the second building component, and the determined spatial relationship type. All relationship records are collected to generate a first relationship pair set.

[0023] Preferably, in step S3, each rule in the rule set is traversed. Each rule includes a condition part and a conclusion part. The condition part includes at least a first component type, a second component type, and a target spatial relationship type between the first component type and the second component type. The conclusion part includes the logical conclusion corresponding to the rule. The first component type, the second component type, and the target spatial relationship type are extracted from the condition part of the currently traversed rule.

[0024] Based on the first component type, a first component set matching the first component type is selected from the building components. Based on the second component type, a second component set matching the second component type is selected from the building components. Each component in the first component set is traversed as the third component, and each component in the second component set is traversed as the fourth component, thus forming the component pair to be queried.

[0025] For each component pair to be queried, check if there is a spatial topology relationship record corresponding to the component pair in the first relationship pair set, and if the spatial relationship type in the spatial topology relationship record is consistent with the target spatial relationship type. If it exists, obtain the logical conclusion from the conclusion part of the rule currently traversed, and generate a second relationship pair between the third and fourth components of the component pair. The second relationship pair contains the identifier of the third component, the identifier of the fourth component, and the relationship type determined by the logical conclusion. Collect all the generated second relationship pairs to form the second relationship pair set.

[0026] Preferably, in step S4, a set of building components is obtained, which contains multiple building components. Each building component corresponds to a component identifier. The set of building components is traversed, and a graph node is created for each building component. The graph node contains a node identifier and a set of node attributes. The node identifier adopts the component identifier of the building component, and the set of node attributes contains at least the component type, spatial location coordinates, and external dimension parameters of the building component.

[0027] Traverse the set of first relation pairs. For each first relation pair in the set of first relation pairs, the first relation pair contains a first component identifier, a second component identifier, and a spatial relation type. Starting from the graph node corresponding to the first component identifier and ending from the graph node corresponding to the second component identifier, create a first graph edge and attach a relation type attribute to the first graph edge. The relation type attribute takes the value of spatial relation type.

[0028] Traverse the set of second relation pairs. For each second relation pair in the set, the second relation pair contains a third component identifier, a fourth component identifier, and a relation type determined by the logical conclusion. Starting from the graph node corresponding to the third component identifier and ending at the graph node corresponding to the fourth component identifier, create a second graph edge and attach a relation type attribute to the second graph edge. The relation type attribute takes the relation type determined by the logical conclusion. Combine all created graph nodes, first graph edges, and second graph edges to obtain the knowledge graph of building engineering review.

[0029] This invention provides a method for constructing an intelligent knowledge graph for building engineering review, which has the following beneficial effects:

[0030] The knowledge graph of this invention uses building components as nodes and spatial topological relationships and normative semantic relationships as edges, so that the spatial association between each pair of components and its corresponding review conclusion are stored in the graph in the form of explicit relationships. When it is necessary to trace the basis of a certain violation conclusion, the original rule entry that generated the conclusion can be directly located through graph traversal. When it is necessary to analyze the compliance status of all components in a certain area, the complete result can be quickly obtained through subgraph query. Attached Figure Description

[0031] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see Figure 1 This invention provides a method for constructing an intelligent knowledge graph for building engineering review, comprising the following steps:

[0034] S1. Obtain the geometric feature data and attribute feature data of building components in the building information model, obtain the building engineering review standard clauses, and parse the standard clauses into a rule set, the rule set containing spatial constraints;

[0035] Furthermore, in S1, the component type, spatial location coordinates, and external dimensions of each building component are extracted from the building information model. Based on the spatial location coordinates and external dimensions, a spatial range representation of each building component is generated. The standard provisions are parsed into a set of rules. Specifically, this includes: performing semantic parsing on the standard provisions, identifying spatial relationship descriptions in the standard provisions, mapping the spatial relationship descriptions to preset spatial relationship types, constructing rule entries based on spatial relationship types, and the rule entries constituting a set of rules.

[0036] Semantic analysis of the normative provisions, identifying spatial relationship descriptions within them, specifically includes the following steps:

[0037] The text data of the standard provisions is obtained, and word segmentation is performed on the text data to divide the continuous natural language text into a word sequence composed of independent words. Part-of-speech tagging is performed on each word in the word sequence to determine the part-of-speech category corresponding to each word. The part-of-speech categories include nouns, verbs, adjectives, prepositions and conjunctions.

[0038] Obtain a preset spatial relationship keyword library, which stores multiple keywords representing spatial location or direction. Traverse the word sequence, compare each word with the keywords in the spatial relationship keyword library, and mark the successfully matched words as candidate spatial relationship words.

[0039] For each candidate spatial relation word, a segment containing the candidate spatial relation word and a predetermined number of adjacent words is extracted from the word sequence as the context segment corresponding to the candidate spatial relation word. Dependency parsing is performed on the context segment to generate a syntactic dependency tree of the context segment. The syntactic dependency tree is marked with the subject-predicate relationship, verb-object relationship, prepositional object relationship and modification relationship between words.

[0040] Based on the syntactic dependency tree, the first and second noun words that have a direct syntactic association with the candidate spatial relation words are identified. The first and second noun words describe the two architectural components involved in the spatial relation, respectively.

[0041] The candidate spatial relation words, the first noun words, and the second noun words are combined to generate a spatial relation description triplet, which is the identified spatial relation description.

[0042] Mapping spatial relationship descriptions to preset spatial relationship types involves the following steps:

[0043] Obtain a preset spatial relation type mapping table. The spatial relation type mapping table contains multiple preset spatial relation types. Each preset spatial relation type corresponds to a type identifier, a set of spatial relation keywords, and a set of syntactic pattern features. For each identified spatial relation description, extract the candidate spatial relation words it contains and the syntactic pattern features in its corresponding syntactic dependency tree.

[0044] Match the candidate spatial relation words with the spatial relation keyword groups corresponding to each preset spatial relation type in the mapping table, and filter out the preset spatial relation types that contain the candidate spatial relation words or their synonyms in the spatial relation keyword groups, as the candidate type set;

[0045] The syntactic pattern features of the spatial relation description are compared with the syntactic pattern features corresponding to each preset spatial relation type in the candidate type set, and the matching degree between the two is calculated. If there is a unique preset spatial relation type with a matching degree exceeding a preset threshold, the preset spatial relation type is determined as the spatial relation type after the spatial relation description is mapped.

[0046] If there are multiple preset spatial relationship types with matching degrees exceeding a preset threshold, then one of the multiple preset spatial relationship types is selected as the mapping result according to the preset priority rules. The priority rules are preset based on common semantic priorities in the field of architectural engineering review.

[0047] If there is no preset spatial relationship type with a matching degree exceeding the preset threshold, the spatial relationship description is marked as a pending item and output to the manual review queue.

[0048] It should be noted that Building Information Modeling (BIM), as a digital carrier in the field of building engineering, stores complete information descriptions of all building components. Among them, geometric feature data defines the shape and position of the component in three-dimensional space, and attribute feature data defines the engineering category and material properties of the component. By directly reading these data from the BIM, an accurate and consistent input basis can be provided for subsequent spatial relationship calculations and rule matching, ensuring that the constructed knowledge graph maintains a strict correspondence with the original design model. Geometric feature data and attribute feature data together constitute the node attributes of each building component in the knowledge graph, which is a bridge connecting physical entities and standard requirements.

[0049] The component type, spatial coordinates, and external dimensions of each building component are extracted from the building information model. The component type determines the applicable clause category in the code review. For example, different types of components such as walls, beams, columns, doors, and windows correspond to different review requirements. The spatial coordinates define the positioning base point or corner point coordinates of the component in the three-dimensional coordinate system. The external dimensions define the extension length of the component in the X-axis, Y-axis, and Z-axis directions. For example, the thickness, height, and length of the wall, or the cross-sectional width, cross-sectional height, and span of the beam. By extracting these three types of basic information, the originally complex building information model can be simplified into structured data records. Each record corresponds to a building component and includes its category identifier and geometric definition.

[0050] After extracting the spatial coordinates and external dimensions of each building component, a spatial extent representation for each component is generated based on these coordinates and dimensions. Discrete coordinate points are combined with dimension parameters to construct a geometric description that fully covers the three-dimensional space occupied by the component. This geometric description is the spatial extent representation. Specifically, if the building component is a regular geometric shape, such as a beam or column with a rectangular cross-section, a three-dimensional bounding box aligned to the axes is generated using the spatial coordinates as the base point and the external dimensions as the extension distances along the three axes, serving as the spatial extent representation of the component. If the building component is an irregular shape, its spatial extent is approximated by combining multiple regular geometric shapes or generating a convex hull. The spatial extent representation transforms abstract coordinate and dimension data into geometric objects capable of spatial operations, enabling subsequent determination of the spatial topological relationships between different components using geometric calculation methods. For example, whether two components intersect, whether one component contains another, or whether two components are adjacent.

[0051] Simultaneously with acquiring building component data, the code provisions of the building engineering review specifications are obtained and parsed into a set of rules. The set of rules contains spatial constraints. The building engineering review specifications exist in the form of natural language text, which contains a large number of qualitative or quantitative requirements regarding the spatial relationships between building components. For example, evacuation doors should open in the evacuation direction, the horizontal distance between door and window openings on both sides of a firewall should be no less than 2 meters, and the support length of a beam should not be less than 180 millimeters. These provisions cannot be directly understood and executed by the computer system and must be parsed into structured rule descriptions. Each rule in the rule set corresponds to the requirements of one or more code provisions. The rule set as a whole constitutes the basis for generating relationship edges in the knowledge graph, that is, what kind of relationship should exist between which components, and whether this relationship conforms to or violates the specifications.

[0052] In the process of parsing the normative clauses, the first step is to perform semantic parsing on the clauses to identify the spatial relationship descriptions. Then, the spatial relationship descriptions are mapped to preset spatial relationship types. Natural language processing technology is used to analyze the text of the normative clauses and extract keywords and phrases describing spatial location or direction, such as adjacent, between, inside, above, parallel, and perpendicular. These keywords and phrases constitute the spatial relationship descriptions. Due to the diversity and ambiguity of natural language expression, it is necessary to map the identified spatial relationship descriptions to a preset set of spatial relationship types with clear geometric meanings. The preset spatial relationship types are a set of finite and mutually exclusive relationship categories predefined according to the common needs of building engineering review, such as adjacent, intersecting, containing, parallel, perpendicular, and distance less than a threshold relationship. Through this mapping, the ambiguous natural language descriptions are converted into spatial relationship types that can be processed by computers.

[0053] After mapping spatial relationship descriptions to preset spatial relationship types, rule entries are constructed based on the spatial relationship types. This further structures the requirements of the normative clauses, forming rule entries consisting of a condition part and a conclusion part. The condition part at least includes the spatial relationship type and may also include component type restrictions, quantity restrictions, or specific numerical thresholds. The conclusion part indicates the review conclusion or derived relationship that should be generated when the condition is met. For example, for the clause that the horizontal distance between door and window openings on both sides of a firewall should be no less than 2 meters, the generated rule entry after parsing can be expressed as: the condition is that the component type is a door or window opening and the spatial relationship type is a distance of less than 2 meters, and the conclusion is that the norm is violated. By constructing such rule entries, the originally descriptive normative clauses are transformed into executable logical judgment units, enabling the system to automatically determine whether to trigger the rule entry for each pair of building components based on their spatial topology and component type, thereby generating the corresponding derived entity relationship pair.

[0054] S2. Based on the geometric feature data, determine the spatial topological relationship between each building component and generate a first set of relation pairs;

[0055] Furthermore, in S2, a first building component is selected from the set of building components, and a second building component is selected from the remaining components to form a pair of components to be judged, and a first spatial range representation corresponding to the first building component and a second spatial range representation corresponding to the second building component are obtained;

[0056] Based on the first spatial range representation and the second spatial range representation, calculate the minimum Euclidean distance between them. Based on the first spatial range representation and the second spatial range representation, perform spatial intersection detection to determine whether there is an intersection region between them. Based on the first spatial range representation and the second spatial range representation, perform inclusion detection to determine whether one spatial range representation is completely inside the other spatial range representation.

[0057] If the minimum Euclidean distance is zero and there is no intersection region, the component pair is determined to satisfy the adjacency relationship. If the minimum Euclidean distance is greater than zero and less than the preset proximity distance threshold and there is no intersection region, the component pair is determined to satisfy the adjacency relationship. If the intersection region is a non-empty set, the component pair is determined to satisfy the intersection relationship.

[0058] If all points in the first spatial range are inside the second spatial range and the volume of the first spatial range is smaller than the volume of the second spatial range, then the component pair is determined to satisfy the inclusion relationship, and the first building component is included by the second building component.

[0059] If all points in the second spatial range are inside the first spatial range and the volume of the second spatial range is smaller than the volume of the first spatial range, then the component pair is determined to satisfy the inclusion relationship, and the second building component is included by the first building component.

[0060] For a pair of components that satisfy at least one of the spatial relationship types of adjacent relationship, intersection relationship, or inclusion relationship, a relationship record is generated. The relationship record contains the first component identifier of the first building component, the second component identifier of the second building component, and the determined spatial relationship type. All relationship records are collected to generate a first relationship pair set.

[0061] It should be noted that the first spatial extent representation is a digital description of the geometric area occupied by the first building component in three-dimensional space. The first spatial extent representation is generated based on the spatial position coordinates and shape dimension parameters of the first building component. Using the spatial position coordinates of the first building component as the positioning base point and the extension length of the shape dimension parameters in the X-axis, Y-axis and Z-axis directions as the boundary, a minimum axis-aligned three-dimensional bounding box that can completely enclose the first building component is constructed. The three-dimensional bounding box is the first spatial extent representation. The first spatial extent representation enables the first building component to have a geometric shape that can participate in spatial calculations in the computer system.

[0062] The second spatial extent representation is a digital description of the geometric area occupied by the second building component in three-dimensional space. The second spatial extent representation is generated based on the spatial position coordinates and shape dimension parameters of the second building component. Using the spatial position coordinates of the second building component as the positioning base point and the extension length of the shape dimension parameters in the X, Y, and Z axes as the boundary, a minimum axis-aligned three-dimensional bounding box that can completely enclose the second building component is constructed. The three-dimensional bounding box is the second spatial extent representation. The second spatial extent representation enables the second building component to have a geometric shape that can participate in spatial calculations in the computer system.

[0063] The first spatial extent representation and the second spatial extent representation have the same mathematical expression form, both of which are axis-aligned three-dimensional bounding boxes defined by six planes. Each bounding box is uniquely determined by its minimum corner coordinates and maximum corner coordinates. Based on the first spatial extent representation and the second spatial extent representation, geometric operations such as spatial intersection detection, minimum Euclidean distance calculation and inclusion detection can be performed to quantitatively determine the spatial topological relationship between the first building component and the second building component.

[0064] Based on the spatial relationship judgment, it is determined whether two building components meet the adjacent relationship. In building engineering, there are many design requirements where components are close to each other but do not penetrate each other, such as the overlap of walls and floors, and the node connection of beams and columns. Geometrically, this relationship is represented by the fact that there is contact between the surfaces of the spatial range representations of the two components but no internal overlap. In specific judgment, the minimum Euclidean distance between the first spatial range representation and the second spatial range representation is first calculated. If the minimum distance is zero or less than the preset adjacent distance threshold, the intersection of the two spatial range representations is further checked. When the minimum distance is zero and there is no internal intersection, it indicates that the two components are exactly in surface contact and can be judged as adjacent. When the minimum distance is greater than zero but less than the preset threshold, it indicates that there is a small gap between the two components but can be regarded as adjacent in engineering, and is also judged as adjacent. The preset threshold is set in advance according to the construction error and design tolerance in the field of building engineering, so as to convert the numerical calculation results into a relationship judgment that conforms to the engineering semantics.

[0065] To determine whether two building components satisfy an intersection relationship, it is important to understand that unintentional spatial overlap of building components is generally not allowed in the design. However, there may be intentional overlaps at certain node structures or pipeline intersections. Geometrically, this relationship is represented by the existence of a common area in the spatial range of the two components. In specific determination, a spatial intersection detection is performed on the first spatial range representation and the second spatial range representation, and the intersection area is calculated. If the intersection area is a non-empty set, that is, the two spatial range representations share a common three-dimensional spatial volume, then it is determined to be an intersection relationship. In actual engineering, the intersection relationship needs to be further distinguished in combination with the component type: for example, the overlap of beams and columns in the node area is a normal structure, while the non-designed intersection of different pipelines is a collision conflict. Therefore, while determining the intersection relationship, the volume and location information of the intersection area are recorded as an auxiliary basis for subsequent rule matching.

[0066] In construction engineering, there are many nested or attached relationships between components. For example, doors are embedded in walls, pipes pass through floors, and equipment bases are attached to structural components. Geometrically, this relationship means that the spatial scope of one component is completely located inside the spatial scope of another component. In specific determination, an inclusion test is performed on the first and second spatial scopes. If all points on the first spatial scope are inside the second spatial scope, and the volume of the first spatial scope is smaller than that of the second spatial scope, then the first component is determined to be included by the second component. Conversely, if all points on the second spatial scope are inside the first spatial scope, then the second component is determined to be included by the first component. The determination of inclusion relationship needs to exclude the special case where the two components completely overlap. The latter should be classified as an overlap relationship or regarded as the same component.

[0067] S3. Match the spatial constraints in the rule set with the geometric feature data, identify the component combinations that satisfy the spatial constraints, and generate a second set of relation pairs based on the matching results;

[0068] Furthermore, in S3, each rule in the rule set is traversed. Each rule includes a condition part and a conclusion part. The condition part includes at least a first component type, a second component type, and a target space relationship type between the first component type and the second component type. The conclusion part includes the logical conclusion corresponding to the rule. The first component type, the second component type, and the target space relationship type are extracted from the condition part of the currently traversed rule.

[0069] Based on the first component type, a first component set matching the first component type is selected from the building components. Based on the second component type, a second component set matching the second component type is selected from the building components. Each component in the first component set is traversed as the third component, and each component in the second component set is traversed as the fourth component, thus forming the component pair to be queried.

[0070] For each component pair to be queried, check if there is a spatial topology relationship record corresponding to the component pair in the first relationship pair set, and if the spatial relationship type in the spatial topology relationship record is consistent with the target spatial relationship type. If it exists, obtain the logical conclusion from the conclusion part of the rule currently traversed, and generate a second relationship pair between the third and fourth components of the component pair. The second relationship pair contains the third component identifier, the fourth component identifier, and the relationship type determined by the logical conclusion. Collect all the generated second relationship pairs to form the second relationship pair set.

[0071] It should be noted that each normative clause is split into two parts, conditions and conclusions, when parsed into rule entries. The condition part describes the scenario in which the rule applies, such as if there are components of type A and components of type B, and the spatial relationship between them is R.

[0072] The conclusion section describes how the relationship between the two components should be labeled when the conditions are met. The logical conclusion is the specific content of the conclusion section, which represents the engineering semantics mapped from the specification stripes, and includes at least one of the following types:

[0073] Affirmative conclusion: This indicates that the component combination meets the specification requirements, such as generating specification-compliant relation edges between component pairs;

[0074] Negative conclusion: indicates that the component combination violates the specification requirements, such as generating a relationship edge that violates the specification between component pairs;

[0075] Derivative conclusions: This indicates that the combination of components triggers further review requirements, such as generating relationship edges between component pairs that require review or special review;

[0076] Building information models typically contain tens of thousands of building components. Performing rule matching on every pair of components would incur a huge computational overhead. Grouping and filtering components by type can limit the matching scope to specific categories related to the rules, greatly improving matching efficiency. By traversing all building components, components whose type attribute values ​​match the first component type are grouped into the first component set, and components whose type attribute values ​​match the second component type are grouped into the second component set. If the first component type and the second component type are the same, then the first component set and the second component set are the same set. In this case, it is necessary to avoid pairing the same component with itself in subsequent matching. The two component sets obtained by filtering constitute the candidate object space for rule matching.

[0077] For each component in the first set of components obtained from the filtering process, designated as the third component, and each component in the second set of components, designated as the fourth component, a query is performed to determine whether there exists a spatial topological relationship between them that matches the target spatial relationship type. Previous steps have generated a first set of relationship pairs containing spatial topological relationships between all building components. This set stores each pair of components with a spatial relationship and its relationship type in the form of key-value pairs or graph database edges. By retrieving the first set of relationship pairs, it is quickly determined whether a specific component pair meets the rule requirements. Specifically, for the component pair consisting of the currently processed third and fourth components, the first set of relationship pairs is searched for a record that simultaneously contains the identifiers of both components. If such a record exists, the spatial relationship type in the record is further compared to the target spatial relationship type. If they match, it indicates that the component pair geometrically satisfies the spatial constraint condition of the rule; if no record exists or the relationship type is inconsistent, it indicates that the component pair does not meet the condition.

[0078] For each pair of components that meet the spatial constraints, a new relation record is generated based on its corresponding logical conclusion. This record contains at least the third component identifier, the fourth component identifier, and the relation type determined by the logical conclusion, such as conforming to the specification, violating the specification, or requiring review. The relation records generated by all the component pairs that meet the conditions are collected to form the second relation pair set.

[0079] S4. Using the building components as nodes and the relationship pairs in the first set of relationship pairs and the second set of relationship pairs as edges, construct a knowledge graph for building engineering review.

[0080] Furthermore, in S4, a set of building components is obtained, which contains multiple building components. Each building component corresponds to a component identifier. The set of building components is traversed, and a graph node is created for each building component. The graph node contains a node identifier and a set of node attributes. The node identifier adopts the component identifier of the building component, and the set of node attributes contains at least the component type, spatial location coordinates, and external dimension parameters of the building component.

[0081] Traverse the set of first relation pairs. For each first relation pair in the set of first relation pairs, the first relation pair contains a first component identifier, a second component identifier, and a spatial relation type. Starting from the graph node corresponding to the first component identifier and ending from the graph node corresponding to the second component identifier, create a first graph edge and attach a relation type attribute to the first graph edge. The relation type attribute takes the value of spatial relation type.

[0082] Traverse the set of second relation pairs. For each second relation pair in the set, the second relation pair contains a third component identifier, a fourth component identifier, and a relation type determined by the logical conclusion. Starting from the graph node corresponding to the third component identifier and ending at the graph node corresponding to the fourth component identifier, create a second graph edge and attach a relation type attribute to the second graph edge. The relation type attribute takes the relation type determined by the logical conclusion. Combine all created graph nodes, first graph edges, and second graph edges to obtain the knowledge graph of building engineering review.

[0083] It should be noted that after generating the first and second sets of relation pairs, a knowledge graph for building engineering review is constructed using building components as nodes and relation pairs from the first and second sets of relation pairs as edges. This structurally integrates the discrete building components in the building information model with the two types of pre-calculated relation pairs, forming a graph data structure that conforms to the definition of graph theory. Each node represents an independent building component in the building information model, and each node carries the component's attribute information, including component type, spatial location coordinates, external dimensions, and spatial range representation. Edges represent the relationships between building components, including the first relation pairs generated by spatial topological relationships and the second relation pairs generated by matching standard rules. Through the organization of nodes and edges, information that was originally scattered in the building information model and standard clauses is uniformly incorporated into the same graph data framework.

[0084] When constructing the knowledge graph, each building component in the Building Information Model (BIM) is first instantiated as a node in the knowledge graph. The component records in the BIM are the basic carriers of knowledge units in the knowledge graph. Each component, as an independent entity, should have a unique representation in the graph. The set of building components is traversed, and a graph node is created for each component. The graph node contains a node identifier and a set of node attributes. The node identifier directly uses the unique identifier of that component in the BIM to ensure traceability of the correspondence with the original model. The node attribute set stores the component type, spatial coordinates, external dimensions, and spatial extent representation extracted in the previous steps. This completely migrates the structured data from the BIM to the node layer of the knowledge graph, enabling each node to represent a building entity with geometric definitions and engineering classifications.

[0085] After instantiating the nodes, relation pairs from the first relation pair set are added as edges to the knowledge graph. The principle behind this is that the first relation pair set records the inherent spatial topological relationships between building components. These relationships reflect the geometric connections between components in the design model and are fundamental to understanding the architectural spatial structure. Specifically, each relation record in the first relation pair set is traversed. Each record contains a first component identifier, a second component identifier, and a spatial relationship type. Starting from the node corresponding to the first component identifier and ending at the node corresponding to the second component identifier, a directed or undirected graph edge is created, and a relation type attribute is attached to this edge. The relation type attribute takes the value of one of the following: adjacency, intersection, or containment. By adding all first relation pairs as edges one by one, a skeleton network describing the architectural spatial structure is formed in the knowledge graph. This network visually presents which components are spatially adjacent, which components intersect each other, and which components are nested and contained within each other.

[0086] After adding the first relation pair, relation pairs from the second relation pair set are added as edges to the knowledge graph. The second relation pair set records semantic relations generated by matching normative clauses. Each relation record in the second relation pair set is traversed. Each record contains a third component identifier, a fourth component identifier, and a relation type determined by logical conclusions. The relation type may be compliant with the norm, violate the norm, or require further review. A graph edge is created with the node corresponding to the third component identifier as the starting point and the node corresponding to the fourth component identifier as the ending point. The edge is then attached with relation type attributes and traceability information. The traceability information points to the original rule entry that generated the relation pair, so as to explain the basis for review later. By adding all the second relation pairs as edges one by one, the semantic annotation of the normative dimension is superimposed on the knowledge graph, so that the graph, which originally simply describes geometric relations, has the knowledge connotation of compliance review.

[0087] Once all nodes and both types of edges are added, a complete knowledge graph for building engineering review is formed. The coexistence of the first and second relation pairs on the same set of nodes enables a deep integration of the geometric attributes and standard semantics of building components. In the final knowledge graph, each building component node establishes spatial associations with surrounding components through the first relation pair and standard semantic associations with related components through the second relation pair. Multiple edges may exist between the same pair of components, such as both spatial adjacency edges and standard compliance edges. These two edges describe the associations between components from different dimensions. For example, it can retrieve all pairs of components that violate the standard within a certain area, or query all other components that are spatially adjacent to a certain component and require further review.

[0088] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0089] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0090] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0091] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0092] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of protection of the described technical solution.

Claims

1. A method for constructing an intelligent knowledge graph for building engineering review, characterized in that, Includes the following steps: S1. Obtain the geometric feature data and attribute feature data of building components in the building information model, obtain the building engineering review standard clauses, and parse the standard clauses into a rule set, the rule set containing spatial constraints; S2. Based on the geometric feature data, determine the spatial topological relationship between each building component and generate a first set of relation pairs; Specifically, a pair of components including a first building component and a second building component is selected from the set of building components. The spatial range representations of the first building component and the second building component in the component pair are obtained. The minimum Euclidean distance between the spatial range representations is calculated, and spatial intersection detection and inclusion detection are performed. For a pair of components that are determined to satisfy at least one of the spatial relationship types of adjacent relationship, intersecting relationship or inclusion relationship, generate a relationship record that includes the first component identifier of the first building component, the second component identifier of the second building component and the determined spatial relationship type, and gather all relationship records to generate a first relationship pair set; S3. Match the spatial constraints in the rule set with the geometric feature data, identify the component combinations that satisfy the spatial constraints, and generate a second set of relation pairs based on the matching results; Each rule in the rule set is traversed. Each rule contains a condition part and a conclusion part. The condition part includes at least the first component type, the second component type, and the target space relationship type between the first component type and the second component type. The conclusion part contains the logical conclusion corresponding to the rule. Based on the first component type and the second component type, select the first component set that conforms to the first component type and the second component set that conforms to the second component type from the building components. Iterate through the components in the first component set as the third component and the components in the second component set as the fourth component to form the component pair to be queried. For each component pair to be queried, check whether there is a corresponding spatial topology relationship record in the first relation pair set and whether its spatial relationship type is consistent with the target spatial relationship type. If it exists, obtain the logical conclusion and generate a second relation pair between the third component and the fourth component, which includes the third component identifier, the fourth component identifier and the relationship type determined by the logical conclusion, and collect them into a second relation pair set. S4. Using the building components as nodes and the relationship pairs in the first set of relationship pairs and the second set of relationship pairs as edges, construct a knowledge graph for building engineering review. Get the set of building components, traverse the set of building components, and create a graph node for each building component; traverse the set of the first relation pairs, starting from the graph node corresponding to the first component identifier and ending from the graph node corresponding to the second component identifier, create the first graph edge, and attach a spatial relation type to the first graph edge; Traverse the second set of relation pairs, starting from the graph node corresponding to the third component identifier and ending at the graph node corresponding to the fourth component identifier, create a second graph edge, and attach the relation type determined by the logical conclusion to the second graph edge. Combine all created graph nodes, first graph edges and second graph edges to obtain the knowledge graph of building engineering review.

2. The intelligent knowledge graph construction method for building engineering review according to claim 1, characterized in that, In step S1, the component type, spatial coordinates, and external dimensions of each building component are extracted from the building information model. Based on the spatial coordinates and external dimensions, a spatial range representation of each building component is generated. The process of parsing normative clauses into a set of rules includes: performing semantic parsing on normative clauses, identifying spatial relationship descriptions in normative clauses, mapping spatial relationship descriptions to preset spatial relationship types, constructing rule entries based on spatial relationship types, and the rule entries constituting a set of rules.

3. The intelligent knowledge graph construction method for building engineering review according to claim 2, characterized in that, Semantic analysis of the normative provisions, identifying spatial relationship descriptions within them, specifically includes the following steps: The text data of the standard provisions is obtained, and word segmentation is performed on the text data to divide the continuous natural language text into a word sequence composed of independent words. Part-of-speech tagging is performed on each word in the word sequence to determine the part-of-speech category corresponding to each word. The part-of-speech categories include nouns, verbs, adjectives, prepositions and conjunctions. Obtain a preset spatial relationship keyword library, which stores multiple keywords representing spatial location or direction. Traverse the word sequence, compare each word with the keywords in the spatial relationship keyword library, and mark the successfully matched words as candidate spatial relationship words. For each candidate spatial relation word, a segment containing the candidate spatial relation word and a predetermined number of adjacent words is extracted from the word sequence as the context segment corresponding to the candidate spatial relation word. Dependency parsing is performed on the context segment to generate a syntactic dependency tree of the context segment. The syntactic dependency tree is marked with the subject-predicate relationship, verb-object relationship, prepositional object relationship and modification relationship between words. Based on the syntactic dependency tree, the first and second noun words that have a direct syntactic association with the candidate spatial relation words are identified. The first and second noun words describe the two architectural components involved in the spatial relation, respectively. The candidate spatial relation words, the first noun words, and the second noun words are combined to generate a spatial relation description triplet, which is the identified spatial relation description.

4. The intelligent knowledge graph construction method for building engineering review according to claim 1, characterized in that, If the minimum Euclidean distance is zero and there is no intersection region, the component pair is determined to satisfy the adjacency relationship. If the minimum Euclidean distance is greater than zero and less than the preset proximity distance threshold and there is no intersection region, the component pair is determined to satisfy the adjacency relationship. If the intersection region is a non-empty set, the component pair is determined to satisfy the intersection relationship. If all points in the first spatial range are inside the second spatial range and the volume of the first spatial range is smaller than the volume of the second spatial range, then the component pair is determined to satisfy the inclusion relationship, and the first building component is included by the second building component. If all points in the second spatial range are inside the first spatial range and the volume of the second spatial range is smaller than the volume of the first spatial range, then the component pair is determined to satisfy the inclusion relationship, and the second building component is included by the first building component.

5. The intelligent knowledge graph construction method for building engineering review according to claim 1, characterized in that, Obtain a set of building components. The set contains multiple building components, each with a corresponding component identifier. Traverse the set of building components and create a graph node for each building component. The graph node contains a node identifier and a set of node attributes. The node identifier uses the component identifier of the building component, and the set of node attributes contains at least the component type, spatial coordinates, and external dimensions of the building component. Traverse the set of first relation pairs. For each first relation pair in the set, the first relation pair contains a first component identifier, a second component identifier, and a spatial relation type. Starting from the graph node corresponding to the first component identifier and ending from the graph node corresponding to the second component identifier, create a first graph edge and attach a relation type attribute to the first graph edge. The relation type attribute takes the value of spatial relation type.

6. The intelligent knowledge graph construction method for building engineering review according to claim 5, characterized in that, Traverse the set of second relation pairs. For each second relation pair in the set, the second relation pair contains a third component identifier, a fourth component identifier, and a relation type determined by the logical conclusion. Starting from the graph node corresponding to the third component identifier and ending at the graph node corresponding to the fourth component identifier, create a second graph edge and attach a relation type attribute to the second graph edge. The relation type attribute takes the relation type determined by the logical conclusion. Combine all created graph nodes, first graph edges, and second graph edges to obtain the knowledge graph of building engineering review.

Images