A method, system, device and medium for constructing an integrated digital model of a full-motion simulator

Abstract

The present application discloses a method, system, device and medium for constructing an integrated digital model of a full-motion simulator, which belongs to the field of simulators. The present application is applicable to constructing digital models based on unstructured heterogeneous data, by obtaining the three-dimensional geometric model and unstructured technical documents of the full-motion simulator, obtaining a structured text information set from the document through named entity recognition and relationship extraction technology, and jointly constructing a heterogeneous knowledge graph containing two types of nodes representing technical entities and three-dimensional parts in combination with the three-dimensional model. The graph neural network is used to perform link prediction on the graph to infer the target edges representing the deep implicit association between the text information and the three-dimensional parts. These target edges are updated to the graph, and all the complete technical information associated with the parts is bound to the corresponding three-dimensional geometric model to generate an information-complete integrated digital model of the full-motion simulator, which can improve the efficiency of digital model construction and the correlation of model information.

Description

Technical Field

[0001] The present application relates to the field of simulators, and in particular to a method, system, equipment and medium for constructing an integrated digital model of a full-motion simulator. Background Art

[0002] As a highly complex and sophisticated system, the design, manufacturing, and installation of a full-motion simulator involve extensive technical data and interdisciplinary collaboration. Building an integrated digital model throughout the entire lifecycle enables information transfer and guidance at different stages simply by referring to the single model. This holds significant promise for improving R&D efficiency, ensuring manufacturing quality, and streamlining the installation process.

[0003] Existing methods for constructing such integrated digital models typically rely on product lifecycle management systems. For example, this involves establishing a unified data model to link design data from different stages. Another common approach is to manually interpret technical requirements from design documents using software such as computer-aided 3D interactive technology and attach them to the corresponding components of the 3D model in the form of annotations or attributes.

[0004] However, existing technical methods have significant limitations when processing massive amounts of heterogeneous unstructured data. The process of manual interpretation and manual association of information is not only time-consuming and labor-intensive, but also prone to errors. In addition, it is often only possible to establish surface correspondences that are clearly recorded in the documents. Existing methods are unable to effectively identify deep implicit associations, which ultimately leads to incomplete digital model information and low knowledge association density. Therefore, existing technical methods have technical problems such as low construction efficiency and insufficient correlation of model information. Summary of the Invention

[0005] The present application provides a method, system, device and computer storage medium for constructing an integrated digital model of a full-motion simulator, which can improve the efficiency of digital model construction and the correlation of model information, so as to solve the technical problems of low construction efficiency and insufficient model information correlation in the existing methods.

[0006] In a first aspect, the present application provides a method for constructing an integrated digital model of a full-motion simulator, which is suitable for constructing a digital model based on unstructured heterogeneous data. The method comprises:

[0007] Obtaining a 3D geometric model and unstructured technical documents of the full-motion simulator, wherein the unstructured technical documents include design technical requirement information and process technical requirement information, and the 3D geometric model includes geometric attribute information;

[0008] By performing named entity recognition and relationship extraction on unstructured technical documents, a structured text information set is obtained. The text information set includes technical entities and technical relationships between technical entities. The technical entities include text semantic information.

[0009] Based on the text information collection and the 3D geometric model, a heterogeneous knowledge graph is constructed. The heterogeneous knowledge graph includes first-class nodes representing technical entities and second-class nodes representing parts in the 3D geometric model, as well as edges between the first-class nodes and the second-class nodes. The edges between the first-class nodes represent the technical relationships between technical entities, and the edges between the second-class nodes represent the geometric adjacency relationships between parts.

[0010] Based on the edges in the heterogeneous knowledge graph, the textual semantic information associated with the first-category nodes, and the geometric attribute information associated with the second-category nodes, a graph neural network is used to perform link prediction to determine the target edge used to connect the first-category nodes with the second-category nodes.

[0011] The target edges are used to update the heterogeneous knowledge graph, and all first-type target nodes connected to the second-type target nodes through target edges are determined in the updated heterogeneous knowledge graph. The technical entities corresponding to the first-type target nodes and the technical relationships between the first-type target nodes are associated with the parts in the three-dimensional geometric model corresponding to the second-type target nodes to generate an integrated digital model of the full-motion simulator.

[0012] In one possible implementation, the method further includes: generating semantic vectors for the first-category nodes using a pre-trained language model based on textual semantic information associated with the first-category nodes, the pre-trained language model being trained based on textual data related to the field of full-motion simulators;

[0013] Based on the edges in the heterogeneous knowledge graph, the textual semantic information associated with the first-category nodes, and the geometric attribute information associated with the second-category nodes, a graph neural network is used to perform link prediction to determine the target edge used to connect the first-category nodes with the second-category nodes, including:

[0014] Based on the edges in the heterogeneous knowledge graph, the semantic vectors of the first-category nodes, and the geometric attribute information associated with the second-category nodes, a graph neural network is used for link prediction to determine the target edges used to connect the first-category nodes with the second-category nodes.

[0015] In one possible implementation, the method further includes: constructing a system dependency network based on the second type of nodes and the edges between the second type of nodes in the heterogeneous knowledge graph;

[0016] By calculating the topological centrality of each second-type node in the system dependency network, we can obtain the system importance index that characterizes the degree of influence of the second-type node on the overall connectivity of the system dependency network.

[0017] Based on the edges in the heterogeneous knowledge graph, the semantic vectors of the first-category nodes, and the geometric attribute information associated with the second-category nodes, a graph neural network is used to perform link prediction to determine the target edge used to connect the first-category nodes with the second-category nodes, including:

[0018] Based on the edges in the heterogeneous knowledge graph, the semantic vectors of the first-category nodes, and the geometric attribute information and system importance index associated with the second-category nodes, a graph neural network is used for link prediction to determine the target edges used to connect the first-category nodes with the second-category nodes.

[0019] In one feasible implementation, link prediction is performed using a graph neural network based on edges in a heterogeneous knowledge graph, semantic vectors of first-category nodes, and geometric attribute information and system importance index associated with second-category nodes to determine a target edge for connecting the first-category nodes with the second-category nodes, including:

[0020] The semantic vector of the first type of node is used as the initial feature vector of the first type of node, and the initial feature vector of the second type of node is generated by jointly encoding the geometric attribute information and system importance index associated with the second type of node;

[0021] Taking any node in the heterogeneous knowledge graph as the central node, the neighborhood aggregation vector is generated based on the weighted summation of the initial feature vectors of all neighboring nodes directly connected to the central node through edges;

[0022] An updated feature vector is generated by concatenating the initial feature vector of any central node with the corresponding neighborhood aggregation vector;

[0023] According to the updated feature vector of any first-category node and the updated feature vector of any second-category node, a scalar value representing the strength of the association between the two is calculated, and a target edge is determined based on the scalar value and a preset scalar threshold.

[0024] In one practicable implementation, by performing named entity recognition and relationship extraction on unstructured technical documents, a structured text information set is obtained, including:

[0025] Mark the component name entity, technical parameter entity, and process action entity identified from the design technical requirement information and process technical requirement information as technical entities, wherein the technical parameter entity includes a numerical parameter and a parameter unit, and the process action entity represents an assembly operation behavior;

[0026] Determine the scope of statements that include the same component name entity in the design technical requirement information and the process technical requirement information, and determine the parameter attribution relationship between the technical parameter entity and the component name entity, as well as the action relationship between the process action entity and the component name entity, to form a technical relationship between the technical entities;

[0027] Construct a structured text information collection based on technical entities and technical relationships.

[0028] In one practicable implementation, a heterogeneous knowledge graph is constructed based on a text information set and a 3D geometric model, including:

[0029] Constructing a first-class node with each technical entity in the text information set, and establishing connecting edges between the first-class nodes based on the technical relationships between the technical entities;

[0030] Constructing a second type of node with each component in the three-dimensional geometric model, and establishing connecting edges between the second type of nodes based on the assembly adjacency relationship between the components;

[0031] By calculating the similarity between the textual semantic information of the first-category nodes and the geometric attribute information of the second-category nodes, an alignment mapping relationship is established between the first-category nodes and the second-category nodes. The alignment mapping relationship is then used to associate the first-category nodes and the second-category nodes representing the same physical object, thereby completing the construction of the heterogeneous knowledge graph.

[0032] In one feasible implementation, a heterogeneous knowledge graph is updated using target edges, all first-type target nodes connected to second-type target nodes via target edges are determined in the updated heterogeneous knowledge graph, and technical entities corresponding to the first-type target nodes and technical relationships between the first-type target nodes are associated with components in the three-dimensional geometric model corresponding to the second-type target nodes, thereby generating an integrated digital model of a full-motion simulator, including:

[0033] Add target edges to the heterogeneous knowledge graph to form an updated heterogeneous knowledge graph, where each target edge connects a first-category node and a second-category node;

[0034] Taking any second-category node in the updated heterogeneous knowledge graph as the second-category target node, traverse each second-category target node in the updated heterogeneous knowledge graph, and determine all first-category nodes directly connected to the second-category target node through the target edge as the first-category target nodes;

[0035] Extracting the textual semantic information of the technical entity corresponding to each first-category target node and the technical relationship associated with the edges between the first-category target nodes to form a set of technical entities and a set of technical relationships associated with the second-category target nodes;

[0036] The textual semantic information of each technical entity in the technical entity set and each technical relationship in the technical relationship set are associated with the data structure of the parts in the three-dimensional geometric model corresponding to the second type of target node to complete the construction of the integrated digital model of the full-motion simulator.

[0037] In a second aspect, the present application provides a system for constructing an integrated digital model of a full-motion simulator, which is suitable for constructing a digital model based on unstructured heterogeneous data. The system includes:

[0038] An acquisition module is used to acquire a three-dimensional geometric model and unstructured technical documents of the full-motion simulator, wherein the unstructured technical documents include design technical requirement information and process technical requirement information, and the three-dimensional geometric model includes geometric attribute information;

[0039] The recognition module is used to obtain a structured text information set by performing named entity recognition and relationship extraction on unstructured technical documents. The text information set includes technical entities and technical relationships between technical entities. The technical entities include text semantic information;

[0040] A construction module is used to construct a heterogeneous knowledge graph based on the text information collection and the 3D geometric model. The heterogeneous knowledge graph includes first-class nodes representing technical entities and second-class nodes representing parts in the 3D geometric model, as well as edges between the first-class nodes and edges between the second-class nodes. The edges between the first-class nodes represent technical relationships between technical entities, and the edges between the second-class nodes represent geometric adjacency relationships between parts.

[0041] A determination module is used to perform link prediction using a graph neural network based on the edges in the heterogeneous knowledge graph, the textual semantic information associated with the first type of nodes, and the geometric attribute information associated with the second type of nodes, so as to determine the target edge for connecting the first type of nodes with the second type of nodes;

[0042] A generation module is used to update the heterogeneous knowledge graph using target edges, determine all first-type target nodes connected to the second-type target nodes through target edges in the updated heterogeneous knowledge graph, and associate the technical entities corresponding to the first-type target nodes and the technical relationships between the first-type target nodes with the parts in the three-dimensional geometric model corresponding to the second-type target nodes to generate an integrated digital model of the full-motion simulator.

[0043] In a third aspect, the present application provides an electronic device comprising: a processor, and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement a method for constructing an integrated digital model of a full-motion simulator as in any embodiment of the first aspect.

[0044] In a fourth aspect, the present application provides a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, a method for constructing an integrated digital model of a full-motion simulator as in any one of the embodiments of the first aspect is implemented.

[0045] The present application implements an integrated digital model construction method, system, device and computer storage medium for a full-motion simulator. By automatically performing semantic parsing and relationship extraction on massive heterogeneous unstructured technical documents, and building a preliminary heterogeneous knowledge graph on this basis, it replaces the time-consuming, labor-intensive and error-prone manual interpretation and manual association process in traditional methods, significantly improving the efficiency of model construction. Graph neural networks are then used to conduct deep learning on the existing structural and attribute information in the graph, which can infer and predict deep implicit associations between text and three-dimensional models that are difficult to discover manually. This automated reasoning and completion mechanism effectively solves the problem of low model knowledge association density caused by existing technologies due to reliance on surface information. Ultimately, by systematically binding complete information to the three-dimensional model, a digital model with more complete information and richer knowledge associations is generated. Therefore, the present application can improve the information integrity and internal association of the digital model while improving construction efficiency.

[0046] Furthermore, by analyzing the physical connection relationship between the components in the full-motion simulator, a system dependency network was constructed, and the system importance index of each component was calculated from it, thereby quantifying the different contributions of each component to the overall structural stability. By introducing the system importance index as a key additional information into the subsequent link prediction process, the reasoning calculation no longer treats all potential associations indiscriminately, but is able to give priority to and identify technical requirements related to more critical structural components. The reasoning method with weights and focus effectively solves the problem that the existing technology may miss key information due to the lack of global structural cognition. Therefore, the present application significantly improves the accuracy of the model's association of key information during the construction process, and improves the accuracy and reliability of information association of the digital model on key components. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0048] Figure 1 This is a flow chart of a method for constructing an integrated digital model of a full-motion simulator provided by one embodiment of the present application;

[0049] Figure 2This is a flowchart of a method for determining a connection edge between a first type of node and a second type of node provided by an embodiment of the present application;

[0050] Figure 3 This is a flow chart of a method for generating an integrated digital model of a full-motion simulator provided by one embodiment of the present application;

[0051] Figure 4 This is a schematic diagram of the structure of a system for building an integrated digital model of a full-motion simulator provided by one embodiment of the present application;

[0052] Figure 5 This is a schematic diagram of the hardware structure of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0053] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present application by illustrating the examples of the present application.

[0054] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, the elements defined by the phrase "comprising..." do not exclude the presence of other identical elements in the process, method, article, or device comprising the elements.

[0055] Existing technical methods have significant limitations when processing massive amounts of heterogeneous unstructured data. The process of manual interpretation and manual association of information is not only time-consuming and labor-intensive, but also prone to errors. In addition, it is often only possible to establish surface-level correspondences that are clearly recorded in the documents. Existing methods are unable to effectively identify and establish deep implicit associations that require contextual context or even comprehensive analysis across documents to discover, ultimately resulting in incomplete digital model information and low knowledge association density. Therefore, existing technical methods have technical problems such as low construction efficiency and insufficient model information correlation.

[0056] To solve the problems of the prior art, the present invention provides a method, system, device, and computer storage medium for constructing an integrated digital model of a full-motion simulator. The following first introduces the method for constructing an integrated digital model of a full-motion simulator provided by the present invention.

[0057] Figure 1 The figure shows a flow chart of a method for constructing an integrated digital model of a full-motion simulator provided by an embodiment of the present application. The method is suitable for constructing a digital model based on unstructured heterogeneous data, such as Figure 1 As shown, the method includes steps S110 to S150.

[0058] This application is applied to scenarios where a unified integrated digital model of a full-motion simulator is constructed based on data from diverse sources and formats that lack a unified predefined data model, such as text-based technical documents that lack a regular structure and structured three-dimensional geometric models. In such scenarios, text information carries abstract knowledge such as product design requirements and process flow, while the three-dimensional model accurately defines the physical form and spatial layout of the product. These two completely different but complementary non-unified structured data sources are effectively linked to construct an integrated full-motion simulator and realize digital management of the full life cycle of the full-motion simulator.

[0059] S110: Acquire a three-dimensional geometric model and unstructured technical documents of the full-motion simulator, where the unstructured technical documents include design technical requirement information and process technical requirement information, and the three-dimensional geometric model includes geometric attribute information.

[0060] A full-motion simulator is a complex mechatronic system used for flight or driving training that can simulate the motion posture of a vehicle. Unstructured technical documents are electronic files that do not have a regular data structure and are mainly described in natural language. They can be PDF documents or Word documents. Design technical requirement information refers to the description of product performance, functions and constraints, including the performance parameters of components, material specifications, tolerance ranges and interface definitions. Process technical requirement information refers to the description of product manufacturing, assembly and testing processes, including assembly sequence, tightening torque, tools used and inspection standards. A three-dimensional geometric model is a data file that digitally represents the physical shape, size and spatial position of a component. It can be a general format file such as STEP or IGES. Geometric attribute information refers to additional data other than the shape in a three-dimensional geometric model, including the name of the component, material code, material properties and assembly constraints that define the spatial relationship between components.

[0061] The system retrieves the required files by accessing the specified data source path. Based on pre-set file type filters, it automatically retrieves and obtains target files from the specified data source. For example, documents with the PDF and DOCX file extensions are classified as unstructured technical documents, and graphic files with the STEP and CATPart file extensions are classified as 3D geometric models. For unstructured technical documents, the system extracts the entire text content. For 3D geometric models, the system analyzes the internal geometric topology data and assembly structure tree, and extracts the metadata embedded in the model as part of the geometric attribute information.

[0062] For example, to build a digital model of a certain flight simulator, the project database is first retrieved through an API interface, including multiple files such as "Cockpit Instrument Panel Layout Specification.pdf" and "Primary Flight Stick Assembly Instructions.docx" as unstructured technical documents. "Cockpit Instrument Panel Layout Specification.pdf" contains textual descriptions of display performance indicators such as resolution and refresh rate, forming the design technical requirements. "Primary Flight Stick Assembly Instructions.docx" contains specific steps and torque requirements for installing the joystick base and connecting sensors, forming the process technical requirements. Simultaneously, the associated "Cockpit Assembly Model.step File" is retrieved as a 3D geometric model. By parsing the 3D geometric model file, the precise 3D shapes of components such as the joystick and base are obtained. The properties of each part are also extracted. For example, the material code for the joystick grip is Stick-Grip-001, and the material is ABS plastic. This data, along with assembly relationships between parts such as contact and alignment, constitutes the geometric property information. Ultimately, the contents of these documents and the 3D model file serve as the initial data for building an integrated digital model of the full-motion simulator.

[0063] S120: Obtain a structured text information set by performing named entity recognition and relationship extraction on the unstructured technical document. The text information set includes technical entities and technical relationships between technical entities. The technical entities include text semantic information.

[0064] Named Entity Recognition (NER) and Relation Extraction (RE) are two core tasks in Natural Language Processing (NLP). Named Entity Recognition (NER) locates and classifies predefined words or phrases with specific meanings from the textual content of unstructured technical documents. Relation Extraction (RE) further determines the semantic connections or functional associations between entities identified through the NER process. The final output of this step is the text information collection, a structured dataset containing machine-readable entity and relationship entries extracted from the original text. A technical entity is a specific word in the text that represents a key engineering concept, such as the name of a component, a performance indicator, or a process operation. A technical relationship is a logical or functional association between two or more technical entities, such as parameter affiliation or action action relationship. Textual semantic information refers to the actual meaning of a technical entity, including not only its literal string but also its conceptual connotation in a specific domain context.

[0065] First, a sequence labeling model is used in conjunction with a preset full-motion simulator domain dictionary to scan the text content of the acquired unstructured technical documents sentence by sentence to identify and label all technical entities that meet the predefined categories. Subsequently, to ensure the accuracy of the information, this process also includes a coreference resolution step to identify and parse pronouns or ambiguous phrases in the text, accurately linking them to the specific technical entities mentioned above. After completing the coreference resolution, relationship extraction is performed to determine whether there is a preset technical relationship type between the entities by analyzing the grammatical structure within the sentence and the co-occurrence pattern between the identified entities. Finally, all identified technical entities and the technical relationships between them are organized into a standardized list to form a structured text information set.

[0066] S130: Based on the text information set and the three-dimensional geometric model, a heterogeneous knowledge graph is constructed. The heterogeneous knowledge graph includes first-class nodes representing technical entities and second-class nodes representing parts in the three-dimensional geometric model, as well as edges between the first-class nodes and edges between the second-class nodes. The edges between the first-class nodes represent the technical relationships between technical entities, and the edges between the second-class nodes represent the geometric adjacency relationships between parts.

[0067] A heterogeneous knowledge graph is a graph data structure used to organize and represent different types of information. It contains multiple types of nodes and edges and is used to integrate information from different data sources into the same network. The first type of node is the node used in the heterogeneous knowledge graph to represent technical entities extracted from unstructured technical documents. Each first-type node carries the textual semantic information of a technical entity. The second type of node is the node used in the heterogeneous knowledge graph to represent physical components in three-dimensional geometric models. Each second-type node is associated with the geometric attribute information of the corresponding component. The edges between the first-type nodes represent the technical relationships between technical entities, and the edges between the second-type nodes represent the geometric adjacency relationships between components.

[0068] First, for each technical entity in the text information collection, a corresponding first-category node is created in the heterogeneous knowledge graph. For each technical relationship, an edge is established between the corresponding first-category nodes. Simultaneously, for each independent component in the 3D geometric model, a corresponding second-category node is created in the heterogeneous knowledge graph. By analyzing the model's assembly structure tree and geometric topology, it is determined whether the components are physically in contact or closely adjacent. If so, an edge is established between the corresponding second-category nodes to represent the geometric adjacency relationship. To fuse the two independent graph structures, an entity alignment process is performed. By calculating the string similarity between the textual semantic information of the first-category nodes and the geometric attribute information of the second-category nodes, two types of nodes that may represent the same physical object are preliminarily associated. For example, the string similarity between component names or material codes is used. Ultimately, a heterogeneous knowledge graph is formed, containing two types of nodes and two types of edges.

[0069] S140: Based on the edges in the heterogeneous knowledge graph, the textual semantic information associated with the first type of nodes, and the geometric attribute information associated with the second type of nodes, a graph neural network is used to perform link prediction to determine a link prediction for connecting the first type of nodes with the second type of nodes.

[0070] Graph neural networks are deep learning models specifically designed for processing graph-structured data. Their core concept is to learn feature representations of nodes by transferring and aggregating information between them. Link prediction is a typical application of graph neural networks. Its goal is to calculate the probability of a connection between any two unconnected nodes in a graph, based on the existing nodes and connections, and to predict the most likely missing connection. A target edge is a high-probability connection edge between a node in the first and a node in the second category that is determined, through the link prediction process, to be expected but currently missing.

[0071] First, an initial feature vector is generated for each first-category node and second-category node in the graph. The feature vectors of first-category nodes can be obtained by encoding their textual semantic information, while the feature vectors of second-category nodes can be obtained by encoding their geometric attribute information. Next, multiple rounds of information transfer and aggregation are performed. In each round, each node collects the feature vectors of all its neighboring nodes and updates its own feature vector using an aggregation function, so that the updated feature vector can incorporate the structural information of the neighborhood. After multiple rounds of iteration, each node has a final feature representation that contains rich contextual information. Finally, the possibility of a connection between any first-category node and any second-category node is evaluated by calculating the similarity or correlation score between the final feature representation of the first-category node and the final feature representation of the second-category node. The connection with the highest probability is selected as the target edge.

[0072] S150: Use target edges to update the heterogeneous knowledge graph, determine all first-type target nodes connected to second-type target nodes through target edges in the updated heterogeneous knowledge graph, and associate the technical entities corresponding to the first-type target nodes and the technical relationships between the first-type target nodes with the parts in the three-dimensional geometric model corresponding to the second-type target nodes to generate an integrated digital model of the full-motion simulator.

[0073] The integrated digital model of the full-motion simulator not only contains the precise geometric information of the components, but also deeply binds all related design requirements and process requirements information from unstructured technical documents to each component in the form of structured data, forming a digital twin with complete information and interactive query.

[0074] First, all target edges are added to the heterogeneous knowledge graph to form an updated knowledge graph with more complete information. To structurally attach this networked knowledge to the 3D model, each second-category node in the graph is traversed. For any second-category node, a graph traversal query is performed to find all first-category nodes directly or indirectly connected to it through the original edges and the newly added target edges. These first-category nodes and their technical relationships are then aggregated into an information set. Next, to facilitate management and access, this information set is structured and reorganized. For example, the information can be categorized into different predefined categories such as design, process, and maintenance based on its nature, forming a hierarchical attribute data structure. Finally, this hierarchical attribute data structure is attached as a whole to the component in the 3D geometric model corresponding to the current second-category node, completing the final mapping from graph knowledge to 3D model attributes.

[0075] This embodiment performs automated semantic parsing and relationship extraction on massive heterogeneous unstructured technical documents, and builds a preliminary heterogeneous knowledge graph on this basis, replacing the time-consuming, labor-intensive and error-prone manual interpretation and manual association steps in traditional methods, significantly improving the efficiency of model construction. Graph neural networks are then used to conduct deep learning on the existing structural and attribute information in the graph, which can infer and predict deep implicit associations between text and three-dimensional models that are difficult to discover manually. This automated reasoning and completion mechanism effectively solves the problem of low model knowledge association density caused by reliance on surface information in existing technologies. Ultimately, by systematically binding complete information to the three-dimensional model, a digital model with more complete information and richer knowledge associations is generated. Therefore, the present application can improve the information integrity and internal association of digital models while improving construction efficiency.

[0076] In one feasible embodiment, the method further includes: generating semantic vectors of the first type of nodes using a pre-trained language model based on text semantic information associated with the first type of nodes, wherein the pre-trained language model is trained based on text data related to the full-motion simulator field.

[0077] A pretrained language model is a deep learning model pre-trained on large-scale, unlabeled text data. Its purpose is to learn general linguistic patterns and knowledge. Further training on specialized text data from the full-motion simulator field allows it to master the terminology and semantic conventions within a specific domain. A semantic vector, generated by a pretrained language model, is a numerical vector that represents the underlying meaning of a word or phrase. Each dimension in the vector corresponds to a potential semantic feature, and words with similar semantics are positioned similarly in the vector space.

[0078] First, a large amount of text data related to the full-motion simulator field is collected. This data can include publicly available aviation standards, equipment manuals, design documents for historical projects, and relevant academic papers. This massive amount of text data is then fed into a general pre-trained language model for secondary training or fine-tuning to adapt it to the language style of the field. The language model can be a BERT model. After training, for each first-category node in the heterogeneous knowledge graph, its associated textual semantic information, namely the name of the technical entity, is fed into the pre-trained language model specifically for this field. The model then outputs a high-dimensional semantic vector that accurately captures the professional connotations of the technical entity.

[0079] Step S140: Based on the edges in the heterogeneous knowledge graph, the textual semantic information associated with the first type of nodes, and the geometric attribute information associated with the second type of nodes, a graph neural network is used to perform link prediction to determine a target edge for connecting the first type of nodes with the second type of nodes, including:

[0080] Step S141: Based on the edges in the heterogeneous knowledge graph, the semantic vectors of the first type of nodes, and the geometric attribute information associated with the second type of nodes, a graph neural network is used to perform link prediction to determine the target edge used to connect the first type of nodes with the second type of nodes.

[0081] The graph neural network uses the semantic vector corresponding to each first-category node as its initial feature vector. For second-category nodes, their initial feature vectors are still encoded by their geometric attribute information. Subsequently, the model performs multiple rounds of information transmission and aggregation. In each round, each node collects the feature vectors of its neighboring nodes and updates its own feature vector through an aggregation function. Since the initial feature vectors of the first-category nodes already embed rich domain knowledge, this high-quality semantic information can be effectively propagated and diffused throughout the graph network during the information transmission process. After multiple rounds of iterations, each node has a final feature representation that combines deep domain knowledge and graph structure information. Finally, by calculating the correlation score between the final feature representation of any first-category node and the final feature representation of any second-category node, the possibility of a connection between them is evaluated, and the connection with the highest probability is selected as the target edge.

[0082] For example, a pre-trained language model based on the Transformer architecture is used, which can be a BERT model. First, hundreds of professional text data such as design specifications, maintenance manuals and research reports related to aviation simulators are collected and used to fine-tune the domain adaptability of a general BERT model so that it can more accurately understand the professional terms in this field. After the training is completed, for each first-class node in the heterogeneous knowledge graph, such as the technical entity "torque", it is input into the fine-tuned domain BERT model. The model outputs a 768-dimensional numerical vector, which is the semantic vector of the "torque" node. In the vector space, it not only captures the literal meaning of "torque", but also contains its deep semantic associations closely related to concepts such as "application", "tightening" and "bolts" in engineering practice.

[0083] The graph neural network then uses the semantic vectors generated in the previous step as the initial features for the first-category nodes and encodes the geometric attribute information as the initial features for the second-category nodes. Through multiple rounds of information transmission and aggregation within the graph network, the features of each node incorporate the structural and semantic information of its neighborhood. Finally, based on the learned final node feature representation, the model calculates a high association score between the first-category node "torque" and the second-category node "M5 bolt." Based on this high score, the method determines the existence of a target edge between them, thereby discovering an implicit technical association that is critical to building a complete digital model.

[0084] In a feasible embodiment, the method also includes: constructing a system dependency network based on the second-class nodes and the edges between the second-class nodes in the heterogeneous knowledge graph; and obtaining a system importance index that characterizes the degree of influence of the second-class nodes on the overall connectivity of the system dependency network by calculating the topological centrality of each second-class node in the system dependency network.

[0085] The system dependency network is a subnetwork extracted from a heterogeneous knowledge graph to represent the dependencies of the full-motion simulator's physical structure. This network contains second-class nodes representing physical components and edges representing the geometric adjacency between them. Topological centrality is a set of metrics used in network science to measure node importance, quantifying its influence by analyzing its position and connection patterns within the network. The system importance index is a numerical value calculated based on topological centrality that characterizes the criticality of a component within the overall system structure. A component with a higher index has a wider impact on other components.

[0086] First, all second-category nodes and their interconnected edges are extracted from the heterogeneous knowledge graph to form a system dependency network. Subsequently, a network analysis algorithm, typically a betweenness centrality algorithm, is applied to calculate the topological centrality of each second-category node. The betweenness centrality algorithm measures the bridge role of a node by counting the number of times the shortest path between all pairs of nodes in the network passes through that node. After the calculation is complete, the betweenness centrality score of each node is normalized to obtain the final system importance index. This index is then attached as a new attribute to the corresponding second-category node in the heterogeneous knowledge graph.

[0087] Step S141: Based on the edges in the heterogeneous knowledge graph, the semantic vectors of the first-type nodes, and the geometric attribute information associated with the second-type nodes, a graph neural network is used to perform link prediction to determine the target edge for connecting the first-type nodes with the second-type nodes, including:

[0088] Step S142: Based on the edges in the heterogeneous knowledge graph, the semantic vectors of the first type of nodes, and the geometric attribute information and system importance index associated with the second type of nodes, a graph neural network is used to perform link prediction to determine the target edge used to connect the first type of nodes with the second type of nodes.

[0089] When generating initial feature vectors for second-category nodes, the graph neural network no longer relies solely on geometric attributes. Instead, it jointly encodes these geometric attributes with the node's system importance index. This results in initial feature vectors for second-category nodes representing key components carrying higher weights or more prominent features. During subsequent information transfer and aggregation, these features can have a greater impact on the neighborhood. Ultimately, when calculating link likelihood, more complete and richer associations are predicted for components with high system importance indices.

[0090] For example, a system dependency network consisting of second-category nodes such as "rocker base," "M5 bolt," "hydraulic pump," and "main power supply" and their physical connections is first extracted from the complete heterogeneous knowledge graph. The topological centrality of each node in this network is then calculated. The "hydraulic pump" node, which connects to multiple actuator subsystems, has high betweenness centrality and is therefore assigned a high system importance index. In contrast, the independent "cockpit interior panel" node, which connects only to a few structural components, has low centrality and is therefore assigned a low system importance index. Subsequently, when the graph neural network generates the initial feature vector for the "hydraulic pump" node, it encodes its geometric properties along with this high importance index. When the model infers which component the "hydraulic system maintenance requirements" technical entity should be associated with, the model is more inclined to establish a target edge between "hydraulic system maintenance requirements" and "hydraulic pump," as the "hydraulic pump" node's features signal its "system importance."

[0091] This embodiment constructs a system dependency network by analyzing the physical connection relationship between the components in the full-motion simulator, and calculates the system importance index of each component, thereby quantifying the different contributions of each component to the overall structural stability. By introducing the system importance index as a key additional information into the subsequent link prediction process, the reasoning calculation no longer treats all potential associations indiscriminately, but is able to give priority to and identify technical requirements related to more critical components in the structure. The reasoning method with weights and focus effectively solves the problem that the existing technology may miss key information due to the lack of global structural cognition. Therefore, the present application significantly improves the accuracy of the model's association of key information during the construction process, and improves the accuracy and reliability of information association of the digital model on key components.

[0092] Figure 2 A flow chart of a method for determining a connection edge between a first type of node and a second type of node provided by an embodiment of the present application is shown. As shown in the figure, the method includes steps S210 to S240.

[0093] In one feasible embodiment, step S142: performing link prediction using a graph neural network based on the edges in the heterogeneous knowledge graph, the semantic vectors of the first-category nodes, and the geometric attribute information and system importance index associated with the second-category nodes to determine a target edge for connecting the first-category nodes with the second-category nodes, includes:

[0094] S210: Using the semantic vector of the first type of node as the initial feature vector of the first type of node, and generating the initial feature vector of the second type of node by jointly encoding the geometric attribute information and the system importance index associated with the second type of node.

[0095] The initial eigenvector is a numerical vector that represents the initial state of each node in a heterogeneous knowledge graph before entering the information transmission and aggregation phase. Joint encoding is a data processing method that aims to integrate and transform information from multiple sources or different properties, such as geometric attributes and system importance indexes, into a numerical vector of uniform dimension.

[0096] For each first-category node, the semantic vector generated in the previous step is used as the initial feature vector of the node. For each second-category node, its associated geometric attribute information, such as a vector representing multiple numerical attributes such as component size, volume, and material, and the node's system importance index, are input into a preset encoding function. This encoding function can be a multi-layer perceptron (MLP) shallow neural network consisting of an input layer, one or more hidden layers, and an output layer, whose internal parameters are learned through model training. The input layer receives the concatenated geometric attribute vector and system importance index. The hidden layer transforms and extracts the input information through neurons with nonlinear activation functions, such as the ReLU function. The output layer maps the transformed features to a numerical vector with the same dimension as the initial feature vector of the first-category node. This vector is the initial feature vector of the second-category node.

[0097] S220: Taking any node in the heterogeneous knowledge graph as the central node, a neighborhood aggregation vector is generated based on the weighted summation of the initial feature vectors of all neighboring nodes directly connected to the central node through edges.

[0098] The neighborhood aggregation vector is a numerical vector generated by aggregating information about the direct neighboring nodes of a central node. It represents the characteristics of the local network environment in which the central node resides. Neighborhood information aggregation is performed with each node in the heterogeneous knowledge graph as the central node. For a selected central node, all neighboring nodes directly connected to the central node by edges are first identified through the adjacency relationships of the graph. The initial feature vectors of these neighboring nodes are then extracted, and a weighted sum operation is performed on these vectors. Weights are learnable parameters in the graph neural network model. During the model training phase, weights are automatically adjusted based on their contribution to the final link prediction task to reflect the differences in the influence of different neighbors on the central node. The final result of the weighted summation is the neighborhood aggregation vector for the central node.

[0099] S230: Generate an updated feature vector by concatenating the initial feature vector of any central node with the corresponding neighborhood aggregation vector.

[0100] The concatenation operation is a vector combination method that connects two or more vectors end to end to form a new vector with a longer dimension. For a selected central node, its initial feature vector and its calculated neighborhood aggregation vector are extracted. Then, a concatenation operation is performed to combine these two vectors into a new vector with the sum of their dimensions. To further enhance the expressiveness of features, this concatenated long vector can be transformed through a nonlinear activation function, such as a rectified linear unit (ReLU) function. The resulting vector is the updated feature vector of the central node, which contains both the original attributes of the node and information about its neighborhood structure.

[0101] S240: Calculate a scalar value representing the strength of the association between any first-category node and any second-category node based on the updated feature vector, and determine a target edge based on the scalar value and a preset scalar threshold.

[0102] The scalar value is a numerical value used to quantify the likelihood or strength of a potential association between a first-category node and a second-category node. The preset scalar threshold is a critical value determined based on historical data and is used to convert continuous association strength scalar values ​​into a binary connection decision, i.e., connect or not connect.

[0103] First, any first-category node and any second-category node in the heterogeneous knowledge graph are selected. Then, the updated feature vectors generated by these two nodes in step S230 are extracted. These two updated feature vectors are input into a similarity calculation function, which can be a shallow neural network classifier. The shallow neural network classifier can be a multi-layer perceptron (MLP) classifier consisting of an input layer, one or more hidden layers, and an output layer. The input layer of this classifier receives the concatenated two updated feature vectors, and the hidden layer interacts and fuses the features through nonlinear transformations. The output layer contains a neuron and, through an activation function such as a sigmoid function, outputs a single scalar value between 0 and 1. This scalar value is the final calculated result representing the strength of the association between the two nodes. Finally, this scalar value is compared with a preset scalar threshold. If the calculated scalar value is greater than the preset scalar threshold, a high-probability association is determined between the two nodes, and a target edge is determined between them.

[0104] For example, the semantic vector of the first-category node, "Hydraulic System Maintenance Requirements," is assigned as the initial feature vector. Meanwhile, for the second-category node, "Hydraulic Pump," its geometric attribute information vector extracted from the 3D model is concatenated with the high system importance index obtained in the previous step. This concatenated vector is then fed into a shallow multilayer perceptron neural network acting as a joint encoder. The network outputs a vector with the same dimension as the first-category node, which serves as the initial feature vector for the "Li-pad-" node.

[0105] Next, with the "Hydraulic Pump" node as the central node, the initial feature vectors of all its neighboring nodes, such as "Actuator A" and "Pipeline B," are aggregated and weighted together to generate a neighborhood aggregate vector for "Hydraulic Pump." The initial feature vector for "Hydraulic Pump" and its neighborhood aggregate vector are then concatenated. This concatenated long vector undergoes a nonlinear transformation using a rectified linear unit (ReLU) function to generate an updated feature vector for "Hydraulic Pump." This updated feature vector now incorporates both the highly important information about the "Hydraulic Pump" itself and the local network environment surrounding its connections to multiple actuators. Finally, the updated feature vector for the "Hydraulic System Maintenance Requirements" node is concatenated with the updated feature vector for the "Hydraulic Pump" node, and the result is input into a multi-layer perceptron classifier. The output layer of the classifier uses a sigmoid function to calculate a scalar value of 0.92, representing the strength of the association between the two nodes. Because this scalar value exceeded the preset scalar threshold of 0.8, it was finally determined that there was a target edge between "Hydraulic System Maintenance Requirements" and "Hydraulic Pump", thereby accurately connecting a high-order maintenance requirement with a key component at the core of the system.

[0106] In one possible implementation, step S120: performing named entity recognition and relationship extraction on unstructured technical documents to obtain a structured text information set includes:

[0107] The component name entity, technical parameter entity and process action entity identified from the design technical requirement information and process technical requirement information are marked as technical entities, where the technical parameter entity includes numerical parameters and parameter units, and the process action entity represents the assembly operation behavior.

[0108] A sequence labeling model fine-tuned on domain corpus is applied. This model predefines entity labels such as part names, technical parameters, and process actions. When the model processes the text content of unstructured technical documents sentence by sentence, it predicts the most likely entity label for each word or phrase in the sentence. For example, when the model encounters the word "joystick base," it labels it as a part name entity based on the patterns learned in the training data. When it encounters "5 Nm," it splits it and labels it as a numerical parameter and a parameter unit, which together form a technical parameter entity. When it encounters the verb "fix," it labels it as a process action entity. In this way, the original plain text is converted into a sequence with richly typed annotations.

[0109] The design technical requirement information and process technical requirement information shall determine the scope of statements that include the same component name entity, and determine the parameter attribution relationship between the technical parameter entity and the component name entity, as well as the action relationship between the process action entity and the component name entity, to form a technical relationship between the technical entities.

[0110] A statement scope is the smallest text unit containing one or more interrelated entities. It can be a complete sentence or an independent technical specification clause. A parameter attribution relationship clarifies to which component name entity a technical parameter entity belongs. An action action relationship clarifies which component name entities a process action entity acts on, as well as the active and passive relationships between them.

[0111] A rule matching component based on dependency grammar analysis can be applied. This component can be a processing pipeline consisting of a dependency grammar parser and a rule engine in series. First, for each sentence in which an entity is identified, the dependency grammar parser analyzes the grammatical structure of the sentence and generates a dependency grammar tree. The tree represents the dependency relationships between words in the sentence, such as subject-predicate relationships, verb-object relationships, and noun-modification relationships. Subsequently, the rule engine loads a set of predefined template rules for extracting technical relationships. These rules are matched based on the paths and labels in the dependency grammar tree. For example, a rule can be defined as: "If a technical parameter entity is directly connected to a part name entity through a noun-modification relationship, then a parameter-attribution relationship is established between the two entities." By applying a series of such rules, the component can accurately extract structured relationships between entities from complex sentence structures.

[0112] A structured text information collection is constructed using technical entities and technical relationships. All identified technical entities, their types, and the technical relationships established between them are stored as triples. A triple consists of a head entity, a relationship, and a tail entity. For example, a parameter ownership relationship can be represented as (part name entity, owns parameter, technical parameter entity), and an action action relationship can be represented as (process action entity, acts on, part name entity). All these triples together form a list, which is the final structured text information collection.

[0113] For example, when processing the sentence "Use M5 bolts to secure the joystick base and apply a torque of 5 Nm" in the "Main Flight Control Stick Assembly Instructions.docx" document, a sequence tagging model fine-tuned with domain data is first used to perform entity recognition on the sentence. The model identifies "M5 bolts" and "joystick base" as part name entities, "5 Nm" as a technical parameter entity (containing the value 5 and the unit Nm), and "secure" and "apply" as process action entities.

[0114] Next, a component consisting of a dependency parser and a rule engine is applied to extract relationships. The dependency parser analyzes the sentence structure and generates a dependency grammar tree. The rule engine then matches this tree. For example, it finds that the process action entity "fix" is grammatically the predicate of the sentence, and "joystick base" is its direct object. Therefore, based on pre-set rules, it establishes an action-action relationship (fix, act on, joystick base) between them. It also finds that the technical parameter entity "5 Nm" is closely associated with the verb "apply," and that the action of "applying torque" is contextually highly relevant to the tightening operation of "M5 bolts." Consequently, it establishes a parameter-attribution relationship (M5 bolt, required torque, 5 Nm). Finally, these entities and relationships extracted from the sentence are constructed into multiple triples and stored in a list, forming part of a structured text information collection. By executing this process on all documents, a complete structured text information collection is ultimately formed.

[0115] In one possible implementation, step S130: constructing a heterogeneous knowledge graph based on the text information set and the 3D geometric model includes:

[0116] A first-class node is constructed with each technical entity in the text information set, and connection edges are established between the first-class nodes based on the technical relationships between the technical entities.

[0117] The algorithm traverses all triples in the text information collection. For each triple's head and tail entities, it checks whether first-class nodes representing these two entities already exist in the graph. If not, it creates new first-class nodes and assigns them the textual semantic information extracted from the text as attributes. Then, based on the relationship type in the triple, it establishes an edge between the pair of first-class nodes representing the head and tail entities, labeling the edge with the relationship type. By processing all triples, a semantic network reflecting the entire textual knowledge is ultimately constructed.

[0118] A second type of node is constructed with each component in the 3D geometric model, and connection edges are established between the second type of nodes based on the assembly adjacency relationship between the components.

[0119] All parts are identified by parsing the assembly structure tree of the three-dimensional geometric model. For each part, a new second-class node is created in the heterogeneous knowledge graph, and the geometric attribute information extracted from the model, such as the part name, material code, and material, is used as the attribute of the node. Then, the assembly adjacency relationship between the parts is determined through geometric calculations. For example, the adjacency relationship between the two parts is determined by detecting whether there is coplanarity, coaxiality, or contact less than a preset distance between the three-dimensional entities of the two parts. If so, an edge is established between the corresponding two second-class nodes. In this way, a geometric network that reflects the physical assembly structure is constructed.

[0120] By calculating the similarity between the textual semantic information of the first-category nodes and the geometric attribute information of the second-category nodes, an alignment mapping relationship is established between the first-category nodes and the second-category nodes. The alignment mapping relationship is then used to associate the first-category nodes and the second-category nodes representing the same physical object, thereby completing the construction of the heterogeneous knowledge graph.

[0121] Traverse all first- and second-category nodes. For any pair of first- and second-category nodes, extract the textual semantic information (i.e., entity name) of the first-category node and the name or material code from the geometric attribute information of the second-category node. Then, apply a string similarity calculation function, which can be an edit distance algorithm or a Jaro-Winkler similarity algorithm, to calculate the similarity score of the two strings. When the calculated similarity score exceeds a preset alignment threshold, the two different types of nodes are considered to point to the same physical object. A special association edge is established between them, and this pairing relationship is recorded in the alignment mapping relationship. In this way, the two originally separate networks are effectively connected, ultimately completing the construction of the heterogeneous knowledge graph.

[0122] For example, first, for the text information set, for each technical entity, such as "stick base", "M5 bolt" and "5 Nm torque", a corresponding first-class node is created. Then, based on the extracted technical relationship, for example, an edge representing the "use" relationship is established between the "stick base" node and the "M5 bolt" node. At the same time, the "cockpit assembly model .step file" is parsed to create a corresponding second-class node for each physical component in the three-dimensional model, such as the stick base named "Stick_Base_Assembly" and the M5 bolt named "Bolt_M5x10", and their geometric attribute information is attached to it. Since these two components are directly assembled in contact in the three-dimensional model, an edge representing the geometric adjacency relationship is also established between the two second-class nodes. At this point, there is a semantic network composed of first-class nodes and a physical network composed of second-class nodes in the graph, and the two are not directly related.

[0123] To merge the two networks, a physical alignment step is performed, traversing all possible pairings of all first-category nodes with all second-category nodes. For example, when processing a first-category node "stick base" and a second-category node "stick_base_assembly," the textual name "stick base" of the first-category node and the geometric name "stick_base_assembly" of the second-category node, as well as the material code "PN-SB-001," are extracted. A string similarity algorithm, such as the edit distance algorithm, is applied to calculate the similarity between "stick base" and "stick_base_assembly." An alias mapping rule is also checked to associate the two. If, after calculation and rule matching, their similarity score exceeds the preset alignment threshold of 0.85, the two nodes are determined to refer to the same physical object. An alignment edge is then established between the two nodes, indicating "equivalent to," and this pairing is recorded in the final alignment mapping. By executing this process for all node pairs, a unified heterogeneous knowledge graph is finally formed, in which the technical requirements of the text file and the physical components of the 3D model are connected through the newly established aligned association edges.

[0124] Figure 3 A flow chart of a method for generating an integrated digital model of a full-motion simulator provided by an embodiment of the present application is shown. As shown in the figure, the method includes steps S310 to S340.

[0125] In one feasible embodiment, step S150: updating the heterogeneous knowledge graph using target edges, determining all first-type target nodes connected to second-type target nodes via target edges in the updated heterogeneous knowledge graph, and associating the technical entities corresponding to the first-type target nodes and the technical relationships between the first-type target nodes to the components in the three-dimensional geometric model corresponding to the second-type target nodes, thereby generating an integrated digital model of the full-motion simulator, including:

[0126] S310: Add target edges to the heterogeneous knowledge graph to form an updated heterogeneous knowledge graph, where each target edge connects a first-category node and a second-category node.

[0127] For each target edge in the set, a new connecting edge is created between the corresponding first-category node and the second-category node in the heterogeneous knowledge graph. To distinguish these inferred edges from the edges existing in the original data, a special type label, such as "inference association," can be attached to these new target edges.

[0128] S320: Take any second-class node in the updated heterogeneous knowledge graph as the second-class target node, traverse each second-class target node in the updated heterogeneous knowledge graph, and determine all first-class nodes directly connected to the second-class target node through the target edge as the first-class target nodes.

[0129] A loop is performed to traverse all second-category nodes in the updated heterogeneous knowledge graph. In each loop, the current second-category node is set as the second-category target node. Then, starting from this second-category target node, all outgoing and incoming edges are queried, and only those connections whose outgoing edges are target edges are selected. All first-category nodes connected to the other ends of these target edges are collected to form a list. This list is the set of first-category target nodes corresponding to the current second-category target node.

[0130] S330: Extracting textual semantic information of the technical entity corresponding to each first-category target node and the technical relationship associated with the edges between the first-category target nodes to form a set of technical entities and a set of technical relationships associated with the second-category target nodes.

[0131] First, the textual semantic information carried by each first-category target node is extracted to form a set of technical entities. Next, within the subgraph consisting of these first-category target nodes, all edges and their labels connecting them are extracted. These edges represent the inherent technical relationships between these technical entities, thus forming a set of technical relationships. Together, these two sets form a complete semantic information unit associated with the current second-category target node.

[0132] S340: Associating the textual semantic information of each technical entity in the technical entity set and each technical relationship in the technical relationship set to the data structure of the parts in the three-dimensional geometric model corresponding to the second type of target node, and completing the construction of the integrated digital model of the full-motion simulator.

[0133] The application programming interface API of the three-dimensional model can be called. For the current second-type target node, first locate the corresponding component in the three-dimensional geometric model through its identifier. Then, the previously formed set of technical entities and technical relationship sets are organized according to the preset hierarchical or key-value pair format. For example, a top-level attribute called "associated technical information" can be created, under which the sub-attributes "design requirements" and "process requirements" are further divided, and different technical entities and relationships are classified under corresponding sub-attributes. Finally, this organized data structure is written as a whole into the attribute data area of ​​the component through the API and saved. When all second-type nodes have completed this operation, an integrated digital model of the full-motion simulator is generated. The integrated digital model of the full-motion simulator can provide information support and guidance for the full-motion simulator in the entire process of design, manufacturing and installation.

[0134] For example, the target edge connecting the first-class node "torque" and the second-class node "M5 bolt" is first added to the heterogeneous knowledge graph, and the edge is assigned a type label of "inference association" to form an updated heterogeneous knowledge graph. Then, all second-class nodes in the graph are traversed. When the second-class node "M5 bolt" is traversed, it is set as the current second-class target node. An edge of type "inference association" starting from this node is found, and the first-class node "torque" connected to the other end of the edge is determined to be the corresponding first-class target node. The "torque" node and its associated information, such as the technical relationship ("torque", the value is "5 Nm"), are extracted, and this information is collectively constructed into a set of technical entities and technical relationships associated with the "M5 bolt" node.

[0135] Finally, the API of the 3D model is called to locate the specific "M5 bolt" component in the "cockpit assembly model .step file" through the identifier. Then, the technical entities and technical relationship sets are organized according to the preset key-value pair format and written into the attribute data structure of the component. For example, a field called "Associated Process Requirements" is created in its property panel, and "Torque: 5 Nm" is used as its value. By executing this process for all components and all inferred target edges, finally, when all second-category nodes have completed the information binding operation, a complete integrated digital model of the full-motion simulator is generated. This model can provide comprehensive data support for subsequent design verification, process planning and installation guidance.

[0136] Based on the same concept, the embodiment of the present application provides a full-motion simulator integrated digital model construction system. Figure 4 The integrated digital model construction system of the full-motion simulator provided in the embodiment of the present application is described in detail.

[0137] Figure 4 This is a structural block diagram of a full-motion simulator integrated digital model construction system shown in an embodiment of the present application.

[0138] like Figure 4 As shown, the system is suitable for building a digital model based on unstructured heterogeneous data, and the system includes:

[0139] An acquisition module 410 is configured to acquire a 3D geometric model and unstructured technical documents of the full-motion simulator, wherein the unstructured technical documents include design technical requirement information and process technical requirement information, and the 3D geometric model includes geometric attribute information;

[0140] Identification module 420, configured to perform named entity recognition and relationship extraction on unstructured technical documents to obtain a structured text information set, wherein the text information set includes technical entities and technical relationships between technical entities, and the technical entities include text semantic information;

[0141] A construction module 430 is configured to construct a heterogeneous knowledge graph based on the text information set and the 3D geometric model, wherein the heterogeneous knowledge graph includes first-type nodes representing technical entities and second-type nodes representing components in the 3D geometric model, as well as edges between the first-type nodes and edges between the second-type nodes, wherein the edges between the first-type nodes represent technical relationships between technical entities and the edges between the second-type nodes represent geometric adjacency relationships between components;

[0142] a determination module 440 for performing link prediction using a graph neural network based on edges in the heterogeneous knowledge graph, textual semantic information associated with the first type of nodes, and geometric attribute information associated with the second type of nodes, to determine a target edge for connecting the first type of nodes with the second type of nodes;

[0143] Generation module 450 is used to update the heterogeneous knowledge graph using target edges, determine all first-type target nodes connected to second-type target nodes through target edges in the updated heterogeneous knowledge graph, and associate the technical entities corresponding to the first-type target nodes and the technical relationships between the first-type target nodes with the parts in the three-dimensional geometric model corresponding to the second-type target nodes to generate an integrated digital model of the full-motion simulator.

[0144] In one embodiment, the determination module 440 is also used to generate semantic vectors of the first type of nodes based on the text semantic information associated with the first type of nodes using a pre-trained language model, where the pre-trained language model is trained based on text data related to the full-motion simulator field; and to perform link prediction using a graph neural network based on the edges in the heterogeneous knowledge graph, the semantic vectors of the first type of nodes, and the geometric attribute information associated with the second type of nodes to determine the target edge used to connect the first type of nodes with the second type of nodes.

[0145] In one embodiment, the determination module 440 is also used to construct a system dependency network based on the second-category nodes and the edges between the second-category nodes in the heterogeneous knowledge graph; by calculating the topological centrality of each second-category node in the system dependency network, a system importance index characterizing the degree of influence of the second-category nodes on the overall connectivity of the system dependency network is obtained; based on the edges in the heterogeneous knowledge graph, the semantic vectors of the first-category nodes, and the geometric attribute information and system importance index associated with the second-category nodes, a graph neural network is used to perform link prediction to determine the target edge used to connect the first-category nodes with the second-category nodes.

[0146] In one embodiment, the determination module 440 is specifically used to use the semantic vector of the first type of node as the initial feature vector of the first type of node, and generate the initial feature vector of the second type of node by jointly encoding the geometric attribute information and system importance index associated with the second type of node; taking any node in the heterogeneous knowledge graph as the central node, a neighborhood aggregation vector is generated based on the weighted summation of the initial feature vectors of all neighboring nodes directly connected to the central node through edges; an updated feature vector is generated by splicing the initial feature vector of any central node with the corresponding neighborhood aggregation vector; based on the updated feature vector of any first type node and the updated feature vector of any second type node, a scalar value representing the strength of the association between the two is calculated, and the target edge is determined based on the scalar value and a preset scalar threshold.

[0147] In one embodiment, the identification module 420 is specifically used to mark the component name entities, technical parameter entities and process action entities identified from the design technical requirement information and the process technical requirement information as technical entities, wherein the technical parameter entities include numerical parameters and parameter units, and the process action entities represent assembly operation behaviors; determine the statement scope including the same component name entity in the design technical requirement information and the process technical requirement information, and determine the parameter attribution relationship between the technical parameter entity and the component name entity and the action relationship between the process action entity and the component name entity to form a technical relationship between the technical entities; and construct a structured text information set based on the technical entities and the technical relationships.

[0148] In one embodiment, the construction module 430 is specifically used to construct a first type of node with each technical entity in the text information set, and establish connecting edges between the first type of nodes based on the technical relationship between the technical entities; construct a second type of node with each component in the three-dimensional geometric model, and establish connecting edges between the second type of nodes based on the assembly adjacency relationship between the components; establish an alignment mapping relationship between the first type of node and the second type of node by calculating the similarity between the text semantic information of the first type of node and the geometric attribute information of the second type of node, and use the alignment mapping relationship to associate the first type of node and the second type of node representing the same physical object to complete the construction of the heterogeneous knowledge graph.

[0149] In one embodiment, the generation module 450 is specifically used to add target edges in the heterogeneous knowledge graph to form an updated heterogeneous knowledge graph, where each target edge connects a first-class node and a second-class node; taking any second-class node in the updated heterogeneous knowledge graph as a second-class target node, traversing each second-class target node in the updated heterogeneous knowledge graph, and determining all first-class nodes directly connected to the second-class target node through the target edge as the first-class target node; extracting the textual semantic information of the technical entity corresponding to each first-class target node and the technical relationship associated with the edges between the first-class target nodes to form a set of technical entities and a set of technical relationships associated with the second-class target nodes; associating the textual semantic information of each technical entity in the technical entity set and each technical relationship in the technical relationship set to the data structure of the parts in the three-dimensional geometric model corresponding to the second-class target node to complete the construction of the integrated digital model of the full-motion simulator.

[0150] Figure 4 Each module in the system shown has the function of implementing Figures 1 to 3 The functions of each step in the embodiment can achieve the corresponding technical effects, which will not be described in detail here for the sake of brevity.

[0151] Figure 5 A schematic diagram of the hardware structure of an electronic device provided in one embodiment of the present application is shown.

[0152] The electronic device may include a processor 510 and a memory 520 storing computer program instructions.

[0153] Specifically, the processor 510 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0154] The memory 520 may include a large-capacity memory for data or instructions. By way of example and not limitation, the memory 520 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 520 may include removable or non-removable (or fixed) media. Where appropriate, the memory 520 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 520 is a non-volatile solid-state memory.

[0155] The memory may include read-only memory (ROM), random access memory (RAM), magnetic disk storage media devices, optical storage media devices, flash memory devices, electrical, optical, or other physical / tangible memory storage devices. Thus, typically, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to the first aspect of the present disclosure.

[0156] The processor 510 reads and executes computer program instructions stored in the memory 520 to implement any one of the methods for constructing an integrated digital model of a full-motion simulator in the above embodiments.

[0157] In one example, the electronic device may further include a communication interface 530 and a bus 540. Figure 5 As shown, the processor 510 , the memory 520 , and the communication interface 530 are connected via a bus 540 and communicate with each other.

[0158] The communication interface 530 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0159] Bus 540 includes hardware, software, or both, and couples the components of the online data traffic metering device to each other. By way of example, and not limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industrial Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industrial Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Area Network (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 540 may include one or more buses. Although the embodiments of the present application describe and illustrate specific buses, the present application contemplates any suitable bus or interconnect.

[0160] The electronic device can execute the method for constructing an integrated digital model of a full-motion simulator in the embodiment of the present application, thereby realizing the integration of Figures 1 to 3 The method for constructing an integrated digital model of a full-motion simulator is described.

[0161] In addition, in conjunction with the above-mentioned methods for constructing an integrated digital model of a full-motion simulator, embodiments of the present application may provide a computer-readable storage medium for implementation. The computer-readable storage medium stores computer program instructions; when executed by a processor, the computer program instructions implement any of the above-mentioned methods for constructing an integrated digital model of a full-motion simulator.

[0162] It should be understood that the present application is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted here. In the above embodiments, several specific steps are described and illustrated as examples. However, the method process of the present application is not limited to the specific steps described and illustrated. Those skilled in the art can make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present application.

[0163] The functional blocks shown in the block diagrams described above can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they may be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, and the like. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments may be stored in a machine-readable medium or transmitted via a data signal carried in a carrier wave over a transmission medium or communication link. "Machine-readable medium" may include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROMs, flash memory, erasable ROMs (EROMs), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, and the like. Code segments may be downloaded via a computer network such as the Internet or an intranet.

[0164] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps. In other words, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0165] Aspects of the present application have been described above with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present application. It should be understood that each block in the flowcharts and / or block diagrams, as well as combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine such that execution of these instructions by the processor of the computer or other programmable data processing device enables the implementation of the functions / actions specified in one or more blocks in the flowcharts and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It should also be understood that each block in the block diagrams and / or flowcharts, as well as combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware that performs the specified functions or actions, or by a combination of dedicated hardware and computer instructions.

[0166] The above description is only a specific embodiment of the present application. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the scope of protection of the present application.

Claims

1. A method for constructing an integrated digital model of a full-motion simulator, suitable for constructing a digital model based on unstructured heterogeneous data, characterized in that: include: Acquire a three-dimensional geometric model and unstructured technical documents of a full-motion simulator, wherein the unstructured technical documents include design technical requirement information and process technical requirement information, and the three-dimensional geometric model includes geometric attribute information; By performing named entity recognition and relationship extraction on the unstructured technical document, a structured text information set is obtained, wherein the text information set includes technical entities and technical relationships between the technical entities, and the technical entities include text semantic information; Based on the text information set and the three-dimensional geometric model, a heterogeneous knowledge graph is constructed, wherein the heterogeneous knowledge graph includes a first type of nodes representing the technical entities and a second type of nodes representing parts in the three-dimensional geometric model, as well as edges between the first type of nodes and edges between the second type of nodes, wherein the edges between the first type of nodes represent technical relationships between the technical entities and the edges between the second type of nodes represent geometric adjacency relationships between the parts; Based on the edges in the heterogeneous knowledge graph, the textual semantic information associated with the first type of nodes, and the geometric attribute information associated with the second type of nodes, a graph neural network is used to perform link prediction to determine a target edge for connecting the first type of nodes with the second type of nodes; The heterogeneous knowledge graph is updated using the target edge, and all first-type target nodes connected to the second-type target nodes through the target edge are determined in the updated heterogeneous knowledge graph, and the technical entities corresponding to the first-type target nodes and the technical relationships between the first-type target nodes are associated with the parts in the three-dimensional geometric model corresponding to the second-type target nodes to generate an integrated digital model of the full-motion simulator.

2. The method according to claim 1, characterized in that The method further comprises: Based on the textual semantic information associated with the first type of nodes, generating semantic vectors for the first type of nodes using a pre-trained language model, wherein the pre-trained language model is trained based on textual data related to the field of full-motion simulators; The link prediction using a graph neural network based on the edges in the heterogeneous knowledge graph, the textual semantic information associated with the first type of nodes, and the geometric attribute information associated with the second type of nodes to determine a target edge for connecting the first type of nodes with the second type of nodes includes: Based on the edges in the heterogeneous knowledge graph, the semantic vectors of the first type of nodes, and the geometric attribute information associated with the second type of nodes, a graph neural network is used to perform link prediction to determine the target edge used to connect the first type of nodes with the second type of nodes.

3. The method according to claim 2, characterized in that The method further comprises: Building a system dependency network based on the second-type nodes and the edges between the second-type nodes in the heterogeneous knowledge graph; By calculating the topological centrality of each second-type node in the system dependency network, a system importance index representing the degree of influence of the second-type node on the overall connectivity of the system dependency network is obtained; The link prediction using a graph neural network based on the edges in the heterogeneous knowledge graph, the semantic vectors of the first type of nodes, and the geometric attribute information associated with the second type of nodes to determine a target edge for connecting the first type of nodes with the second type of nodes includes: Based on the edges in the heterogeneous knowledge graph, the semantic vectors of the first type of nodes, and the geometric attribute information and system importance index associated with the second type of nodes, a graph neural network is used to perform link prediction to determine the target edge used to connect the first type of nodes with the second type of nodes.

4. The method according to claim 3, characterized in that The link prediction using a graph neural network based on the edges in the heterogeneous knowledge graph, the semantic vectors of the first-type nodes, and the geometric attribute information and system importance index associated with the second-type nodes to determine a target edge for connecting the first-type nodes with the second-type nodes includes: Using the semantic vector of the first type of node as the initial feature vector of the first type of node, and generating the initial feature vector of the second type of node by jointly encoding the geometric attribute information and the system importance index associated with the second type of node; Taking any node in the heterogeneous knowledge graph as a central node, performing weighted summation based on the initial feature vectors of all neighboring nodes directly connected to the central node through the edges to generate a neighborhood aggregation vector; Generate an updated feature vector by concatenating the initial feature vector of any central node with the corresponding neighborhood aggregation vector; According to the updated feature vector of any first-category node and the updated feature vector of any second-category node, a scalar value representing the strength of the association between the two is calculated, and the target edge is determined based on the scalar value and a preset scalar threshold.

5. The method according to claim 1, wherein The structured text information set is obtained by performing named entity recognition and relationship extraction on the unstructured technical document, including: Marking the component name entity, technical parameter entity, and process action entity identified from the design technical requirement information and the process technical requirement information as technical entities, wherein the technical parameter entity includes a numerical parameter and a parameter unit, and the process action entity represents an assembly operation behavior; Determining the range of statements including the same component name entity in the design technical requirement information and the process technical requirement information, and determining the parameter attribution relationship between the technical parameter entity and the component name entity, and the action relationship between the process action entity and the component name entity, to form a technical relationship between the technical entities; A structured text information set is constructed using the technical entities and the technical relationships.

6. The method according to claim 1, wherein The step of constructing a heterogeneous knowledge graph based on the text information set and the three-dimensional geometric model includes: Constructing the first type of node with each of the technical entities in the text information set, and establishing connection edges between the first type of nodes based on the technical relationship between the technical entities; Constructing the second type of node with each component in the three-dimensional geometric model, and establishing connecting edges between the second type of nodes based on the assembly adjacency relationship between the components; By calculating the similarity between the textual semantic information of the first type of nodes and the geometric attribute information of the second type of nodes, an alignment mapping relationship is established between the first type of nodes and the second type of nodes, and the first type of nodes and the second type of nodes representing the same physical object are associated using the alignment mapping relationship to complete the construction of the heterogeneous knowledge graph.

7. The method according to claim 1, characterized in that The method of updating the heterogeneous knowledge graph by using the target edge, determining all first-type target nodes connected to the second-type target nodes via the target edge in the updated heterogeneous knowledge graph, and associating the technical entities corresponding to the first-type target nodes and the technical relationships between the first-type target nodes to the components in the three-dimensional geometric model corresponding to the second-type target nodes, and generating an integrated digital model of a full-motion simulator, includes: Adding the target edge to the heterogeneous knowledge graph to form an updated heterogeneous knowledge graph, wherein each target edge connects a node of the first type and a node of the second type; Taking any second-category node in the updated heterogeneous knowledge graph as a second-category target node, traversing each second-category target node in the updated heterogeneous knowledge graph, and determining all first-category nodes directly connected to the second-category target node through the target edge as the first-category target nodes; Extracting textual semantic information of the technical entity corresponding to each first-category target node and the technical relationship associated with the edge between the first-category target nodes to form a set of technical entities and a set of technical relationships associated with the second-category target nodes; The textual semantic information of each technical entity in the technical entity set and each technical relationship in the technical relationship set are associated with the data structure of the parts in the three-dimensional geometric model corresponding to the second type of target node to complete the construction of the integrated digital model of the full-motion simulator.

8. A full-motion simulator integrated digital model construction system, suitable for constructing digital models based on unstructured heterogeneous data, characterized by: The system comprises: an acquisition module, configured to acquire a three-dimensional geometric model and unstructured technical documents of the full-motion simulator, wherein the unstructured technical documents include design technical requirement information and process technical requirement information, and the three-dimensional geometric model includes geometric attribute information; an identification module configured to obtain a structured text information set by performing named entity recognition and relationship extraction on the unstructured technical document, wherein the text information set includes technical entities and technical relationships between the technical entities, and the technical entities include text semantic information; a construction module for constructing a heterogeneous knowledge graph based on the text information set and the three-dimensional geometric model, wherein the heterogeneous knowledge graph includes a first type of nodes representing the technical entities and a second type of nodes representing parts in the three-dimensional geometric model, as well as edges between the first type of nodes and edges between the second type of nodes, wherein the edges between the first type of nodes represent technical relationships between the technical entities and the edges between the second type of nodes represent geometric adjacency relationships between the parts; a determination module, configured to perform link prediction using a graph neural network based on the edges in the heterogeneous knowledge graph, the textual semantic information associated with the first type of nodes, and the geometric attribute information associated with the second type of nodes, to determine a target edge for connecting the first type of nodes with the second type of nodes; A generation module is used to update the heterogeneous knowledge graph using the target edge, determine all first-type target nodes connected to the second-type target nodes through the target edge in the updated heterogeneous knowledge graph, and associate the technical entities corresponding to the first-type target nodes and the technical relationships between the first-type target nodes to the parts in the three-dimensional geometric model corresponding to the second-type target nodes, so as to generate an integrated digital model of the full-motion simulator.

9. An electronic device, characterized in that: The device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the method for constructing an integrated digital model of a full-motion simulator as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the method for constructing an integrated digital model of a full-motion simulator as described in any one of claims 1 to 7.

Images