Complex structure part-based knowledge graph construction method

Abstract

The invention provides a method for constructing a knowledge graph based on complex structural parts. According to the method, feature parameters are fused in a hidden layer through a neural network model, and a high-precision classification result is output to define a knowledge graph category; defining and standardizing a process knowledge concept from a data source to form a mode layer of a tree structure; extracting entities from technical documents and historical regulations in combination with a bottom-up method to construct a data layer; multi-dimensional mapping of a mode layer and a data layer is realized through a Neo4j graph database, and a visual knowledge graph containing a process knowledge relationship is generated; and finally, generating a process route of the complex structural part by using the semantic association network in the map. According to the method, the knowledge graph is constructed on the basis of the complex structural parts, scattered technical documents, expert experience and historical data are integrated, a unified knowledge network is formed, data islands are effectively broken, the structured level, relevance and reuse efficiency of complex structural part process knowledge are enhanced, and the machining efficiency of the complex structural parts is effectively improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of complex part knowledge graph construction, and particularly relates to a method for constructing a complex structure part knowledge graph. BACKGROUND

[0002] High-end equipment manufacturing field has put forward very high performance requirements for complex structure parts, and the complexity of the structure and the multi-disciplinary coupling characteristics make process design a core challenge. Complex structure parts need to withstand extreme temperature, pressure and radiation environment, and the manufacturing process involves precise regulation of parameters in multiple fields such as material science, fluid mechanics and thermodynamics, requiring process knowledge to have high systematicness and traceability. The existing process knowledge management system is dominated by document storage, resulting in massive heterogeneous data scattered in two-dimensional drawings, test reports, expert experience and other different carriers, forming high-redundancy unstructured information islands. Traditional database systems lack semantic association and reasoning ability, and cannot realize automatic integration and conflict detection of cross-field knowledge, which seriously restricts the iteration efficiency and quality stability of process design. This limitation is particularly prominent in the development process of new complex structure parts, and process personnel need to spend a lot of time reconstructing knowledge context from fragmented information, which is difficult to meet the urgent needs of rapid innovation and collaborative design under the background of intelligent manufacturing.

[0003] Therefore, in the face of the urgent need to improve the processing efficiency of complex structure parts, there is currently an urgent need for an efficient processing method based on a knowledge graph. This method realizes the systematic and structured management of complex structure part process knowledge, intelligently infers process routes and optimizes process parameters, and processes complex structure parts according to the complex structure part processing route and the optimized parameter scheme, ultimately significantly improving the processing efficiency of complex structure parts. SUMMARY

[0004] To solve the problem of dispersed and weakly associated complex structure part process knowledge, which cannot efficiently process complex structure parts, the present application proposes a method for constructing a complex structure part knowledge graph. This method takes a hybrid knowledge graph system with a "four-layer architecture-ontology-driven-dynamic update" as the core driving engine, realizes intelligent inference of process routes and precise collaborative optimization of process parameters, and significantly improves processing efficiency and quality.

[0005] A method for constructing a complex structure part knowledge graph, comprising the following steps:

[0006] Step S1, extract the key process features of the complex structure part, input the feature code into the neural network model, fuse the process parameters in the model hidden layer, output the process classification label through the neural network, and define the knowledge graph construction range through the process classification label to generate a sub-graph construction domain for each type of part;

[0007] Step S2, in the sub-graph construction domain, based on the historical machining process plan of the complex structure part process database, the process elements are refined by adopting the hierarchical tree structure, forming a four-layer knowledge expression system including part layer, process layer, process step layer and feature layer, and an ontology model including class, object, attribute and data attribute is generated by adopting the ontology modeling tool, and the mode layer design of the knowledge graph is completed in a top-down manner,

[0008] Step S3, based on the structure and rules defined in the mode layer, the ontology concepts and their attributes are extracted from the heterogeneous data sources, and the data layer is constructed in a bottom-up manner through knowledge extraction, fusion and relationship mapping,

[0009] Step S4, mapping the mode layer and the data layer into the Neo4j graph database to realize the instantiation storage of the knowledge graph; through the ontology extension of the mode layer and the entity update of the data layer, the dynamic evolution of the knowledge graph is supported,

[0010] Step S5, based on the process constraint chain, resource matching rule and parameter association network in the constructed knowledge graph, the machining process route of the complex structure part and the optimized parameter scheme are automatically generated, and the complex structure part is machined according to the machining process route of the complex structure part and the optimized parameter scheme.

[0011] In order to optimize the technical scheme, the further improvement scheme is as follows:

[0012] Step S1 specifically includes the following steps:

[0013] Step S11, extraction and coding of key process features of complex structure parts:

[0014] The key process features of the complex structure part are extracted, including the plane, curved surface, groove and hole of the part, and the features are numerically coded: the plane is mapped into normal vector and area parameters, the curved surface is converted into curvature matrix, and the groove and hole are coded into depth-width ratio and position coordinates. The numerically coded features are input into the input layer of the neural network, and the process attributes including material hardness and tolerance grade are simultaneously integrated into the complex structure part process attributes. The feature normalization processing is realized by tensor operation to provide structured input for hidden layer parameter fusion,

[0015] Step S12, intelligent classification and graph delimitation of process parameter fusion:

[0016] The process parameters including curvature, hole diameter and groove depth are fused in the hidden layer of the neural network, and the feature interaction calculation is realized through the nonlinear activation function and weight optimization; the output layer generates process classification labels through Softmax normalization, and the process classification labels include generating thin-walled parts, deep hole parts and multi-curvature special-shaped parts, and the knowledge graph construction range is delimited through the process classification labels.

[0017] Step S2 specifically comprises the following steps:

[0018] Step S21, hierarchical expression of machining process knowledge:

[0019] In the sub-graph construction domain, the historical machining process plan in the complex structure part process database is taken as the object, and the structured and unstructured data therein are summarized, including part information, machining method and process parameters. A complete part process plan contains a process route, and in each process, there are several steps, and in the step content, the machining information of one or more features is described. The progressive decomposition logic of process route-step-feature machining information is defined, and the four-layer association model of part-process-step-feature is established to ensure the completeness and traceability of process knowledge.

[0020] Step S22, ontology modeling construction mode layer:

[0021] The addition of class, object attribute and data attribute is realized through the modeling tool, the part ontology, process ontology, step ontology and feature ontology are defined by using OWL language, the inheritance relationship between classes, object attributes and data attributes are defined, the mode layer design of knowledge graph is completed, the ontology relationship constructed is reasoned and corrected by using Hermi reasoning machine, the logical conflicts are automatically corrected, and the accuracy of the semantic of mode layer is ensured.

[0022] Step S3 specifically comprises the following steps:

[0023] Step S31, extracting data source ontology concept to construct data layer:

[0024] The data layer is constructed according to the structure level and rules set in the mode layer, the ontology concept and its attributes are extracted from the data source, the mode layer instance is constructed, and the data layer of the knowledge graph is formed,

[0025] Step S32, knowledge extraction, knowledge fusion and ontology relationship establishment:

[0026] Based on the historical machining process data source of complex structure parts, the regular expression rule matching knowledge extraction method is adopted to realize the knowledge extraction of process knowledge entity and attribute; the similarity of entity names is calculated by using edit distance, the entities with similarity greater than the threshold value are clustered and fused, and the same entity naming in the dictionary is unified to achieve the effect of entity alignment to eliminate the ambiguity of concepts; the relationship table is established to connect the extracted entities, all ontology instance IDs in the instance table obtained by knowledge extraction are put out, and the instances with relationship are one-to-one corresponding by ID, and finally four relationship tables are generated: "feature has step", "process has step", "part has feature" and "part has process".

[0027] Step S4 specifically comprises the following steps:

[0028] Step S41, knowledge graph instantiation storage:

[0029] The mode layer and the data layer are respectively imported into the Neo4j graph database, the mode layer is combined with the data layer through mapping of classes, object attributes and data attributes, mapping of the mode layer ontology to Neo4j labels, attributes and relationships is realized through a Cypher query language,

[0030] Step S42, process knowledge graph update:

[0031] The entities, concepts, relationships, attributes and their types in the mode layer are subjected to adding, deleting and modifying operations, and the new entities, attributes and attribute values are remapped to the knowledge graph, so that the update of the data layer of the knowledge graph is realized.

[0032] Step S5 specifically comprises the following steps:

[0033] Step S51, constraint rule driven process route generation:

[0034] Based on the transitive object attribute of the process ontology and the class equivalence constraint of the resource ontology in the knowledge graph, the OWL inference machine is used to analyze the complex structure part ontology features and match the process step ontology template; the relationship chain of the graph mode layer is dynamically traversed to generate a process step sequence, the data layer instance is bound to the resource, and a complex structure part machining process route conforming to the semantic logic is output, so that the rapid reasoning of the machining route is realized,

[0035] Step S52, knowledge graph driven parameter decision:

[0036] Based on the historical optimal parameters inherited by the data attribute of the process step ontology in the knowledge graph, the SWRL rule of the mode layer is used to realize conflict detection and trigger the parameter degradation mechanism; the material attribute of the complex structure part ontology and the equipment constraint of the resource ontology are fused, and the parameter instance of the data layer of the knowledge graph is dynamically reconstructed, so that the optimization of the process parameters of the complex structure part is realized, and finally the complex structure part is machined according to the machining process route of the complex structure part and the optimized parameter scheme.

[0037] Compared with the prior art, the present application has the following significant advantages:

[0038] The application realizes fine expression and efficient reuse of process knowledge of complex structure parts by constructing hierarchical process knowledge graph. The four-layer architecture of part-process-step-feature is combined with ontology modeling technology to establish a process knowledge network containing semantic association, which significantly improves the automatic integration capability of cross-domain knowledge. The dynamic updating mechanism supports ontology evolution in the mode layer and real-time mapping of instances in the data layer, ensuring the timeliness and consistency of the knowledge base. Based on Hermi inference engine and Neo4j graph database, the process route and process parameters of complex structure parts are inferred based on knowledge graph, providing knowledge-driven decision support for process design, achieving intelligent optimization of process parameters and abnormal root cause tracing, and realizing efficient machining of complex structure parts. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 The design idea of the knowledge graph of the application is shown in the figure.

[0040] Figure 2 The implementation flowchart of the classification of complex structure parts of the application is shown in the figure.

[0041] Figure 3 The structure diagram of the knowledge storage path and method of the application is shown in the figure.

[0042] Figure 4 The knowledge graph instance diagram of the overall blisk part of the application is shown in the figure.

[0043] Figure 5 The flowchart of the process inference of the knowledge graph of the application is shown in the figure. DETAILED DESCRIPTION

[0044] The embodiments of the application will be further described in detail below with reference to the accompanying drawings.

[0045] As shown in the figure, Figure 1 The application proposes a method based on complex structure part knowledge graph construction. To clearly show the core mechanism and implementation effect of this method, representative complex structure parts such as overall blisk, casing, blade, turbine disc, etc. are selected for example. The significant feature of this method is that it relies on the constructed hierarchical knowledge graph system, and the core implementation steps include:

[0046] Step S1, process feature classification driven graph category implementation:

[0047] As shown in the figure, Figure 2As shown, the key process features of complex structure parts such as whole blade discs are extracted, including planes, curved surfaces, grooves, holes and other geometric elements, which are numerically coded to form a multi-dimensional feature vector input neural network model; in the hidden layer, the curvature, hole diameter, groove depth and other process parameters are fused, and a high confidence process classification label is output through nonlinear transformation. The classification result accurately defines the construction category of the knowledge graph, generates a vertical domain of each type of part, and realizes the aggregation and efficient reuse of process knowledge.

[0048] Step S1 includes:

[0049] Step S11, complex part process feature extraction and coding:

[0050] The core geometric features of complex structure parts such as whole blade discs are extracted, including planes, curved surfaces, grooves, holes and other geometric elements, which are respectively coded as normal vector-area parameter group, curvature matrix eigenvalue, depth-width ratio-position coordinate vector; fuse material hardness value and tolerance grade scalar, construct multi-dimensional input tensor, input full connection neural network layer after Z-score normalization to eliminate dimension difference; through linear transformation and ReLU activation function to generate standardized feature primitive, provide structured input for hidden layer process parameter fusion, significantly improve model convergence speed and classification accuracy.

[0051] According to the feature recognition and classification results, the parts are roughly divided into four categories: "blade class", "whole blade class", "casing class" and "turbine disc class", and then according to the characteristics of different structure features in the same large category of parts, continue to be refined into multiple subcategories. Among them, "blade class" is divided according to whether it has cooling holes, whether it contains rim plate boss, and mortise type (dove / fir tree); "whole blade class" is divided according to whether the flow channel contains a sieve, whether the disc core contains a hole, and whether the blade disc combination area contains a boss; "casing class" is divided according to whether it contains a mounting edge, whether the thin wall contains a reinforcing rib, and whether the inner cavity contains a groove; "turbine disc class" is divided according to whether the disc core is perforated, whether the web is asymmetric, and whether the outer rim contains a mortise.

[0052] Step S12, intelligent classification and graph delimitation of process parameter fusion:

[0053] Based on the feature primitive in the neural network hidden layer, the curvature, hole diameter, groove depth and other core process parameters are fused; through ReLU nonlinear activation function and gradient back propagation to optimize the weight matrix to realize feature interaction calculation; the output layer is normalized by Softmax function to generate high confidence process classification label including thin-walled parts, deep hole parts, multi-curvature special-shaped parts, etc.; the classification result accurately defines the construction category of the knowledge graph and drives the generation of class-specific sub-graph domain.

[0054] Step S2, building of complex structure part knowledge graph mode layer:

[0055] Based on the historical machining process plan of the complex structure part process database, the process elements are refined using hierarchical tree structure, forming a four-layer knowledge expression system including part layer, process layer, step layer and feature layer. And the ontology model containing class, object, attribute and data attribute is generated using ontology modeling tool, and the knowledge graph mode layer design is completed in a top-down manner.

[0056] Step S2 includes:

[0057] Step S21, hierarchical expression of machining process knowledge:

[0058] The historical machining process plan in the complex structure part process database is taken as the object, and the structured and unstructured data such as part information, machining method and process parameters are summarized.

[0059] A complete part process plan contains a process route, contains several steps in each process, and describes the machining information of one or more features in the step content. The progressive decomposition logic of process route-step-feature machining information is defined, the four-layer association model of part-process-step-feature is established, and the integrity and traceability of process knowledge are ensured.

[0060] The part layer expresses the management information of part design and manufacturing, including part basic information, material information and machining information. The part basic information is the description of each part instance, which is convenient for searching and reusing.

[0061] The process layer expresses the process sequence of part machining and method. The process knowledge of process layer mainly includes process ID, process code, process name and some tooling information used in the process.

[0062] The step layer is the description of each machining action in a process. The process knowledge of step layer includes step ID, step code, machining allowance and dimensional tolerance and the description of the specific content of the machining process.

[0063] The process knowledge of feature layer includes feature ID, feature code, feature name and the description of each feature size parameter.

[0064] Step S22, build mode layer using ontology modeling:

[0065] Ontology is a structured framework for describing concepts and their relationships in a specific field. It provides a unified and unique semantic basis for knowledge sharing in the shared range by clearly defining core elements such as classes, attributes, relationships, instances and axioms.

[0066] The four elements of complex part structure are modeled by ontology, forming the ontology model of class, object attribute and data attribute, including part ontology, process ontology, step ontology and feature ontology.

[0067] Part ontology: According to the structural characteristics and application scenarios of parts, complex structural parts are divided into various sub-classes.

[0068] Process ontology: Each part ontology has a processing process sequence, mainly including milling, pliers, insertion, boring, turning, planing, grinding, and is refined into various sub-classes according to different processing methods.

[0069] Step ontology: Each process contains multiple processing steps, including "tapping", "tool setting" and other steps; and processing structural features, such as "drilling", "milling plane", "cylindrical grinding" and other steps.

[0070] Feature ontology: According to the different positions of features in parts, it can be classified as "front end", "rear end", "left end", "right end", "upper surface", "lower surface" and other features; and according to the position of the structural feature, including "boss", "hole", "U-shaped groove" and other features.

[0071] The ontology modeling tool Protégé is used to model the process knowledge ontology according to the schema layer architecture, and the class, object attribute and data attribute are added; then the Hermi inference engine is used to infer and correct the ontology relationship, and saved as an OWL file; the visual plug-in OntoGraf is used to visualize the semantic relationship of ontology.

[0072] Through steps S21-S22, the knowledge graph schema layer is built, and the part process plan is summarized into four layers: the part layer stores basic information and material specifications, the process layer defines the processing flow, the step layer refines the operation parameters, and the feature layer describes the geometric properties. The hierarchical model strengthens the process integrity, logical correlation and decision-making reasoning efficiency through structured expression.

[0073] Step S3, building of complex structural part knowledge graph data layer:

[0074] Based on the structure level and rules set by the schema layer, the ontology concepts and their attributes are extracted from the process knowledge data source, the constructed schema layer is instantiated, and the knowledge graph data layer is built from bottom to top through knowledge extraction, knowledge fusion and ontology relationship establishment.

[0075] Step S3 includes:

[0076] Step S31, extract data source ontology concept to build data layer:

[0077] The knowledge graph data layer uses a relational database to store a large number of complex structure part machining instances. The MySQL database is used to store the machining instances and relationships of the complex structure parts. The data layer is constructed according to the structure level and rules set by the schema layer, and the ontology concepts and their attributes are extracted from the data source, so that the constructed schema layer is instantiated.

[0078] Step S32, knowledge extraction, knowledge fusion and ontology relationship establishment:

[0079] The construction of the typical complex structure part machining process knowledge data layer includes three parts of knowledge extraction, knowledge fusion and ontology relationship establishment.

[0080] Knowledge extraction is one of the core technologies of knowledge graph construction, which extracts triples from different data sources and different structure data and stores them in the knowledge graph. The knowledge extraction method of regular expression rule matching is used in the present application. The feature size parameter is extracted with "guarantee size" as the extraction rule starting point and period as the extraction end mark, and the regular expression is "(?< = guarantee size). * (? =.)".

[0081] The structured data in the process file includes entities and related data attributes of parts, processes and steps, which are extracted by using the PDFplumber library to generate instance tables and stored in the MySQL database.

[0082] Through knowledge fusion, the repeated and incorrect information in the extraction result is cleaned and integrated, different entity semantics in the data source are linked to the same entity, the entity semantic disambiguation and coreference resolution are completed, and the entity is uniformly named.

[0083] For the extracted entities, the similarity of the entity names is calculated by using the edit distance, the entities with a similarity greater than a threshold value are clustered and fused, and the same entity name in the dictionary is unified, so that the effect of entity alignment is achieved. The edit distance refers to the minimum number of editing operations required to change one string into another, and the editing operations include replacement, insertion and deletion. The fusion result is saved in the MySQL database.

[0084] The extracted entities are connected by establishing a relationship table, all ontology instance IDs in the instance table obtained by knowledge extraction are extracted, the instances with relationships are one-to-one corresponding by ID, and the following four relationship tables are generated: "feature has step", "process has step", "part has feature" and "part has process", as shown in the following table. Figure 2

[0085] ​Through the steps of S31-S32, knowledge extraction, knowledge fusion and ontology relationship establishment are performed on the data source by a bottom-up method, and entities and information are matched and filled into the schema layer, so that the construction of the knowledge graph data layer is completed.

[0086] Step S4, storage and update of the complex structure part knowledge graph:

[0087] Based on the constructed complex structure part knowledge graph, the schema layer and the data layer are imported into the graph database, and the combination of the class, object attribute and data attribute is realized through the mapping, so that the instantiation of the knowledge graph is realized; the update of the schema layer and the data layer can be realized to realize the update of the knowledge base.

[0088] Step S4 includes:

[0089] Step S41, knowledge graph instantiation storage:

[0090] The knowledge storage of the present application mainly adopts the method of combining MySQL relational database and Neo4j graph database: the MySQL database is used as a medium for storing data, and various ontology instances are mainly stored; the Neo4j graph database is used as the center, and all triples are visualized and displayed in the Neo4j graph database.

[0091] The storage method is as shown in Figure 3 The constructed schema layer is exported in the form of an OWL file, and is imported into the Neo4j database by using the Neosemanticsjar tool; the connection and data transmission between the Neo4j and the MySQL database are realized through the Apoc plug-in in the Neo4j and the MySQL JDBC driver; various entity nodes, entity relationships and attributes are established in the Neo4j. Taking the relationship of 'part has feature' as an example, the following is the Cypher statement for importing the data in the MySQL database into the Neo4j graph database to generate triples:

[0092] call apoc.load.jdbc('jdbc:mysql: / / localhost: / complex structure part?user=root&password=root&useUnicode=true&characterEncoding=utf8','select*from part has feature')yield row

[0093] match(from:part instance{part id:row.part id}),(to:feature instance{feature id:row.feature id})

[0094] merge(from)-[r:has feature]->(to)

[0095] The mode layer and the data layer are respectively imported into the Neo4j graph database, the mode layer is combined with the data layer through the mapping of classes, object attributes and data attributes, and the instantiation of the knowledge graph is realized.

[0096] Based on the machining process characteristics of the typical parts with complex structure, the Cypher query language is used to construct a structured knowledge graph instance as shown in Figure 4 The graph realizes the explicit expression of process knowledge through three-layer architecture of feature layer, process step layer and process layer: the feature layer accurately defines the typical features covering the key functional structures of the blisk, such as flow passage, grid, disc core hole, etc.; the process step layer establishes the mapping relationship between features and machining chains: the flow passage feature is associated with the five-axis rough milling, semi-finishing milling and finishing milling process step sequence, the grid feature is bound to the turning, rough grinding and fine grinding compound process step, and the disc core hole feature is linked to deep hole drilling and reaming; the process layer integrates related process steps to form manufacturing units such as milling, turning and grinding. This architecture realizes the structured integration of process knowledge from feature definition to process execution, provides an example for process intelligent decision-making based on knowledge graph, and realizes efficient machining of complex structure parts.

[0097] Step S42, updating the process knowledge graph:

[0098] After the construction of the knowledge graph system, the knowledge base needs to be updated regularly. From a logical point of view, the update of the knowledge base is realized by updating the two levels of the mode layer and the data layer.

[0099] The entities, concepts, relationships, attributes and their types in the mode layer are added, deleted or modified. The updated results are re-imported into the Neo4j graph database to generate new nodes. The new entities, attributes and attribute values are remapped to the knowledge graph. For a small amount of data update, Cypher statements can be used in the Neo4j graph database to add, delete and modify instances, attributes and relationships.

[0100] For batch data update, the results of knowledge extraction are generated in CSV file format using Python, and then Cypher statements are used to directly import the triples into the Neo4j graph database and update them in various nodes. The new instances and relationships obtained by knowledge extraction are saved in CSV file format and placed in the import folder under the installation directory of Neo4j, and then Cypher statements are input into the Neo4j input box.

[0101] Step S5, process knowledge graph driven process intelligent reasoning:

[0102] As shown in Figure 5As shown, based on the process constraint chain, resource matching rule and parameter association network in the constructed knowledge graph, the machining process route and parameter scheme of the complex structure part are automatically generated. The semantic association of the graph provides the core basis for process step sequence optimization, equipment selection and cutting parameter decision, realizing the transition of process design from experience dependence to data driving.

[0103] Step S5 includes:

[0104] Step S51, constraint rule driven process route generation:

[0105] By analyzing the transitive object properties of the process ontology through the OWL inference machine, the topological dependency chain between the steps is constructed; based on the class equivalence constraint rules of the resource ontology, the equipment resource binding is triggered; the feature vector of the part ontology is matched with the cosine similarity of the step ontology template to locate the optimal process template; the step sequence is dynamically generated by traversing the mode layer semantic relationship chain, such as drilling, reaming and reaming, and finally the executable process route is output.

[0106] Step S52, knowledge graph driven parameter decision:

[0107] Analyzing the data attribute inheritance history optimal parameter value of the step ontology, loading the pre-stored process parameter database; detecting process conflicts in real time through mode layer SWRL rules, including speed limit detection and material hardness mutation identification; triggering parameter degradation mechanism to execute dynamic threshold adjustment; using particle swarm optimization algorithm to solve the optimal parameter combination with the maximum material removal rate and the minimum tool wear as the multi-objective function; combining the physical constraint attributes of the bound equipment instances, including the maximum torque threshold and power upper limit value of the spindle, dynamically reconstructing the cutting speed and feed rate parameter chain; after conflict resolution and boundary correction, the executable process parameter scheme is output, effectively improving the machining efficiency of new complex parts.

[0108] In summary, the method for constructing a knowledge graph based on a complex structure part of the application has the following beneficial effects:

[0109] 1. Hierarchical structure improves process design efficiency and accuracy: Through the four-layer architecture of part layer, process layer, step layer and feature layer, complex process knowledge is decomposed into modular and hierarchical structure. This hierarchical method makes the process design process more clear, facilitating quick positioning and adjusting parameters of specific links, thereby improving design efficiency. Using ontology modeling tools and inference machines, the consistency of process rules can be automatically verified, reducing human errors.

[0110] 2. Unified semantic framework enhances knowledge sharing and reuse: The ontology provides a unified semantic foundation for the process knowledge of complex structural parts, ensuring consistent understanding of concepts such as "part", "process", "step", and "feature" by different systems. This eliminates term ambiguity and promotes cross-team knowledge sharing. The visualization of the knowledge graph is achieved through the graph database Neo4j, which intuitively presents the associations between process knowledge. For example, through the "part has feature" and "process has step" relationship tables, users can quickly query all the machining features of a part and their corresponding processes, supporting cross-domain knowledge integration and reuse.

[0111] 3. Support process design recommendation and optimization: Based on the large number of machining examples stored in the knowledge graph, process parameter optimization and machining route recommendation can be performed. By analyzing the historical machining data of similar parts, the recommended machining sequence and parameter combination can be recommended, which can shorten the process design cycle and significantly improve the machining efficiency of complex structural parts. By regularly updating the pattern layer and data layer, the knowledge graph can continuously learn new processes and methods. Only by updating the ontology model and relationship table can new technologies be integrated into the existing knowledge graph to meet the needs of technology iteration.

[0112] The above is only the preferred embodiment of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solutions falling within the scope of the present application should be considered within the protection scope of the present application. It should be noted that for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the present application should be considered within the protection scope of the present application.

Claims

1. A method for constructing a knowledge graph based on a complex structure part, characterized in that, The method comprises the following steps: Step S1, extracting key process features of complex structure parts, inputting the features into a neural network model through feature coding, fusing process parameters in the hidden layer of the model, outputting process classification labels through the neural network, and determining the knowledge graph construction range through the process classification labels to generate a sub-graph construction domain for each type of part; Step S2, in the sub-graph construction domain, based on the historical machining process procedures of the complex structure part process database, the process elements are refined using a hierarchical tree structure to form a four-layer knowledge expression system including part layer, process layer, process step layer and feature layer, and an ontology modeling tool is used to generate an ontology model including class, object, attribute and data attribute, and the mode layer design of the knowledge graph is completed in a top-down manner, Step S3, based on the structure and rules defined in the mode layer, the ontology concepts and their attributes are extracted from the heterogeneous data sources, and the data layer is constructed from bottom to top through knowledge extraction, fusion and relationship mapping, Step S4, mapping the mode layer and the data layer into a Neo4j graph database to realize the instantiation storage of the knowledge graph; through ontology expansion of the mode layer and entity update of the data layer, the dynamic evolution of the knowledge graph is supported, Step S5, based on the process constraint chain, resource matching rules and parameter association network in the constructed knowledge graph, the machining process route of the complex structure part and the optimized parameter scheme are automatically generated, and the complex structure part is machined according to the machining process route of the complex structure part and the optimized parameter scheme.

2. The method for constructing a knowledge graph based on complex structure parts according to claim 1, characterized in that, Step S1 specifically comprises the following steps: Step S11, key process feature extraction and coding of complex structure parts: Extract key process features of complex structure parts, including planes, curved surfaces, grooves and holes. Numerical coding is performed on the features: plane is mapped to normal vector and area parameter, curved surface is converted to curvature matrix, and groove and hole are coded as depth-width ratio and position coordinates. The numerically coded features are input into the input layer of the neural network, and the complex structure part process attributes are simultaneously integrated, including material hardness and tolerance grade. Feature normalization is realized through tensor operation to provide structured input for hidden layer parameter fusion, Step S12, intelligent classification and graph delimitation of process parameter fusion: In the hidden layer of the neural network, the process parameters of the complex structure part are fused, including curvature, hole diameter and groove depth. Feature interaction calculation is realized through nonlinear activation function and weight optimization; the output layer generates process classification labels through Softmax normalization, including generating thin-walled parts, deep hole parts and multi-curvature special-shaped parts. The process classification labels are used to determine the knowledge graph construction range.

3. The method of claim 1, wherein, Step S2 specifically comprises the following steps: Step S21, hierarchical expression of machining process knowledge: In the sub-graph construction domain, the historical machining process plan in the complex structure part process database is taken as the object, the structured and unstructured data in the historical machining process plan are summarized, the structured and unstructured data include part information, machining method and process parameter, a complete part process plan contains a process route, in each process, a plurality of working steps are contained, in the working step content, the machining information of one or more features is described, the progressive decomposition logic of process route-working step-feature machining information is defined, the four-layer association model of part-process-working step-feature is established, the integrity and traceability of process knowledge are ensured, Step S22, the ontology modeling construction mode layer: Through the modeling tool, the addition of class, object attribute and data attribute is realized, the part ontology, process ontology, working step ontology and feature ontology are defined by using the OWL language, the inheritance relationship between classes, object attributes and data attributes are defined, the mode layer design of the knowledge graph is completed, the ontology relationship constructed is reasoned and corrected by using the Hermi reasoning machine, the logical conflicts are automatically corrected, and the accuracy of the mode layer semantics is ensured.

4. The method for constructing a knowledge graph based on complex structure parts according to claim 1, characterized in that, Step S3 specifically includes the following steps: Step S31, extracting the data source ontology concept to construct the data layer: The data layer is constructed according to the structure level and rules set in the mode layer, the ontology concept and its attributes are extracted from the data source, the mode layer instance is constructed, and the data layer of the knowledge graph is formed, Step S32, knowledge extraction, knowledge fusion and ontology relationship establishment: Based on the historical machining process data source of complex structure parts, the regular expression rule matching knowledge extraction method is adopted to realize the knowledge extraction of process knowledge entities and attributes; the similarity of entity names is calculated by using the edit distance, the entities with a similarity greater than a threshold value are clustered and fused, and are unified into the same entity naming in the dictionary to achieve the effect of entity alignment to eliminate the ambiguity of concepts; the established relationship table is used to connect the extracted entities, all ontology instance IDs in the instance table obtained by knowledge extraction are put out, the instances with relationships are one-to-one corresponding by using the ID, and finally four relationship tables are generated: "feature has working step", "process has working step", "part has feature" and "part has process".

5. The method of claim 1, wherein, Step S4 specifically includes the following steps: Step S41, knowledge graph instantiation storage: The mode layer and the data layer are respectively imported into the Neo4j graph database, the mode layer and the data layer are combined by mapping the class, object attribute and data attribute, the mapping of the mode layer ontology to the Neo4j label, attribute and relationship is realized by using the Cypher query language, Step S42, process knowledge graph update: The entities, concepts, relationships, attributes and their types in the mode layer are added, deleted and modified, the new entities, attributes and attribute values are remapped to the knowledge graph, and the update of the data layer of the knowledge graph is realized.

6. The method for constructing a knowledge graph based on complex structure parts according to claim 1, characterized in that, Step S5 specifically includes the following steps: Step S51, constraint rule driven process route generation: Based on the inheritance of the object properties of the process ontology and the class equivalence constraints of the resource ontology in the knowledge graph, the complex structure part ontology features are parsed and the process ontology template is matched through the OWL inference machine; the process sequence is generated by dynamically traversing the relationship chain of the graph pattern layer, the resource is bound by triggering the data layer instance, and the complex structure part processing route conforming to the semantic logic is output, realizing the rapid reasoning of the processing route, Step S52, parameter decision driven by knowledge graph: Based on the inheritance of the process ontology data properties in the knowledge graph, the conflict detection is realized through the SWRL rules of the pattern layer and the parameter degradation mechanism is triggered; the complex structure part ontology material properties and the resource ontology equipment constraints are fused, the knowledge graph data layer parameter instance is dynamically reconstructed, and finally the optimization of the complex structure part process parameters is realized, and finally the complex structure part is processed according to the complex structure part processing route and the optimized parameter scheme.

Images