Hole making process parameter recommendation method based on knowledge graph

Abstract

The invention discloses a hole making process parameter recommendation method based on a knowledge graph, which uses an optimized seven-step method, and provides an ontology evaluation system and an ontology updating method, so that a hole making process ontology can be updated and iterated according to different processing requirements; according to the method, the fine-tuning UIE model is adopted to perform knowledge extraction on the hole-making process text data, good extraction accuracy can be obtained only by performing fine-tuning on the UIE model by using small sample data, the marking cost of knowledge extraction of the hole-making process text data is reduced, and the extraction efficiency is improved; according to the method, an entity alignment algorithm and a processing case filtering algorithm based on text similarity and word list mapping are designed, so that the repetition rate of knowledge in the knowledge graph is reduced, and the retrieval efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of hole making process technology, and in particular to a method for recommending hole making process parameters based on a knowledge graph. Background Art

[0002] Aerospace technology has increasingly higher requirements for aircraft assembly technology, and the quality of assembly directly affects aircraft performance and safety. In aircraft manufacturing, 70% of assembly tasks are mechanical connections, and the main connection methods are riveting and bolting, both of which require hole making. Therefore, the efficiency and accuracy of the hole making process parameter recommendations will greatly affect the speed and quality of aircraft assembly.

[0003] The current parameter recommendation for hole-making process mainly adopts traditional methods such as table lookup method, orthogonal experiment method and expert recommendation method. These methods are difficult to recommend process parameters quickly and accurately, and have poor autonomous reasoning ability, making it difficult to meet the needs of hole-making processing with a high level of automation. By storing hole-making process data and knowledge in the form of a graph structure, while improving search efficiency, it is also possible to combine reasoning algorithms to mine hidden relationships between data. The existing parameter recommendation methods based on knowledge graphs in the field of hole-making process have the defects of poor hole-making process ontology construction effect, high training and annotation costs of the knowledge extraction algorithm used, and poor results. Summary of the invention

[0004] In order to solve the shortcomings of the above-mentioned hole making process parameter recommendation method, the present invention proposes a hole making process parameter recommendation method based on knowledge graph. A hole making process parameter recommendation method based on knowledge graph includes:

[0005] Step S1, constructing the main body of the hole making process:

[0006] Collect data and knowledge related to the hole-making process, complete the demand analysis for hole-making process ontology construction, clarify the target scope of the hole-making process ontology based on this, reuse the ontology related to the hole-making process, and then refine the core concepts in the hole-making process field, clarify the classification relationship between concepts, define the relationship between concepts and the attributes of concepts, and finally determine whether the ontology needs to be updated through ontology evaluation. If the ontology passes, use Protégé software to visualize the ontology to form the model layer of the knowledge graph.

[0007] Step S2, using a pattern-based knowledge extraction method to complete the knowledge extraction of the hole-making process table data; after one-to-one correspondence between the table data header and the concepts in the pattern layer, the data in the table is converted into triples based on the relationship between the concepts in the pattern;

[0008] Step S3, using the UIE model to extract knowledge from the hole-making process text data: clean the collected text data to reduce the interference of irrelevant data. Annotate the collected text data based on the knowledge graph model layer, and use the annotated data to fine-tune the UIE model; finally, use the trained UIE model to extract triples from the hole-making process text data;

[0009] Step S4, knowledge fusion and storage of hole making process knowledge graph based on Neo4j graph database:

[0010] The entity fusion is completed through the similarity and the mapping table between the Chinese name of the material and its English abbreviation; the relationship fusion of the same processing cases is completed by comparing whether the data parameters of the processing cases are the same; finally, the fused hole-making process knowledge graph is stored in the Neo4j database and visualized;

[0011] Step S5, searching in the constructed knowledge graph based on the processing requirements to obtain the final recommended processing parameters.

[0012] Preferably, the step S1 specifically includes:

[0013] Step S1.1, consider the construction requirements of the hole-making process ontology and clarify the target scope of the hole-making process ontology:

[0014] First, the domain of the hole-making process is known from the demand analysis of the hole-making process. Second, the quality requirements of the hole-making process are determined.

[0015] Step S1.2, extracting the core concepts of the hole-making process and constructing the ontology of the hole-making process, you can refer to the existing ontology and reuse it;

[0016] Step S1.3, first you need to understand the domain knowledge, and then list as many domain concepts as possible that need to be expressed to the user; the formation of these concepts mainly comes from the collected hole-making process data;

[0017] Step S1.4, based on all the concepts of hole making process extracted in step S1.3, these concepts are sorted out to form a conceptual system framework of hole making process:

[0018] Firstly, the hole-making process is divided into five basic concepts: processing holes, processing plate materials, processing tools, processing quality assessment and processing process parameters; then the basic concepts are further subdivided to form specific process parameter concepts and clear classification relationships; finally, it is analyzed whether the concepts are classified correctly and whether the concepts meet the minimum ontology principle. If not, they are sorted out again until a complete hole-making process concept hierarchy system is formed;

[0019] Step S1.5, define the relationship and attributes between hole making process concepts:

[0020] Relationship refers to the possible connection between two concepts, which can be divided into object relationship and data type relationship. For a newly defined processing case class, the relationship between it and the processing hole, processing tool, processing process parameters and processing quality is used to describe a specific processing case.

[0021] Step S1.6, ontology evaluation:

[0022] The ontology evaluation method in this paper mainly evaluates the semantic depth, knowledge scope, reliability, portability and scalability of the ontology. If the ontology evaluation fails, it will be rebuilt and analyzed starting from step S1.1;

[0023] Step S1.7, using the Protégé ontology building tool to visualize the hole making process ontology.

[0024] Preferably, the step S2 specifically includes:

[0025] The table data comes from an Excel file that records hole-making process cases; the file consists of a header and cells that express specific hole-making process data, and each row represents a feasible processing case. When extracting table data, it is extracted row by row: first, each row corresponds to an entity of the processing case class, and its attribute is the number of rows. Then, the other cells in the row are analyzed in turn, one cell corresponds to an entity, the class of the entity is the class corresponding to the cell header in the pattern layer, and the attribute of the entity is the specific value of the cell; finally, a triple is constructed with the processing case entity as the head entity, the relationship between the cell class and the processing case class in the pattern layer as the relationship, and the cell entity as the tail entity, and the triple is stored in a json file. Each line of the json file contains all triples of the same processing case entity.

[0026] Preferably, the step S3 specifically includes:

[0027] Step S3.1, data cleaning of the collected literature;

[0028] Step S3.2, randomly select a set number of cleaned text data and perform data annotation on the Doccano platform. The annotated labels and relationships come from the classification and relationship in the model layer;

[0029] Step S3.3, deployment and fine-tuning of the UIE model, specifically: export the labeled data from Doccano as a json file; during training, divide the labeled data into a training set, a validation set, and a test set; and introduce a set proportion of negative examples into the training set, that is, introduce entities or relations that should not be included in the extraction results, and learn the ability to avoid errors during fine-tuning;

[0030] Step S3.4, extract hole-making process knowledge from unlabeled text using the fine-tuned UIE model, and convert the obtained hole-making process knowledge into easy-to-process triples. This process requires the extraction of process knowledge using the fine-tuned UIE and the conversion of the UIE output into triples.

[0031] When using UIE, you first need to set the structured extraction language to specify the entities and relations that need to be extracted, which is in the form of {head entity type: [relation 1, relation 2]}; according to the constructed pattern layer, the structured extraction language for hole-making process knowledge extraction is defined as {processing case: [tool type is, tool material is, aperture is, processing plate material is, rotation speed is, feed speed is, top angle is, recommended back angle is, processing plate thickness is, helix angle is, roughness is, burr height is, aperture error is, tool diameter is, feed amount per revolution is]}; then call the fine-tuned UIE model to obtain the UIE predictor; use the sentence to be predicted as the input of the predictor, and then its output is the extracted entities and relations; finally, convert the extraction result into a triple structure in the knowledge graph.

[0032] Preferably, the collected text data is cleaned to reduce the interference of irrelevant data.

[0033] Preferably, in step 1, when the processing target changes and the hole-making process body needs to be updated on a large scale, the construction is restarted from S1.1; when the hole-making processing technology is updated and the hole-making process body needs to be updated on a small scale, the core concepts of the field are revised starting from S1.3.

[0034] Preferably, in step 4, the table data triples and the text data triples are entity aligned and the process case duplicates are filtered, and the triples are stored in the Neo4j graph database to form a hole making process knowledge graph, which specifically includes:

[0035] Step S4.1, selecting Neo4j, a graph database developed based on Java, as a storage platform for the hole-making process knowledge graph;

[0036] Step S4.2, aligning entities of the knowledge extraction results according to similarity;

[0037] Step S4.3, filtering the processed cases;

[0038] Step S4.4, the hole process knowledge graph is queried, and the query nodes and relationships are returned through a visual interface.

[0039] Preferably, the step 5 specifically includes:

[0040] First, input the name of the processing plate material to be queried, evaluate the similarity between the material name and the attribute values ​​of all processing plate material nodes in the graph, and output multiple nodes with the highest similarity; then, the user confirms the material name to be queried at last; find the processing plate material node with the material name as the attribute value, and then find all the processing case nodes related to it, output the aperture nodes related to these processing cases, and then the user selects the aperture for the hole; finally, output the processing case nodes that are related to the selected aperture node and processing plate material node at the same time, and then find other nodes that are related to these process case nodes. The attribute values ​​of these nodes are the hole-making process parameters recommended based on the query knowledge graph.

[0041] Preferably, in step S3.4, a json file is used to store the triples, and one line of the json file contains all triples of the same processing case entity.

[0042] Preferably, when aligning the entities of the knowledge extraction results in step S4.2, the similarity between the texts is calculated by taking the average value of the Jaccard similarity and the Levenshtein similarity.

[0043] Preferably, in step S4.3, the filtering weight of the processing case specifically includes:

[0044] First, create a two-dimensional list variable and use the processing case entity of the row to create a processing case node; then, query the tail node in all triples of this row. If the tail node is found to already exist, use the existing node to establish a relationship and find the processing case node with the same relationship as the tail node, and store all the found nodes in a row of the list variable. If the tail node does not exist, create a new tail node and relationship, and add a blank row to the list variable; after the traversal is completed, if each row of the table variable has a common node, it means that this processing case node already exists and does not need to be created, so this processing case node is deleted.

[0045] In view of the above technical solution, the present invention has the following beneficial effects:

[0046] 1. The present invention uses an optimized seven-step method, provides an ontology evaluation system and an ontology update method, so that the hole-making process ontology can be updated and iterated according to different processing requirements.

[0047] 2. The present invention adopts a fine-tuned UIE model to extract knowledge from the hole-making process text data. Only a small sample data is needed to fine-tune the UIE model to obtain a good extraction accuracy, thereby reducing the annotation cost of knowledge extraction from the hole-making process text data and improving the extraction efficiency.

[0048] 3. The present invention designs an entity alignment algorithm and a processing case filtering algorithm based on text similarity and vocabulary mapping to reduce the repetition rate of knowledge in the knowledge graph and improve retrieval efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is an overall flow chart of the knowledge graph construction method for hole making process of the present invention;

[0050] Figure 2 This is a flow chart of the optimized seven-step method of the present invention;

[0051] Figure 3 A visualization result diagram of the hole-making process entity constructed by the present invention;

[0052] Figure 4 It is a flow chart of the entity alignment algorithm of the present invention;

[0053] Figure 5 This is a flowchart of the weight filtering algorithm for processing cases of the present invention;

[0054] Figure 6 The visualization result of the knowledge graph of the hole-making process of the present invention;

[0055] Figure 7 The following is a flow chart of the recommended hole making process parameters of the present invention. DETAILED DESCRIPTION

[0056] In order to describe the above steps S1 to S5 in more detail, they are described in detail below with reference to the accompanying drawings and examples. The specific implementation is as follows:

[0057] Step S1, using the optimized seven-step method to construct the ontology of the hole-making process. To construct the knowledge graph of the hole-making process, the ontology must first be constructed, that is, the model layer must be constructed to obtain the knowledge system of the hole-making process and provide guidance for knowledge extraction. The specific steps for constructing the ontology of the hole-making process according to the optimized seven-step method are as follows: Steps S1.1 to S1.7, and the flow chart is as follows: Figure 2 shown.

[0058] Step S1.1, clarify the target scope of the hole-making process ontology. The ontology development first needs to comprehensively consider the construction requirements of the hole-making process ontology and determine the target scope of the ontology construction. First of all, from the demand analysis of the hole-making process, it can be known that the field of its ontology is mainly limited to the field of aviation hole-making processes dominated by drilling. Secondly, the hole-making process ontology is mainly used for hole-making process parameter recommendations. The parameter recommendation depends on the correct processing case. What is important in a case is the properties of the hole-making processing plate, the hole-making processing parameters, the processing tools, and the quality requirements for the holes. Finally, the above two points can be updated as the hole-making process parameter recommendations are needed.

[0059] Step S1.2, reuse the existing ontology. Extract the core concepts of the hole-making process. Before constructing the ontology of the hole-making process, you can refer to the existing ontology and reuse it appropriately. At present, there is no existing ontology for reference in the field of hole-making technology, and the cutting tool ontology is referenced during construction.

[0060] Step S1.3, extract the core concepts of the hole-making process. The construction of the hole-making process ontology first requires understanding the domain knowledge, and then listing as many domain concepts as possible that need to be expressed to users. The formation of these concepts mainly comes from the collected hole-making process data.

[0061] Step S1.4, define the conceptual hierarchy of the hole-making process. Based on the extraction of all the concepts of the hole-making process in step S1.3, this step is mainly to sort out these concepts to form a well-defined conceptual framework of the hole-making process. First, the hole-making process is divided into five basic concepts: processing holes, processing plate materials, processing tools, processing quality assessment, and processing process parameters. These basic concepts cover almost all parameters involved in the hole-making process. Then the basic concepts are further subdivided to form specific process parameter concepts and a clear classification relationship. For example, the processing process parameters include special processes involved in the processing, the feed speed, rotation speed, and feed rate of the hole-making robot, and the feed rate if tool change or layered processing is required. Finally, analyze whether the concepts are classified correctly and whether the concepts meet the principle of minimum ontology. If not, sort them out again until a complete conceptual hierarchy of the hole-making process is formed.

[0062] Step S1.5, define the relationship and attributes between hole-making process concepts. A relationship refers to the possible connection between two concepts. This connection can be divided into object relationships and data type relationships. Data type relationships are usually called attributes, and attributes are a kind of relationship in a broad sense. A new processing case class is defined, and its relationship with processing holes, processing tools, processing process parameters, and processing quality can be used to describe a specific processing case.

[0063] Step S1.6, ontology evaluation. The ontology evaluation method in this paper mainly evaluates the semantic depth, knowledge scope, reliability, portability and scalability of the ontology. If the ontology evaluation fails, it will be rebuilt and analyzed from step S1.1.

[0064] Step S1.7, ontology visualization. The constructed hole-making process ontology is visualized using the Protégé ontology building tool. The visualization results are shown in Figure 1. Figure 3 shown.

[0065] When the processing target changes and the hole-making process ontology needs to be updated on a large scale, it is rebuilt from S1.1; when the hole-making processing technology is updated and the hole-making process ontology needs to be updated on a small scale, the core concepts of the field are revised starting from S1.3.

[0066] Step S2, knowledge extraction of pattern-based tabular data. The tabular data comes from an excel file that records hole-making process cases. The file consists of a header and cells that express specific hole-making process data, and each row represents a feasible processing case. When extracting tabular data, row-by-row extraction is adopted. The following takes the extraction of a row of data as an example. First, each row corresponds to an entity of a processing case class, and its attribute is the number of rows. Then, the other cells in the row are analyzed in turn. One cell corresponds to an entity. The class of the entity is the class corresponding to the cell corresponding to the header in the pattern layer, and the attribute of the entity is the specific value of the cell. Finally, a triple is constructed with the processing case entity as the head entity, the relationship between the cell class and the processing case class in the pattern layer as the relationship, and the cell entity as the tail entity, and the triple is stored in a json file. Each line of the json file contains all triples of the same processing case entity.

[0067] Step S3, using the fine-tuned UIE model to extract knowledge from the hole-making process text data. The text data comes from relevant literature that records hole-making process knowledge and cases. The UIE model is a pre-trained large language model. The UIE model fine-tuned using the hole-making process data can extract entities and relationships in the hole-making process text according to the set structured extraction language. This step completes the fine-tuning of the UIE model in the field of hole-making technology and uses the fine-tuned UIE to extract hole-making process text data and convert it into triples.

[0068] Step S3.1, text data cleaning. First, the collected documents need to be cleaned, and the steps include:

[0069] 1) Locate key text positions based on keywords;

[0070] 2) Convert all characters into half-width characters;

[0071] 3) Remove all spaces and line breaks;

[0072] Step S3.2, training data annotation. Randomly select 50 cleaned text data for data annotation on the Doccano platform. The annotated labels and relationships come from the classification and relationship in the model layer. The detailed steps of annotation are as follows:

[0073] 1) Create a project on the Doccano platform that supports sequence labeling tasks, and check "Allow entity overlap" and "Use relationship labels".

[0074] 2) Save the data to be uploaded in a txt file, and separate the sentences to be marked with line breaks. Select this file when uploading data in Doccano, and set the upload data type to TextLine.

[0075] 3) According to the model layer and the data to be annotated, determine the entities and relationships that need to be annotated. Add corresponding labels in Doccano.

[0076] 4) Finally, mark the text. Select the corresponding entity by clicking the text with the mouse. Select two entities to select the relationship between them.

[0077] Step S3.3, deployment and fine-tuning of UIE model. The UIE model is deployed on the local platform. The specific training environment is shown in Table 1.

[0078] Table 1 Training environment configuration

[0079]

[0080] Export the labeled data from Doccano as a json file. During training, the labeled data is divided into a training set, a validation set, and a test set in a ratio of 3:1:1. In order to make the fine-tuned UIE adapt to information extraction tasks under different settings and solve the exposure error of autoregressive Teacher-Forcing, a certain proportion of negative examples are introduced in the training set, that is, entities or relations that should not be included in the extraction results are introduced to learn the ability to avoid errors during fine-tuning. The ratio of the number of negative examples to positive examples is called negative ratio.

[0081] The basic UIE model is selected as the training model, and the hyperparameters selected during training are shown in Table 2.

[0082] Table 2 Training hyperparameters

[0083] Parameter Type Parameter Value negative ratio 0.05 Learning rate 5e-6 Batch size 8

[0084] After training, the F1 value in the validation set is 0.775, and the F1 value in the test set is 0.614.

[0085] Step S3.4, use the fine-tuned UIE model to extract knowledge. Use the fine-tuned UIE model to extract hole-making process knowledge from unlabeled text, and convert the obtained hole-making process knowledge into easy-to-process triples. This process requires the extraction of process knowledge using the fine-tuned UIE and the conversion of the UIE output into triples.

[0086] When using UIE, you first need to set the structured extraction language to specify the entities and relations that need to be extracted, which is in the form of {head entity type: [relation 1, relation 2]}. According to the pattern layer designed in Chapter 2, the structured extraction language for hole-making process knowledge extraction can be defined as {processing case: [tool type is, tool material is, aperture is, processing plate material is, rotation speed is, feed speed is, top angle is, recommended back angle is, processing plate thickness is, helix angle is, roughness is, burr height is, aperture error is, tool diameter is, feed per revolution is]}. Then call the fine-tuned UIE model to obtain the UIE predictor. The sentence to be predicted is used as the input of the predictor, and its output is the extracted entities and relations. Finally, the extraction result is converted into a triple structure in the knowledge graph.

[0087] In order to facilitate further processing of triple files, json files are used to store triples. One line of the json file contains all triples of the same processing case entity.

[0088] Step S4, knowledge fusion and storage of hole-making process knowledge graph based on Neo4j graph database. In this step, the table data triples and text data triples are entity aligned and process case duplicates are filtered, and the triples are stored in the Neo4j graph database to form a hole-making process knowledge graph.

[0089] Step S4.1, select the storage platform of the hole-making process knowledge graph. Neo4j is a graph database developed based on Java. It can store graph-structured data directly in the network structure instead of the table structure used by the relational database, so its data has more comprehensive object characteristics. Neo4j does not rely on Schema when storing data, has good compatibility with data, and has the characteristics of index-free adjacency. Combined with the Cypher language provided by Neo4j, efficient queries on the graph database can be achieved. The attribute graph structure of Neo4j can be used to store data in the knowledge graph and visualize it. Therefore, the Neo4j graph database is used for the hole-making process knowledge graph, and entities are expressed as nodes and relationships are expressed as edges when storing.

[0090] Step S4.2, entity alignment of knowledge extraction results. Text data comes from different documents, and there may be problems such as content duplication, semantic diversity, and uneven quality. When describing the same physical quantity or physical quantity attribute, different documents may have large differences. For example, for CFRP / Al laminated materials, some documents directly use "CFRP / Al", while other documents use "CFRP / Al alloy". Therefore, the same entity may appear in the system in different semantic forms. Organically integrating knowledge from multiple sources is a necessary condition for building a high-quality hole-making process knowledge graph.

[0091] When aligning entities, it is necessary to determine the similarity between entity attribute values. Evaluating the similarity between texts using Jaccard similarity and Levenshtein similarity will make the determination of similarity more reasonable.

[0092] If a and b represent two short texts, the Jaccard similarity calculation formula is as shown in Formula 1.

[0093]

[0094] Among them, A and B represent the character sets that make up the short text, |A∩B| represents the number of characters shared by the two sets, and |A∪B| represents the total number of characters in the two sets.

[0095] The Levenshtein similarity coefficient calculation formula is:

[0096]

[0097] Among them, d(A,B) is the Levenshtein minimum edit distance, and its formula is:

[0098]

[0099] Where |A| and |B| represent the lengths of strings A and B, respectively. A[1:] and B[1:] represent strings A and B after removing the first character. δ(A[1], B[1]) is a function that compares the first characters of A and B and returns 1 if they are the same, otherwise it returns 0.

[0100] The final similarity is the average of Jaccard similarity and Levenshtein similarity, and the formula is:

[0101]

[0102] Professional terms such as material names are often expressed by their English abbreviations. This step also uses the method of constructing a mapping vocabulary to avoid interference between different representation methods on similarity. Before comparing the similarity, it is necessary to construct new matching terms that replace the professional terms with their English abbreviations, which together with the original text form a group of terms to be matched. Then the similarity of the attribute values ​​of the same nodes is calculated with each term in the group of terms to be matched, and the maximum similarity is taken as the similarity between the new node and the node. Sort the similarities between the new node and all the nodes in the graph. If there is a node with a similarity of 1, it means that the same node already exists and does not need to be created. Return the node for subsequent operations. If there is no node with a similarity of 1 but there is a node with a higher similarity, manually select whether to replace it. If replaced, return the selected node. If not replaced, return the newly created node. The specific flow chart is as follows Figure 4 shown.

[0103] Step S4.3, filtering duplicate processing cases. In order to prevent the same specific processing cases from being repeatedly established, it is necessary to implement filtering duplicate processing cases. Take reading a line of a json file as an example. First, establish a two-dimensional list variable and use the processing case entity of the line to establish a processing case node. Then, query the tail nodes in all triples in this line. If it is found that the tail node already exists, use the existing node to establish a relationship and find the processing case node with the same relationship as the tail node. All the found nodes are stored in a row of the list variable. If the tail node does not exist, create a new tail node and relationship, and add a blank line to the list variable. After the traversal is completed, if each row of the table variable has a common node, it means that this processing case node already exists and does not need to be created, so this processing case node is deleted. The specific flow chart is as follows Figure 5 shown.

[0104] Step S4.4, visualization of the hole-making process knowledge graph. In the Neo4j database, write a Cypher statement to query the knowledge graph, and return the query nodes and relationships through a visualization interface. By entering the statement "MATCH(n)RETURN n" in the Neo4j database, the overall visualization effect of the graph can be obtained, such as Figure 6 shown.

[0105] Step S5, use the constructed knowledge graph to recommend parameters. Other hole making process parameters are recommended based on the processing plate material and hole diameter. The flowchart of the query process is as follows: Figure 7 As shown. First, input the name of the processing plate material to be queried, use the average Jaccard similarity and Levenshtein similarity to evaluate the similarity between the material name and the attribute values ​​of all processing plate material nodes in the graph, and output the nodes with the highest similarity. Then, the user confirms the name of the material to be queried. Find the processing plate material node with the material name as the attribute value, then find all the processing case nodes related to it, output the aperture nodes related to these processing cases, and then the user selects the aperture for the hole. Finally, output the processing case nodes that are related to the selected aperture node and processing plate material node at the same time, and then find other nodes that are related to these process case nodes. The attribute values ​​of these nodes are the hole-making process parameters recommended based on the query knowledge graph.

Claims

1. A method for recommending hole-making process parameters based on knowledge graph, characterized in that: include: Step S1, constructing the main body of the hole making process: Collect hole-making process-related data and knowledge, complete the demand analysis for hole-making process ontology construction, clarify the target scope of hole-making process ontology based on this, reuse the ontology related to hole-making process, and then refine the core concepts in the field of hole-making process, clarify the classification relationship between concepts, define the relationship between concepts and the attributes of concepts, and finally determine whether the ontology needs to be updated through ontology evaluation; if the ontology passes, use Protégé software to visualize the ontology to form the model layer of the knowledge graph; Step S2, using a pattern-based knowledge extraction method to complete the knowledge extraction of the hole-making process table data; after one-to-one correspondence between the table data header and the concepts in the pattern layer, the data in the table is converted into triples based on the relationship between the concepts in the pattern; Step S3, using the UIE model to extract knowledge from the hole-making process text data: performing data cleaning on the collected text data; The collected text data is annotated based on the pattern layer of the knowledge graph, and the UIE model is fine-tuned using the annotated data. Finally, the trained UIE model is used to extract triples from the hole-making process text data. Step S4, knowledge fusion and storage of hole making process knowledge graph based on Neo4j graph database: The entity fusion is completed through the similarity and the mapping table between the Chinese name of the material and its English abbreviation; the relationship fusion of the same processing cases is completed by comparing whether the data parameters of the processing cases are the same; finally, the fused hole-making process knowledge graph is stored in the Neo4j database and visualized; Step S5, searching in the constructed knowledge graph based on the processing requirements to obtain the final recommended processing parameters.

2. A method for recommending hole-making process parameters based on knowledge graph according to claim 1, characterized in that: The step S1 specifically includes: Step S1.1, consider the construction requirements of the hole-making process ontology and clarify the target scope of the hole-making process ontology: First, the domain of the hole-making process is known from the demand analysis of the hole-making process. Second, the quality requirements of the hole-making process are determined. Step S1.2, extracting the core concepts of the hole-making process and constructing the ontology of the hole-making process, you can refer to the existing ontology and reuse it; Step S1.3, first you need to understand the domain knowledge, and then list as many domain concepts as possible that need to be expressed to the user; the formation of these concepts mainly comes from the collected hole-making process data; Step S1.4, based on all the concepts of hole making process extracted in step S1.3, these concepts are sorted out to form a conceptual system framework of hole making process: Firstly, the hole-making process is divided into five basic concepts: processing holes, processing plate materials, processing tools, processing quality assessment and processing process parameters; then the basic concepts are further subdivided to form specific process parameter concepts and clear classification relationships; finally, it is analyzed whether the concepts are classified correctly and whether the concepts meet the minimum ontology principle. If not, they are sorted out again until a complete hole-making process concept hierarchy system is formed; Step S1.5, define the relationship and attributes between hole making process concepts: Relationship refers to the possible connection between two concepts, which can be divided into object relationship and data type relationship. For a newly defined processing case class, the relationship between it and the processing hole, processing tool, processing process parameters and processing quality is used to describe a specific processing case. Step S1.6, ontology evaluation: The ontology evaluation method judges the semantic depth, knowledge scope, reliability, portability and scalability of the ontology. If the ontology evaluation fails, it will be rebuilt and analyzed starting from step S1.1; Step S1.7, using the Protégé ontology building tool to visualize the hole making process ontology.

3. A method for recommending hole-making process parameters based on knowledge graph according to claim 1, characterized in that: The step S2 specifically includes: The table data comes from an Excel file that records hole-making process cases; the file consists of a header and cells that express specific hole-making process data, and each row represents a feasible processing case; when extracting table data, row-by-row extraction is adopted: first, each row corresponds to an entity of the processing case class, and its attribute is the row number; then, the other cells of the row are analyzed in turn, one cell corresponds to an entity, and the class of the entity is the class corresponding to the cell header in the pattern layer, and the attribute of the entity is the specific value of the cell; finally, a triple is constructed with the processing case entity as the head entity, the relationship between the cell class and the processing case class in the pattern layer as the relationship, and the cell entity as the tail entity, and the triple is stored in a json file. Each line of the json file contains all triples of the same processing case entity.

4. A method for recommending hole making process parameters based on knowledge graph according to claim 1, characterized in that: The step S3 specifically includes: Step S3.1, data cleaning of the collected literature; Step S3.2, randomly select a set number of cleaned text data and perform data annotation on the Doccano platform. The annotated labels and relationships come from the classification and relationship in the model layer; Step S3.3, deployment and fine-tuning of the UIE model, specifically: export the labeled data from Doccano as a json file; during training, divide the labeled data into a training set, a validation set, and a test set; and introduce a set proportion of negative examples into the training set, that is, introduce entities or relations that should not be included in the extraction results, and learn the ability to avoid errors during fine-tuning; Step S3.4, using the fine-tuned UIE model to extract hole-making process knowledge from unlabeled text, and converting the obtained hole-making process knowledge into easy-to-process triples; this process requires the extraction of process knowledge using the fine-tuned UIE and the conversion of the UIE output into the triple format; When using UIE, you first need to set the structured extraction language to specify the entities and relations that need to be extracted, which is in the form of {head entity type: [relation 1, relation 2]}; according to the constructed pattern layer, the structured extraction language for hole-making process knowledge extraction is defined as {processing case: [tool type is, tool material is, aperture is, processing plate material is, rotation speed is, feed speed is, top angle is, recommended back angle is, processing plate thickness is, helix angle is, roughness is, burr height is, aperture error is, tool diameter is, feed amount per revolution is]}; then call the fine-tuned UIE model to obtain the UIE predictor; use the sentence to be predicted as the input of the predictor, and then its output is the extracted entities and relations; finally, convert the extraction result into a triple structure in the knowledge graph.

5. The method for recommending hole making process parameters based on knowledge graph according to claim 1, characterized in that: In step 1, when the processing target changes and the hole-making process ontology needs to be updated on a large scale, it is rebuilt from S1.1; when the hole-making processing technology is updated and the hole-making process ontology needs to be updated on a small scale, the core concepts of the field are revised from S1.

3.

6. A method for recommending hole making process parameters based on knowledge graph according to claim 1, characterized in that: In step 4, the table data triples and text data triples are entity aligned and the process case duplicates are filtered, and the triples are saved in the Neo4j graph database to form a hole-making process knowledge graph, which specifically includes: Step S4.1, selecting Neo4j, a graph database developed based on Java, as a storage platform for the hole-making process knowledge graph; Step S4.2, aligning entities of the knowledge extraction results according to similarity; Step S4.3, filtering the processed cases; Step S4.4, the hole process knowledge graph is queried, and the query nodes and relationships are returned through a visual interface.

7. The method for recommending hole-making process parameters based on knowledge graph according to claim 1, characterized in that: The step 5 specifically includes: First, input the name of the processing plate material to be queried, evaluate the similarity between the material name and the attribute values ​​of all processing plate material nodes in the graph, and output multiple nodes with the highest similarity; then, the user confirms the material name to be queried at last; find the processing plate material node with the material name as the attribute value, and then find all the processing case nodes related to it, output the aperture nodes related to these processing cases, and then the user selects the aperture for the hole; finally, output the processing case nodes that are related to the selected aperture node and processing plate material node at the same time, and then find other nodes that are related to these process case nodes. The attribute values ​​of these nodes are the hole-making process parameters recommended based on the query knowledge graph.

8. The method for recommending hole making process parameters based on knowledge graph according to claim 1, characterized in that: In step S3.4, a json file is used to store triples, and one line of the json file contains all triples of the same processing case entity.

9. The method for recommending hole making process parameters based on knowledge graph according to claim 1, characterized in that: When aligning the entities of the knowledge extraction results in step S4.2, the similarity between the texts is calculated by taking the average of the Jaccard similarity and the Levenshtein similarity.

10. The method for recommending hole making process parameters based on knowledge graph according to claim 1, characterized in that: In step S4.3, the filtering weight of the processing case specifically includes: First, create a two-dimensional list variable and use the processing case entity of the row to create a processing case node; then, query the tail node in all triples of this row. If the tail node is found to already exist, use the existing node to establish a relationship and find the processing case node with the same relationship as the tail node, and store all the found nodes in a row of the list variable. If the tail node does not exist, create a new tail node and relationship, and add a blank row to the list variable; after the traversal is completed, if each row of the table variable has a common node, it means that this processing case node already exists and does not need to be created, so delete this processing case node.

Images