Manufacturing industry process parameter optimization method based on knowledge graph

Abstract

The invention discloses a manufacturing industry process parameter optimization method based on a knowledge graph, and belongs to the field of manufacturing industry intellectualization. The technical problems that in the prior art, parameter optimization depends on artificial experience, systematic management is lacked, intelligent reasoning is difficult to conduct, and knowledge retrieval precision is low are solved. The method comprises the steps of firstly collecting and preprocessing manufacturing industry process data; then, constructing a semantic vector embedding model, and converting the process description into vector representation by adopting a pre-trained bidirectional encoder; defining a process knowledge ontology model, and establishing a parameter association relationship through semantic coding to form a knowledge graph; storing the coded knowledge vector into a distributed database and establishing a retrieval index; user requirements are received, and parameter matching is carried out through cosine similarity calculation; optimizing the matching parameters based on the knowledge graph association relationship; and finally collecting feedback to continuously update the knowledge graph. Through continuous optimization of the feedback learning mechanism, the precision and efficiency of technological parameter optimization in the manufacturing industry are improved.

Description

TECHNICAL FIELD

[0001] The application relates to a manufacturing process parameter optimization method, in particular to a manufacturing process parameter optimization method based on a knowledge graph, and belongs to the technical field of manufacturing informatization. BACKGROUND

[0002] With the rapid development and application of information technology, intelligent manufacturing has become an important direction for the transformation and upgrading of manufacturing industry. In the manufacturing production process, process parameter optimization has an important influence on product quality and production efficiency. Traditional process parameter optimization methods mainly rely on expert experience and have strong subjectivity, which is difficult to adapt to complex and variable production environments.

[0003] In the prior art, there are some data-driven optimization methods, such as optimizing process parameters through statistical analysis, machine learning and other technologies. However, these methods often ignore the structured expression and reasoning of knowledge, and it is difficult to fully utilize the knowledge and experience of field experts. At the same time, with the explosive growth of manufacturing data, how to effectively extract knowledge from massive data and realize process parameter optimization has become a problem to be solved.

[0004] As a knowledge representation method, the knowledge graph can represent entities and their relationships in a graphical manner and has strong semantic expression ability. Applying knowledge graph technology to manufacturing process parameter optimization is expected to solve the problem that traditional methods cannot fully utilize domain knowledge and realize more intelligent parameter optimization.

[0005] However, the current manufacturing process parameter optimization method based on knowledge graph still has the following shortcomings: The efficiency of constructing a knowledge graph is low, and it is difficult to quickly extract effective knowledge from a large number of manufacturing documents; The knowledge representation is not accurate enough to accurately capture the complex relationships between process parameters; The optimization reasoning ability is limited, and it is difficult to fully utilize the semantic analysis advantage of the knowledge graph; The system is not very practical, and it is not closely combined with the actual manufacturing process.

[0006] Therefore, there is an urgent need for a manufacturing process parameter optimization method that can efficiently construct a knowledge graph, accurately represent parameter relationships, have intelligent reasoning ability and be effectively applied to actual production. SUMMARY

[0007] The application aims to provide a manufacturing process parameter optimization method based on a knowledge graph, which aims to solve the problem that the field knowledge and experience cannot be fully utilized in the process of process parameter optimization in the prior art, and realizes intelligent optimization of process parameters by constructing a process knowledge graph.

[0008] To achieve the above objectives, this invention provides a method for optimizing manufacturing process parameters based on knowledge graphs, comprising the following steps: S1: Process knowledge acquisition and preprocessing: Collect process knowledge sources such as manufacturing process documents, operation manuals, and expert experience, and form standardized text through intelligent segmentation and preprocessing; S2: Construction of process knowledge graph. Natural language processing technology is used to extract process entities and their relationships from preprocessed text and construct a process knowledge graph. S3: Process knowledge graph storage and indexing, which stores the constructed knowledge graph in a graph database and establishes a vector index to achieve efficient querying and retrieval; S4: Semantic query of process parameters, based on the user's query intent, uses semantic vector matching to retrieve relevant process parameter nodes and relationships; S5: Process parameter relationship analysis, based on the topology of knowledge graph, analyzes the influence relationship and constraints between process parameters; S6: Process parameter optimization reasoning. Based on parameter relationships and manufacturing objectives, a graph-based reasoning algorithm is used to generate process parameter optimization schemes.

[0009] The present invention has the following beneficial effects: Through intelligent text analysis technology, the ability to efficiently extract knowledge from a large number of unstructured process documents has been achieved, which greatly improves the efficiency of knowledge graph construction. By employing a refined entity relationship extraction method, we can accurately capture the complex relationships between process parameters, providing a reliable knowledge base for subsequent optimization reasoning. The storage scheme based on graph databases and vector indexes supports efficient semantic queries and similarity searches, significantly improving the system's retrieval performance. Introducing graph-based reasoning algorithms can comprehensively consider the influence relationships and constraints between parameters, and generate more reasonable process parameter optimization schemes. The overall approach is closely integrated with the actual manufacturing process, making it highly practical and adaptable, and applicable to a variety of manufacturing scenarios. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0011] Figure 1 A flowchart illustrating a knowledge graph-based method for optimizing manufacturing process parameters, as provided in an embodiment of the present invention.

[0012] Figure 2 This is a structural diagram of the process knowledge graph construction module in an embodiment of the present invention.

[0013] Figure 3 This is a schematic diagram illustrating the process of reasoning for optimizing process parameters in an embodiment of the present invention. Detailed Implementation

[0014] To enable those skilled in the art to better understand the technical solutions of the present invention, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0015] Please refer to Figure 1 As shown in the figure, this invention provides a method for optimizing manufacturing process parameters based on knowledge graphs, which mainly includes six steps: process knowledge acquisition and preprocessing, process knowledge graph construction, process knowledge graph storage and indexing, process parameter semantic query, process parameter relationship analysis, and process parameter optimization reasoning. Each step will be described in detail below with reference to specific implementation methods.

[0016] Step S1: Acquisition of process knowledge and pretreatment This step mainly involves acquiring and preprocessing manufacturing process knowledge, providing standardized basic data for subsequent knowledge graph construction.

[0017] First, acquire relevant manufacturing process knowledge. Sources of this knowledge include, but are not limited to, textual materials such as process documents, operation manuals, product specifications, expert experience summaries, production logs, and fault records. These materials may come from the company's internal management systems, document libraries, and other professional databases or resource repositories.

[0018] Secondly, the acquired process knowledge text is intelligently segmented. Specifically, a semantic-based intelligent segmentation algorithm is used to divide the long text into semantic units of appropriate size, ensuring that each text block has complete semantic content. During the segmentation process, strong breakpoints such as periods, exclamation marks, and question marks are given priority as segmentation criteria, followed by line breaks as secondary breakpoints, and finally weak breakpoints such as commas and semicolons are considered. Simultaneously, minimum and maximum target block size parameters are set to ensure that the segmented text blocks are of moderate size, facilitating subsequent processing.

[0019] Finally, the segmented text blocks undergo preprocessing, including text cleaning, standardization, and format conversion. Text cleaning primarily removes irrelevant information, special characters, and redundant whitespace; standardization includes terminology standardization, unit conversion, and format specification; and format conversion ensures that all text conforms to the unified format required by the system.

[0020] Through the above processing, the original unstructured process knowledge is transformed into standardized text blocks, laying the foundation for subsequent knowledge graph construction.

[0021] Step S2: Construction of Process Knowledge Graph Please refer to Figure 2 As shown, this step mainly involves extracting entities and relationships from the preprocessed process knowledge text to construct a process knowledge graph.

[0022] First, entity extraction is performed on each text block. The entity extraction module is responsible for identifying key concepts in the text, including but not limited to: process parameters (such as temperature, pressure, speed, etc.), equipment components, material properties, process steps, quality indicators, etc. Entity extraction uses natural language processing technology combined with domain ontology knowledge to ensure that the extracted entities are domain-relevant and accurate.

[0023] Secondly, relationship extraction is performed on the identified entities. The relationship extraction module is responsible for discovering semantic associations between entities, mainly including: parameter influence relationships (such as "temperature affects hardness"), parameter constraint relationships (such as "pressure must be within a specific range"), causal relationships (such as "excessive rotation speed causes vibration"), and compositional relationships (such as "heat treatment includes quenching and tempering"). During the relationship extraction process, in addition to identifying the relationship type, descriptive information of the relationship is also recorded, such as the degree of influence and constraint conditions.

[0024] Next, the extracted entities and relationships are integrated and standardized. This step mainly addresses issues such as entity redundancy and relationship consistency, including entity disambiguation (identifying and merging different expressions referring to the same concept), relationship verification (ensuring that the extracted relationships conform to domain rules), and attribute supplementation (adding necessary attribute information to entities).

[0025] Finally, an initial process knowledge graph is constructed. Normalized entities are used as graph nodes, and relationships are used as graph edges, forming the initial knowledge graph structure. Each node contains information such as entity name, entity type, and attributes; each edge contains information such as relationship type, relationship description, and source text block identifier. Through this process, the transformation from unstructured text to a structured knowledge graph is completed, effectively capturing entity concepts and their relationships within process knowledge.

[0026] Step S3: Storage and Indexing of Process Knowledge Graph This step mainly involves storing and indexing the process knowledge graph, providing efficient support for subsequent queries and retrievals.

[0027] First, select an appropriate graph database for knowledge graph storage. Graph databases natively support graph structure data, making them suitable for storing and querying complex entity relationship networks. During storage, nodes, edges, and their attributes in the knowledge graph are mapped to the corresponding structures in the graph database, ensuring data integrity and consistency. Simultaneously, appropriate data models and constraints are set to guarantee the standardization of storage.

[0028] Secondly, a vector index is built for the knowledge graph. Specifically, a text embedding model is used to convert textual information such as entity names and descriptions into high-dimensional vector representations, capturing the semantic features of the entities. These vectors are stored and indexed to support subsequent semantic similarity searches. The establishment of the vector index significantly improves the efficiency and accuracy of the system in processing semantic queries.

[0029] Secondly, establish a knowledge graph metadata management mechanism. Metadata includes version information, construction time, data source, entity and relationship statistics, etc. Metadata management helps in the maintenance, updating, and quality control of the knowledge graph.

[0030] Finally, a knowledge graph update mechanism was designed and implemented. As new technological knowledge continues to emerge, the knowledge graph needs to be updated regularly to maintain its timeliness and completeness. The update mechanism includes various methods such as incremental updates (adding new entities and relationships), corrective updates (modifying error messages), and expansion updates (expanding the attributes of existing entities).

[0031] Through the above processing, efficient storage and indexing of the process knowledge graph were achieved, providing basic support for subsequent parameter querying and optimization.

[0032] Step S4: Semantic query of process parameters This step mainly implements the semantic query function for process parameters based on user intent in order to obtain relevant process parameter information.

[0033] First, the system receives and parses the user's query request. The query request can be in natural language (e.g., "What heat treatment parameters affect the hardness of steel?") or structured form (e.g., query conditions specifying particular process parameters). The system performs semantic understanding on the query request, extracting the query intent and key concepts.

[0034] Secondly, the query is converted into a vector representation. Using the same text embedding model as the knowledge graph index, the query text is converted into vector form to enable semantic matching with entity vectors in the knowledge graph.

[0035] Next, a vector similarity search is performed. Based on the vector index, the similarity between the query vector and the entity vectors in the knowledge graph is calculated to identify the entity nodes most relevant to the query. Similarity calculation uses cosine similarity and other measurement methods, and a similarity threshold is set to filter the results.

[0036] Next, relationship expansion is performed based on the initial matching results. The system not only returns entities directly related to the query, but also explores related nodes and relationships of these entities using a graph traversal algorithm to obtain a more complete knowledge network. During relationship expansion, a traversal depth parameter can be set to control the expansion range.

[0037] Finally, the query results are organized and sorted. Based on factors such as entity relevance scores and relationship importance, the results are comprehensively sorted to ensure that the most relevant process parameter information is displayed first. At the same time, the results are organized into a hierarchical structure for easy user understanding and use.

[0038] Through the above process, an efficient mapping from user natural language queries to the process parameter knowledge network is achieved, providing an accurate data foundation for subsequent parameter relationship analysis.

[0039] Step S5: Analysis of process parameter relationships This step primarily uses the topological structure of the knowledge graph to analyze the complex relationships and constraints between process parameters.

[0040] First, a parameter influence network is constructed. From the knowledge subgraph obtained from the query, all process parameter nodes and their interrelationships are extracted to construct the parameter influence network. Nodes in the network represent process parameters, and edges represent the influence relationships between parameters. The attributes of the edges include information such as influence type, influence degree, and relationship description.

[0041] Secondly, parameter importance analysis is performed. Based on graph centrality algorithms (such as degree centrality, betweenness centrality, eigenvector centrality, etc.), the importance index of each parameter node in the network is calculated. Importance analysis helps identify core parameters that have a critical impact on the process, providing a focus for optimization.

[0042] Next, identify the parameter constraints. Extract parameter constraint information from the knowledge graph, including value range constraints (e.g., "temperature must be between 800-900℃"), parameter correlation constraints (e.g., "temperature should not exceed a specific value when pressure increases"), and process specification constraints (e.g., "parameter configuration requirements for a specific process according to industry standards"). These constraints are rules that must be followed during parameter optimization.

[0043] Next, parameter sensitivity is analyzed. Based on the parameter relationship descriptions recorded in the knowledge graph, the sensitivity of each parameter to the target indicator is assessed, and sensitive parameters that significantly affect product quality or process effectiveness are identified. The sensitivity analysis results will guide parameter adjustment strategies in subsequent optimization processes.

[0044] Finally, a parameter optimization objective model is constructed. Based on user needs and process characteristics, the objective function for parameter optimization is determined, such as maximizing product quality, minimizing energy consumption, or balancing multiple indicators. The objective model will serve as the evaluation basis for parameter optimization inference.

[0045] The above analysis has led to a comprehensive understanding of the relationships between process parameters, laying the foundation for further parameter optimization reasoning.

[0046] Step S6: Process parameter optimization reasoning Please refer to Figure 3 As shown, this step mainly uses graph reasoning algorithms based on the results of parameter relationship analysis to generate optimized schemes for process parameters.

[0047] First, initialize the parameter configuration. Based on typical configurations in the knowledge graph or parameter values ​​currently used in production, set the initial state of the parameters. The initial configuration should meet the parameter constraints and serve as the starting point for optimization.

[0048] Secondly, a reasoning rule base is constructed. Parameter influence rules, expert experience rules, and process theory rules are extracted from the knowledge graph to form a structured reasoning rule base. The rules can be expressed as conditional statements like "if...then..." or as mathematical expressions relating parameters.

[0049] Then, graph-based parameter optimization inference is performed. The inference process employs heuristic search and constraint propagation algorithms. Under the premise of satisfying all constraints, the parameter configuration is gradually adjusted according to the influence relationships between parameters, moving closer to the optimization objective. During the inference process, the system comprehensively considers factors such as the importance, sensitivity, and interdependencies of the parameters to ensure the rationality and effectiveness of the adjustments.

[0050] Next, the effectiveness of the optimization scheme is evaluated. Based on empirical data or theoretical models in the process knowledge graph, the generated parameter configuration scheme is evaluated to predict its potential process effects or product quality levels. The evaluation results will serve as the basis for adjusting the scheme.

[0051] Finally, a parameter optimization recommendation report is generated. The report includes the optimized parameter configuration, expected results, the basis and rationale for parameter adjustments, and implementation suggestions. The report is presented in an easy-to-understand format for easy reference and implementation by process engineers.

[0052] Through the above reasoning process, the process knowledge contained in the knowledge graph is transformed into specific parameter optimization schemes, realizing knowledge-driven intelligent optimization of process parameters.

[0053] Through the above six steps, this invention realizes a complete process from process knowledge acquisition to parameter optimization reasoning, and constructs a set of manufacturing process parameter optimization methods based on knowledge graphs, which can effectively improve the intelligence level of manufacturing process parameter optimization and provide technical support for improving product quality and production efficiency.

Claims

1. A method for optimizing manufacturing process parameters based on knowledge graphs, characterized in that, Includes the following steps: Collect and preprocess the process parameters of the manufacturing industry, including process documents, parameter records and expert experience, and perform intelligent block segmentation, noise reduction and structured processing on the text to identify and extract key information such as process parameters, process flow and quality indicators; A knowledge graph extraction system based on a large language model is built, which intelligently divides process documents into blocks according to a preset range of character counts and prioritizes the use of delimiters to process text breakpoints. Ontology analysis is performed on each text block using a large language model to extract core entities and relationships between entities in the process domain, and the extracted results are formatted into structured data containing nodes and edges. The structured data is stored in a graph database, and process entity nodes and relationship edges between entities are established in the graph database to form a manufacturing process knowledge graph. Construct a semantic vector embedding model to transform the textual description of process entities into high-dimensional vector representations, and store the vector representations in the attributes of graph database nodes; Receive user process requirement description and perform semantic encoding, convert the requirement description into vector representation, calculate the similarity between the query vector and the vector in the knowledge base based on vector similarity, and filter the matching results according to a preset threshold; Based on the matched entity nodes, retrieve related entity nodes and relationships from the knowledge graph, and construct a context-dependent subgraph that contains the relationships between process parameters; Based on the relationships in the knowledge subgraph, the matched process parameters are optimized and combined to generate process parameter optimization suggestions; Collect user feedback on optimization suggestions, adjust the weights of similarity calculation based on the feedback results, continuously update the knowledge graph, and improve the accuracy of system recommendations.

2. The method according to claim 1, characterized in that, The intelligent block division step includes: Set the minimum and maximum target block sizes; Define the priority of delimiters, including strong breakpoints, secondary breakpoints, and weak breakpoints; Find the best breakpoint within the ideal block range according to priority order; If the backtracking search fails, search forward for the breakpoint within the allowed overflow range; When all breakpoint searches fail, truncate at the ideal location.

3. The method according to claim 1, characterized in that, The ontology analysis steps include: Identify key terms in the text, including entities such as process parameters, equipment, product characteristics, and quality indicators; Analyze the relationships between terms to determine which terms are related; Extract the relationship description between each pair of related terms; The extracted results are organized into triplets containing source nodes, target nodes, and relation descriptions.

4. The method according to claim 1, characterized in that, The knowledge graph storage steps include: Create session nodes as organizational units of the knowledge graph; The extracted entities are stored as nodes in the graph database; The relationships between entities are stored as edges in a graph database; Add descriptive attributes to each relation edge, including relation type and relation description; Establish the attribution relationship between entity nodes and session nodes.

5. The method according to claim 1, characterized in that, The vector embedding model steps include: Based on a pre-trained text vector model, entity names and entity descriptions are vector-encoded. The generated vector representation is stored as an attribute of the entity node; Build vector indexes to support efficient similarity retrieval.

6. The method according to claim 1, characterized in that, The similarity retrieval steps include: Transform user queries into vector representations; Calculate the similarity between the query vector and the entity vectors in the knowledge base; Filter the most relevant entities based on similarity thresholds; Sort the results by similarity.

7. The method according to claim 1, characterized in that, The step of constructing the context-related subgraph includes: Starting with the matching entity node, explore adjacent nodes in the graph database; The boundaries of the subgraph are determined according to the preset context depth parameters; Extract all nodes and relationships from the subgraph; Sort the elements in the subgraph according to their relevance to the query.

8. The method according to claim 1, characterized in that, The process parameter optimization steps include: Analyze the entity relationships in the subgraph to extract causal relationships and parameter influence relationships; Based on the correlation rules between process parameters, parameter combination schemes are generated; The effectiveness of parameter combinations is evaluated based on quality indicators from historical data; Generate optimization suggestions that include recommended parameters and expected results.

9. The method according to claim 1, characterized in that, The feedback update step includes: Record users' adoption of system-recommended parameters and the optimization effect; The weighting coefficients for similarity calculation are dynamically adjusted based on feedback results. Regularly update the knowledge graph, adding new process parameter entities and relationships; Optimize the vector representation model to improve the accuracy of semantic retrieval.

10. The application of the method according to any one of claims 1-9 in a manufacturing production optimization system, characterized in that, The application constructs a knowledge base for process parameters based on knowledge graphs, enabling intelligent recommendation, automatic optimization, and continuous quality improvement of process parameters.

Images