Mechanical part cost decision support system and method driven by industrial knowledge graph

Abstract

The invention discloses a mechanical part cost decision support system and method driven by an industrial knowledge graph, and relates to the technical field of mechanical manufacturing cost decision, and the system comprises a multi-modal data collection module which collects multi-type data of the whole process of mechanical manufacturing and processes the multi-type data to form a unified set; the multi-modal knowledge graph construction module constructs a graph and establishes an updating mechanism; the intelligent reasoning decision module inputs new part parameters to screen an optimal scheme; the dynamic optimization adjustment module collects data in real time to update a result reversely; the decision result output module generates a visual report and supports multi-format export; according to the method, manufacturing data dispersion pain points are broken, multi-source data fusion precipitation and knowledge systematization are realized, and accurate support is provided for cost decision making; meanwhile, an optimal scheme is screened through multi-algorithm cooperation and multi-target evaluation, field data are collected in real time to dynamically optimize decisions, the crossing from static estimation to dynamic optimization is achieved, the core target is effectively balanced, and the economical efficiency and the stability are improved.

Description

Technical Field

[0001] This invention relates to the field of mechanical manufacturing cost decision-making technology, specifically to an industrial knowledge graph-driven mechanical parts cost decision support system and method. Background Technology

[0002] In the modern machinery manufacturing industry, the parts production process encompasses multiple stages, including design, process planning, equipment processing, and quality inspection. Cost decisions require comprehensive consideration of various factors such as material consumption, processing time, equipment utilization, and process adaptation. The accuracy of these decisions directly impacts a company's production efficiency and market competitiveness. With the rapid development of intelligent manufacturing technology, the data generated during the manufacturing process exhibits multimodal characteristics, including unstructured process document texts and semi-structured parts drawing data, as well as structured equipment operating parameters and material consumption records. Traditional data management methods struggle to effectively integrate multi-source data. Knowledge graph technology, with its associative knowledge organization, efficient knowledge retrieval, and reasoning capabilities, is gradually becoming an important tool for knowledge management in the manufacturing industry. However, current industry solutions for cost decisions in mechanical parts have not yet fully realized the deep integration and knowledge transformation of multimodal manufacturing data, making it difficult to build a process knowledge system covering the entire process. This fails to provide sufficient support for dynamic and accurate cost decisions, making the industry's demand for knowledge-driven intelligent cost decision support technology increasingly urgent.

[0003] Traditional cost decision-making techniques for mechanical parts have several limitations and struggle to meet the refined demands of modern intelligent manufacturing. Firstly, at the data processing level, traditional methods primarily analyze single types of data, failing to effectively integrate unstructured, semi-structured, and structured multimodal data. This results in severe data fragmentation, incomplete information for cost accounting, and potential biases in the accounting basis. Secondly, in terms of knowledge management, key knowledge such as historical process parameters, cost data, and equipment adaptation experience are often scattered across different systems or documents, lacking a systematic knowledge organization and updating mechanism. This leads to low knowledge reuse rates and makes cost decisions for new parts difficult. While traditional methods rely on rapid matching of valid historical samples and experience, they suffer from low decision-making efficiency. Furthermore, in terms of decision-making logic, traditional methods often depend on human experience or a single algorithm for estimating working hours and costs, neglecting the coupling relationship between processes and equipment. They also fail to form a multi-objective comprehensive evaluation system that covers cost, efficiency, pass rate, and equipment utilization, resulting in insufficient scientific rigor and comprehensiveness in the decision-making results. In addition, traditional technologies cannot collect dynamic data from the production site in real time. When deviations occur in equipment operating status or material consumption, cost decisions cannot be adjusted in a timely manner, leading to a large gap between the estimated results and the actual production costs, which can easily cause resource waste or production plan delays. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an industrial knowledge graph-driven decision support system and method for mechanical parts cost. The system acquires data from the entire mechanical manufacturing process through a multimodal data acquisition module, and constructs a multimodal process knowledge graph after preprocessing. The intelligent reasoning decision module uses the graph to calculate the similarity between new parts and historical parts, and selects the optimal process route and cost scheme. The dynamic optimization and adjustment module collects production data in real time and adjusts process parameters. Finally, it outputs a visualized decision report containing cost details, process routes, etc., and supports export in multiple formats and historical version traceability.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On one hand, an industrial knowledge graph-driven decision support system for mechanical parts cost, the system comprising: Multimodal data acquisition module: It connects to relevant systems throughout the entire mechanical manufacturing process through standardized interfaces, collects unstructured text data, semi-structured drawing data, and structured data, filters invalid information after data verification, classifies and stores the data to form a unified data set; Multimodal knowledge graph construction module: Receives a unified dataset, merges and preprocesses it, transforms the multimodal data into a unified feature vector through a multimodal manufacturing feature fusion algorithm, combines the BERT-BiLSTM-CRF model to complete entity recognition and relationship extraction, and after knowledge fusion to eliminate conflicts and redundancies, constructs and stores a multimodal process knowledge graph, while establishing an incremental update mechanism for the graph. Intelligent reasoning and decision-making module: When the feature parameters of a new part are input, the comprehensive manufacturing similarity between the new part and historical parts in the knowledge graph is calculated by the graph-enhanced manufacturing similarity algorithm. After screening matching samples, the total processing time is calculated by the process-equipment coupled time algorithm. The total cost of candidate process routes is calculated by the multi-dimensional coupled cost algorithm. The optimal process route and cost scheme are screened by the multi-objective evaluation system, and the initial decision result is output. The specific steps for using a multi-objective evaluation system to select the optimal process route and cost scheme are as follows: First, determine the evaluation objective dimensions, including four core objectives: cost, production efficiency, yield rate, and equipment utilization rate. The weights of each objective are calculated using the analytic hierarchy process (AHP), with cost having the following weight: Production efficiency weight is The pass rate weight is Equipment utilization rate weighting is And satisfy The target data for each candidate process route are standardized, and the cost is standardized using an inverse standardization formula, the expression of which is: Production efficiency, pass rate, and equipment utilization rate are calculated using a forward standardized formula, expressed as follows: ,in, It is the first Candidate route number Standardized data of the project target For the first Candidate route number Original data of the project target, For the first The maximum value of the project target, For the first Minimum value of the project target The data is standardized; a weighted summation method is used to calculate the overall score of each candidate process route. The calculation formula is as follows: ,in, For the first The overall score of the candidate routes. For the first Project target weight, For the first Candidate route number Standardized data of the project target For cost, For production efficiency, For the pass rate, To improve equipment utilization, According to the overall score Candidate process routes are sorted in descending order, and the route with the highest score is selected as the optimal process route; if two or more candidate routes have the same overall score, the cost-standardized data is used. The highest-performing route is considered the optimal solution. If the cost standardization data remains the same, then the production efficiency standardization data is used. The highest route is considered the optimal solution. The dynamic optimization and adjustment module collects real-time data on equipment operation, material consumption, labor hours, and quality from the production site. It compares this data with baseline data to analyze deviations, generates process parameter adjustments using a manufacturing stability dynamic optimization algorithm, updates labor hours and cost accounting results in reverse, and outputs the final optimization result. The specific methods for real-time data collection from the production site are: collecting equipment operation data through industrial IoT sensors and equipment PLC interfaces; collecting material consumption data through the material management system and weighing sensors; collecting labor hours data through the MES system; and collecting quality data through quality inspection equipment and on-site detection sensors. Decision output module: Integrates the final optimization results to generate a visual decision report that includes cost details, process routes, parameter adjustment basis and implementation suggestions, and supports export in multiple formats, online sharing and historical version tracking.

[0006] Furthermore, the standardized interfaces used in the multimodal data acquisition module include RESTful API, OPCUA industrial communication protocol, and CAD software secondary development interface. The CAD software secondary development interface is compatible with SolidWorks API, UG OpenAPI, and AutoCAD ObjectARX. The data verification rules specifically include: character encoding consistency verification for unstructured text data, validity verification for blank characters accounting for no more than 5%; format integrity verification and geometric parameter dimension chain closure verification for semi-structured drawing data; field type verification for structured data, numerical range verification based on the 3σ range of enterprise historical data, and related data encoding matching verification.

[0007] Furthermore, in the multimodal knowledge graph construction module, the mathematical expression for the multimodal manufacturing feature fusion algorithm is: ,in, This is a multimodal fusion feature vector. For modal type identification, Create weights for the modes. For the first The number of feature dimensions of a modality For the first Modal first Manufacturing influencing factors of each characteristic For the first Modal first The original vector of each feature, For the first Modal first The data credibility of each feature, and satisfying .

[0008] Furthermore, the specific steps for constructing a multimodal process knowledge graph in the graph construction module are as follows: 1. Receive a unified dataset and perform standardized preprocessing on text data, semi-structured drawing data, and structured data respectively. 2. Apply a multimodal manufacturing feature fusion algorithm to transform the preprocessed multimodal data into a unified feature vector. 3. Use the BERT-BiLSTM-CRF model to perform entity recognition and relation extraction on the feature vectors and preprocessed data, forming original entity-relation pairs. 4. Perform knowledge fusion on the original entity-relation pairs to eliminate semantic conflicts and redundant data. 5. Transform the processed knowledge into entity-relation-attribute triples to construct a five-level multimodal process knowledge graph. 6. Store the multimodal process knowledge graph in Neo4j graph database. The database is configured, and an incremental update mechanism for the graph is established. This mechanism specifically involves setting update trigger conditions: the amount of new data reaches 5% of the total existing graph data, or the new data contains uncovered part types or process types. After triggering the update, incremental knowledge extraction and fusion are performed on the new data, without processing the knowledge already stored in the existing graph. The incrementally processed entity-relationship-attribute triples are stored in the Neo4j graph database using an incremental write mode. During the update process, the query and retrieval functions of the existing graph remain normal, and the graph index is automatically synchronized after the update is complete. The Neo4j graph database supports fast queries by entity type and relation type, and a scheduled data backup mechanism can be configured, with a backup cycle customizable between 1 and 7 days.

[0009] Furthermore, the fine-tuning steps of the BERT-BiLSTM-CRF model are as follows: a training set is constructed based on annotated corpus in the mechanical manufacturing field, with the ratio of training set, validation set, and test set set set to 7:2:1. The number of iterations is configured to 50 rounds, and the learning rate adopts a cosine annealing scheduling strategy with an initial learning rate set to 2e-5. The priority rule for resolving semantic conflicts in knowledge fusion is: design document annotation information > PLM system storage information > MES system feedback information > manually entered information. Redundant data filtering adopts a hash value comparison method, and different attribute descriptions of the same entity are merged using a weighted average method.

[0010] Furthermore, in the intelligent reasoning and decision-making module, the mathematical expression for the graph-enhanced manufacturing similarity algorithm is: ,in, For new parts With historical parts Overall manufacturing similarity, For feature vector similarity weights, Fuse feature vectors for new parts. For historical parts, feature vectors are fused. It is the feature vector of the new part fusion Feature vector fusion with historical parts cosine similarity, Create a set of relationships for the core of the knowledge graph. For the first Weight of class relationships For relationship matching degree, The new parts involve the first Class creates relational instances, It is the first historical part involved Class creates relational instances, The total number of core manufacturing relationships; the similarity threshold can be customized within the range of 0.6-0.95. During screening, samples are first sorted in descending order of similarity, and samples that are incompatible with the existing equipment are removed. Finally, 3-5 valid matching samples are retained. If there are not enough, the threshold range is automatically expanded by 0.05 until the quantity requirement is met.

[0011] Furthermore, in the intelligent reasoning and decision-making module, the mathematical expression for the process-equipment coupled time algorithm is: ,in, Total processing time The average baseline time for historical parts. The number of key manufacturing factors that affect working hours. The time sensitivity coefficient of the manufacturing factor. This represents the actual manufacturing factor for the new part. This represents the average manufacturing factor of historical parts. This is a function to correct equipment utilization. This represents the actual utilization rate of the equipment.

[0012] Furthermore, in the intelligent reasoning and decision-making module, the mathematical expression for the multi-dimensional coupling cost algorithm is: ,in, For total cost, For material base cost, This represents the real-time material loss rate. The cost per unit of labor hour Total processing time This refers to the equipment depreciation cost per unit of labor hour. This is the equipment load correction factor. This is the unit process switching cost coefficient. For the first The time required for each process This represents the total number of processes.

[0013] Furthermore, in the dynamic optimization adjustment module, the mathematical expression for the dynamic optimization algorithm for manufacturing stability is: ,in, This refers to the adjustment amount of process parameters. To optimize the response coefficient, To determine the actual cost of production in real time, To predict the total cost, To ensure manufacturing stability, process parameter adjustments include machining speed, feed rate, depth of cut, tool selection, and process sequence optimization. The adjustment step size is 5%-10% of the initial parameters, and each adjustment must meet the rated parameter range of the equipment and process quality requirements.

[0014] On the other hand, there is an industrial knowledge graph-driven decision support system and method for cost control of mechanical parts. The specific steps of this method are as follows: S1, Multimodal Data Acquisition: Through standardized interfaces, it connects to relevant systems throughout the entire mechanical manufacturing process to collect unstructured text data, semi-structured drawing data, and structured data. After data verification and filtering out invalid information, the data is classified and stored to form a unified data set. S2, Multimodal Knowledge Graph Construction: Receive a unified dataset and perform preprocessing. Transform the multimodal data into a unified feature vector using a multimodal manufacturing feature fusion algorithm. Combine the BERT-BiLSTM-CRF model to complete entity recognition and relation extraction. After knowledge fusion to eliminate conflicts and redundancies, construct a multimodal process knowledge graph and store it in the Neo4j graph database. At the same time, establish an incremental update mechanism for the graph. S3, Intelligent Reasoning and Decision Making: Input the feature parameters of the new part, calculate the comprehensive manufacturing similarity between the new part and historical parts in the knowledge graph through the graph-enhanced manufacturing similarity algorithm, screen 3-5 valid matching samples, calculate the total processing time through the process-equipment coupled time algorithm, calculate the total cost of the candidate process route through the multi-dimensional coupled cost algorithm, and then screen the optimal process route and cost scheme through the multi-objective evaluation system, and output the initial decision result; S4, Dynamic Optimization and Adjustment: Real-time collection of equipment operation, material consumption, working hours and quality data at the production site, comparison and analysis of deviations with benchmark data, generation of process parameter adjustment through manufacturing stability dynamic optimization algorithm, reverse update of working hours and cost accounting results, and output of final optimization results; S5, Decision Output: Integrates the final optimization results to generate a visual decision report that includes cost details, process routes, parameter adjustment basis, and implementation suggestions. It supports export in multiple formats, online sharing, and historical version tracking.

[0015] Compared with existing technologies, this industrial knowledge graph-driven decision support system and method for mechanical parts cost has the following advantages: I. This invention connects to relevant systems throughout the entire mechanical manufacturing process through standardized interfaces, integrating unstructured text, semi-structured drawings, and various types of structured data. After verification, filtering, and classification, a unified dataset is formed. Then, a multimodal manufacturing feature fusion algorithm is used to transform the data into a unified feature vector. Combined with entity recognition, relation extraction, and knowledge fusion, a multimodal process knowledge graph is constructed, and an incremental update mechanism is established. This breaks through the industry pain points of scattered and heterogeneous data throughout the manufacturing process, achieving deep integration and structured accumulation of multi-source data. This allows scattered manufacturing knowledge to form a reusable and scalable knowledge system, significantly improving data utilization efficiency and knowledge reuse value. It provides comprehensive and accurate knowledge support for cost decisions, avoids decision-making biases caused by data fragmentation, and promotes the transformation of manufacturing knowledge from scattered accumulation to systematic management.

[0016] Second, this invention uses a graph-enhanced manufacturing similarity algorithm to accurately match historical part samples, combines a process-equipment coupled time algorithm and a multi-dimensional coupled cost algorithm to calculate core indicators, and then uses a multi-objective evaluation system to comprehensively weigh various key factors to select the optimal solution. At the same time, it relies on the Industrial Internet of Things and various production systems to collect on-site operation data in real time, and uses a manufacturing stability dynamic optimization algorithm to generate process parameter adjustment amounts and update the decision results in reverse. This realizes the leap from static estimation to dynamic optimization of cost decision-making. It not only ensures the scientificity and accuracy of decision-making through multi-algorithm collaboration and multi-objective comprehensive evaluation, but also responds to dynamic changes in the production site in real time, corrects deviations in a timely manner, and effectively balances the core objectives of cost, production efficiency, pass rate and equipment utilization rate. It significantly improves the economy and stability of the manufacturing process, provides enterprises with more practical decision support, and helps manufacturing enterprises achieve refined cost control and efficient production operation.

[0017] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art 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.

[0019] Figure 1 Flowchart of a module for a mechanical parts cost decision support system driven by industrial knowledge graph; Figure 2 A flowchart illustrating the steps of a knowledge graph-driven method for supporting cost decisions in mechanical parts. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] Example 1: Cost decisions and process optimization scenarios for mass production of precision gears.

[0022] Multimodal data acquisition module: By connecting to enterprise PLM systems via RESTful API interfaces, production equipment control systems via OPCUA industrial communication protocol interfaces, and design software via CAD software secondary development interfaces, data related to the entire precision gear manufacturing process is collected. Unstructured text data includes gear design specifications and process technical specifications; semi-structured drawing data includes 3D gear models and 2D engineering drawings; and structured data involves gear module, number of teeth, precision grade, material type, equipment operating parameters, and historical processing times. Multi-dimensional validation is performed on the collected data: ensuring consistent character encoding in text data and that whitespace characters do not exceed 5% to avoid affecting subsequent text parsing due to encoding confusion or invalid characters; verifying the completeness of drawing formats and the closure of geometric parameter dimension chains to ensure that the drawing data accurately reflects the gear structural characteristics; and verifying the correctness of structured data field types, that values ​​are within the 3σ range of historical enterprise data, and that associated data encoding matches, filtering data exceeding reasonable ranges or with incorrect associations. Finally, the data is categorized and stored to form a unified data set, providing comprehensive and accurate basic data support for subsequent knowledge graph construction. Figure 1 As shown.

[0023] Multimodal knowledge graph construction module: After receiving the unified dataset, standardization preprocessing is performed on the text, semi-structured drawings, and structured data respectively to eliminate format differences and inconsistencies in representation among different data types. A multimodal manufacturing feature fusion algorithm is then applied to transform the preprocessed three modalities into a unified feature vector. The mathematical expression of the multimodal manufacturing feature fusion algorithm is as follows: ,in, This is a multimodal fusion feature vector. For modal type identification, Create weights for the modes. For the first The number of feature dimensions of a modality For the first Modal first Manufacturing influencing factors of each characteristic For the first Modal first The original vector of each feature, For the first Modal first The data credibility of each feature, and satisfying This approach breaks down barriers between multimodal data, enabling effective integration of data features. The BERT-BiLSTM-CRF model is used to perform entity recognition and relation extraction on feature vectors and preprocessed data, accurately identifying core entities such as "gear," "gear hobbing machine," "heat treatment process," "module," and "machining time," as well as relationships between entities such as "adopted," "required," and "associated," forming original entity-relationship pairs. Semantic conflicts are resolved according to a priority rule of "design document annotation information > PLM system stored information > MES system feedback information > manually entered information," ensuring the authority of knowledge representation. A hash value comparison method is used to filter redundant data, reducing the storage pressure on the graph. Weighted average methods are used to merge different attribute descriptions of the same entity, ensuring the completeness and accuracy of entity attributes. The processed knowledge is transformed into entity-relationship-attribute triples, constructing a multimodal process knowledge graph containing five levels: parts, processes, equipment, parameters, and costs. This graph is stored in the Neo4j graph database, leveraging its efficient graph structure query capabilities to meet subsequent similarity calculation and process retrieval needs. Simultaneously, an incremental update mechanism for the atlas is established. An update is triggered when the amount of new data reaches 5% of the total existing atlas data or when the new data includes gear types or process types that are not covered. During the update, incremental knowledge extraction and fusion are performed only on the new data to avoid duplicate processing of existing data. This ensures that the existing atlas query and call functions remain normal. After the update, the atlas index is automatically synchronized to ensure that the atlas can reflect the latest manufacturing knowledge in real time.

[0024] Intelligent reasoning and decision-making module: After inputting the feature parameters of the new batch of precision gears, the comprehensive manufacturing similarity between the new gear and historical gears in the knowledge graph is calculated using a graph-enhanced manufacturing similarity algorithm. The mathematical expression for the graph-enhanced manufacturing similarity algorithm is as follows: ,in, For new parts With historical parts Overall manufacturing similarity, For feature vector similarity weights, Fuse feature vectors for new parts. For historical parts, feature vectors are fused. It is the feature vector of the new part fusion Feature vector fusion with historical parts cosine similarity, Create a set of relationships for the core of the knowledge graph. For the first Weight of class relationships For relationship matching degree, The new parts involve the first Class creates relational instances, It is the first historical part involved Class creates relational instances, The total number of core manufacturing relationships is used. The similarity threshold is customizable within the range of 0.6-0.95. During screening, samples are first sorted in descending order of similarity, and those incompatible with existing equipment are removed. 3-5 valid matching samples are retained. If insufficient, the threshold range is automatically expanded by 0.05 until the quantity requirement is met. The similarity of feature vectors and the matching degree of core manufacturing relationships are combined to comprehensively measure the manufacturing correlation between new and old parts. The initial similarity threshold is set to 0.8. After sorting in descending order of similarity, samples incompatible with existing gear hobbing and grinding machines in the current workshop are removed, ultimately retaining 4 valid matching samples to ensure the feasibility of subsequent process reference cases. A process-equipment coupled time algorithm is adopted, combining historical average gear baseline time, key manufacturing factors affecting time, time sensitivity coefficients of manufacturing factors, and current equipment utilization. It fully considers the impact of equipment status and manufacturing parameters on time, accurately calculating the total processing time of the new gear. The mathematical expression of the process-equipment coupled time algorithm is: ,in, Total processing time The average baseline time for historical parts. The number of key manufacturing factors that affect working hours. The time sensitivity coefficient of the manufacturing factor. This represents the actual manufacturing factor for the new part. This represents the average manufacturing factor of historical parts. This is a function to correct equipment utilization. This represents the actual utilization rate of the equipment. Through a multi-dimensional coupled cost algorithm, it integrates basic material costs, real-time material loss rate, unit processing cost, total processing time, unit equipment depreciation cost, equipment load correction coefficient, unit process switching cost coefficient, and total number of processes. This comprehensively covers the cost components of materials, processing, equipment, and process switching, calculating the total cost of four candidate process routes. The mathematical expression of the multi-dimensional coupled cost algorithm is: ,in, For total cost, For material base cost, This represents the real-time material loss rate. The cost per unit of labor hour Total processing time This refers to the equipment depreciation cost per unit of labor hour. This is the equipment load correction factor. This is the unit process switching cost coefficient. For the first The time required for each process The total number of processes. A multi-objective evaluation system is used to select the optimal solution: the weights of cost, production efficiency, pass rate, and equipment utilization rate are determined using the analytic hierarchy process (AHP) to ensure that the importance of each evaluation dimension aligns with the company's production goals; the target data for each candidate route are standardized to eliminate dimensional differences between different indicators; cost is standardized using an inverse standardization formula, expressed as: Production efficiency, pass rate, and equipment utilization rate are calculated using a forward standardized formula, expressed as follows: ,in, It is the first Candidate route number Standardized data of the project target For the first Candidate route number Original data of the project target, For the first The maximum value of the project target, For the first Minimum value of the project target The data is standardized; the comprehensive score is calculated using a weighted summation method, and the formula is as follows: ,in, For the first The overall score of the candidate routes. For the first Project target weight, For the first Candidate route number Standardized data of the project target For cost, For production efficiency, For the pass rate, To improve equipment utilization, According to the overall score Candidate process routes are sorted in descending order, and the route with the highest score is selected as the optimal process route; if two or more candidate routes have the same overall score, the cost-standardized data is used. The highest-performing route is considered the optimal solution. If the cost standardization data remains the same, then the production efficiency standardization data is used. The highest route is taken as the optimal solution, and the initial decision result is output to provide a data-driven basis for the selection of production solutions.

[0025] Dynamic optimization and adjustment module: Real-time operational data from gear hobbing and grinding machines is collected via industrial IoT sensors and equipment PLC interfaces to accurately monitor equipment status. Consumption data for 20CrMnTi raw materials is collected through a material management system and weighing sensors to understand material usage. Actual processing time consumption data is collected through the MES system to compare planned and actual time differences. Quality inspection equipment and on-site sensors collect quality data such as gear tooth profile and direction accuracy to monitor product quality. By comparing this real-time data with benchmark data, deviations such as a slightly higher material loss rate and extended grinding time are identified, pinpointing the problems in the production process. A manufacturing stability dynamic optimization algorithm is applied to generate process parameter adjustments. The mathematical expression for this algorithm is: ,in, This refers to the adjustment amount of process parameters. To optimize the response coefficient, To determine the actual cost of production in real time, To predict the total cost, To improve manufacturing stability, process parameter adjustments include machining speed, feed rate, depth of cut, tool selection, and process sequence optimization. The adjustment step size is 5%-10% of the initial parameters. Each adjustment must meet the equipment's rated parameter range and process quality requirements. Adjustments to machining speed, feed rate, and process sequence are controlled within 5%-10% of the initial parameters, strictly adhering to the equipment's rated parameter range and gear machining quality requirements to avoid over-adjustment leading to equipment failure or product scrap. Based on the adjustment amount, the total machining time and total cost calculation results are updated in reverse, outputting the final optimized result. This ensures that costs and process solutions are adapted to real-time production status, improving production economy and stability.

[0026] Decision result output module: The final optimization results are integrated to generate a visualized decision-making report. This report clearly presents the cost details of the new batch of precision gears, helping production and finance departments understand the cost structure; it marks the optimal process route, providing clear process guidance for workshop production; it explains the basis for parameter adjustments, making the adjustments more convincing; and it provides implementation suggestions, directly guiding on-site operations. The report supports export to PDF and Excel formats, adapting to the data usage habits of different departments; it can be shared online through the company's internal system, promoting collaboration among design, production, and finance departments; and it supports historical version tracking, facilitating subsequent queries about the cost decision-making process for this batch of gears and providing a reference for cost control of similar products.

[0027] In summary, in the mass production of precision gears, the system integrates data from the entire process through a multimodal data acquisition interface, forming a precise dataset after verification. It then constructs a knowledge graph based on a multimodal manufacturing feature fusion algorithm and a BERT-BiLSTM-CRF model, combined with an incremental update mechanism to ensure data timeliness. Subsequently, it uses graph-enhanced manufacturing similarity algorithms to screen and match samples, calculate labor hours and costs, determine the optimal process through a multi-objective evaluation system, dynamically optimize modules to adapt to real-time production deviations, and finally generate a visual report. The entire process covers the entire chain from data acquisition to decision output, supported by key algorithms, enabling precise cost decision-making and process optimization in gear mass production, improving production economy and collaborative efficiency.

[0028] Example 2: Cost decisions and process optimization scenarios for customized engine block production.

[0029] S1, Multimodal Data Acquisition: By connecting to enterprise ERP systems via RESTful API interfaces, workshop automated production lines via OPCUA industrial communication protocol interfaces, and 3D design software via CAD software secondary development interfaces, data from the entire process of customized engine block production is collected. Unstructured text data includes cylinder block customization requirement documents, technical briefing records, and quality inspection reports, accurately capturing customization requirements and quality standards. Semi-structured drawing data covers 3D assembly drawings and cross-sectional drawings of key components, detailing the structural features of the customized cylinder block. Structured data includes cylinder diameter, number of cylinders, wall thickness, materials, processing equipment models, historical production pass rates, and material consumption quotas, quantifying key parameters for customized production. Strict validation is performed on the collected data: ensuring consistent character encoding and a compliant proportion of whitespace characters in text data to avoid confusing customization requirements; verifying the completeness of drawing formats and the closure of geometric parameter dimensional chains to ensure accurate conversion of the customized cylinder block's structural design into production data; checking that structured data field types are correct, values ​​are within the enterprise's historical 3σ range, and related data codes match, filtering out abnormal data that does not meet customized production requirements, and finally classifying and storing the data to form a unified data set. This provides targeted data support for knowledge graph construction in customized scenarios, such as... Figure 2 As shown.

[0030] S2, Multimodal Knowledge Graph Construction: After receiving the unified dataset, standardized preprocessing is performed on the text, semi-structured drawings, and structured data, respectively, to eliminate special formatting and expression differences in the customized production data and ensure data consistency, taking into account the characteristics of customized production. A multimodal manufacturing feature fusion algorithm is then used to transform the preprocessed multimodal data into a unified feature vector. The mathematical expression of the multimodal manufacturing feature fusion algorithm is as follows: This approach achieves cross-modal integration of customized data, highlighting the feature weights of customized parameters. A BERT-BiLSTM-CRF model is used for entity recognition and relation extraction, focusing on identifying entities closely related to customized production, such as "engine cylinder block," "CNC machining center," "die-casting process," "cylinder diameter," and "yield rate," as well as relationships between entities such as "dependency," "containment," and "influence," forming original entity-relationship pairs. Conflicting data is processed according to pre-defined semantic conflict resolution priority rules to ensure the accuracy and authority of customized knowledge; hash value comparison is used to filter redundant information, reducing duplicate customized case data in the graph; different attribute descriptions of the same entity are merged to improve the attribute information of entities related to customized cylinder blocks. The processed knowledge is transformed into entity-relationship-attribute triples, constructing a multimodal process knowledge graph containing five levels: parts, processes, equipment, parameters, and costs. This graph is stored in the Neo4j graph database, leveraging its graph structure advantages to quickly retrieve historical process and cost data required for customized production. Configure an incremental update mechanism for the atlas. An update is triggered when the amount of new data reaches 5% of the total existing atlas data or when a new cylinder block customization type or an uncovered processing technology appears. Only the new data is incrementally extracted and integrated, and stored in the database in an incremental write mode. The update process does not affect the query and use of the existing atlas. The index is automatically synchronized after the update is completed. At the same time, a scheduled data backup is configured every 3 days to prevent the loss of customized production-related knowledge data and ensure the long-term stable availability of the atlas.

[0031] S3, Intelligent Reasoning and Decision Making: Input the feature parameters of the customized engine block, and calculate the comprehensive manufacturing similarity between the customized block and historical customized blocks in the knowledge graph using a graph-enhanced manufacturing similarity algorithm. The mathematical expression of the graph-enhanced manufacturing similarity algorithm is as follows: The similarity threshold is customizable within the range of 0.6-0.95. During screening, samples are first sorted in descending order of similarity, and those incompatible with existing equipment are removed. 3-5 valid matching samples are retained. If insufficient, the threshold range is automatically expanded by 0.05 until the quantity requirement is met. Combining the similarity of the feature vectors of customized parameters with the matching degree of the core relationship of customized production, the most valuable historical cases are selected. The initial similarity threshold is set to 0.75. After sorting in descending order of similarity, samples incompatible with the current workshop's die-casting equipment and CNC machining center are removed, ultimately retaining 3 valid matching samples to ensure that the reference cases are suitable for existing production conditions. A process-equipment coupled time algorithm is adopted, combining the average baseline time of similar customized cylinder blocks in history, key manufacturing factors, the time sensitivity coefficient of manufacturing factors, and the actual utilization rate of current equipment. The impact of customized parameters on processing time is fully considered to accurately calculate the total processing time of customized cylinder blocks. The mathematical expression of the process-equipment coupled time algorithm is: Through a multi-dimensional coupled cost algorithm, the basic cost of 6061 aluminum alloy raw materials, real-time material loss rate, unit processing cost, total processing time, unit equipment depreciation cost, equipment load correction coefficient, unit process switching cost coefficient, and total number of processes are integrated to comprehensively cover various cost items in customized production. The total cost of three candidate process routes is calculated. The mathematical expression of the multi-dimensional coupled cost algorithm is as follows: The optimal solution is selected using a multi-objective evaluation system: the weights of cost, production efficiency, pass rate, and equipment utilization rate are determined through the analytic hierarchy process (AHP) to balance the economic and efficiency requirements of customized production; the target data of each candidate route are standardized, and the cost is calculated using an inverse standardization formula, expressed as: Production efficiency, pass rate, and equipment utilization rate are calculated using a forward standardized formula, expressed as follows: To eliminate the dimensional differences between indicators, a weighted summation method is used to calculate the comprehensive score, and the calculation formula is as follows: According to the overall score Candidate process routes are sorted in descending order, and the route with the highest score is selected as the optimal process route; if two or more candidate routes have the same overall score, the cost-standardized data is used. The highest-performing route is considered the optimal solution. If the cost standardization data remains the same, then the production efficiency standardization data is used. The highest route is taken as the optimal solution, and the initial decision result is output to provide an optimal solution reference for the production of customized cylinder blocks.

[0032] S4, Dynamic Optimization and Adjustment: Real-time monitoring of key equipment in customized production is achieved by collecting operational data from die-casting machines and CNC machining centers via industrial IoT sensors and equipment PLC interfaces. This includes real-time monitoring of raw material consumption data for 6061 aluminum alloy through a material management system and weighing sensors to promptly identify potential material waste during customized production. The MES system collects actual man-hour consumption data for each process, comparing planned and actual man-hours. Quality inspection equipment and on-site sensors collect quality data such as cylinder wall thickness tolerance and cylinder bore roundness to ensure customized cylinders meet quality standards. Comparison of real-time data with baseline data reveals deviations such as high material loss rates in the die-casting process and insufficient equipment utilization in the finishing process, clarifying directions for production optimization. A manufacturing stability dynamic optimization algorithm is applied to generate process parameter adjustments. The mathematical expression for this algorithm is: The adjustments include die-casting pressure, cutting depth, tool selection, and process sequence. The adjustment step size is controlled within 5%-10% of the initial parameters and must comply with the equipment's rated parameters and cylinder block machining quality requirements to avoid affecting the precision of the customized product. Based on the adjustment amount, the total machining time and total cost calculation results are updated in reverse, outputting the final optimized result. This allows the cost and process plan of customized production to dynamically adapt to the actual on-site conditions, improving the economy and efficiency of customized production.

[0033] S5, Decision Result Output: The final optimization results are integrated to generate a visualized decision report. This report details the cost breakdown of the customized engine block, helping companies clearly understand the cost structure of customized projects; it clarifies the optimal process route, providing specific process guidance for customized production in the workshop; it explains the basis for parameter adjustments, ensuring the rationality of these adjustments is traceable; and it provides implementation suggestions, directly guiding on-site operations in customized production. The report supports exporting to Word and PDF formats, facilitating sharing between the customized project team and clients, as well as internal departments; it can be shared online through enterprise collaboration platforms, promoting collaboration among design, production, and procurement departments; and it supports historical version tracking, allowing for subsequent review of the cost decisions and process optimization process for this customized engine block, providing experience and reference for similar customized projects in the future.

[0034] In summary, for customized engine block production, the system collects and rigorously verifies relevant multimodal data. It integrates this data using a multimodal manufacturing feature fusion algorithm to construct a knowledge graph containing customized entities, and ensures data security and availability through regular backups and incremental updates. Based on customization parameters, it uses graph-enhanced manufacturing similarity algorithms to select suitable samples, calculate labor costs, and determine the optimal solution through multi-objective evaluation. It dynamically optimizes and addresses issues of loss and equipment utilization in customized production, ultimately outputting a report tailored to the customized requirements. The system is fully integrated with the characteristics of customization, achieving precise control over the cost and process of customized engine blocks, and accumulating experience for similar customized projects.

[0035] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An industrial knowledge graph-driven decision support system for mechanical parts cost, characterized in that, The system includes: Multimodal data acquisition module: It connects to relevant systems throughout the entire mechanical manufacturing process through standardized interfaces, collects unstructured text data, semi-structured drawing data, and structured data, filters invalid information after data verification, classifies and stores the data to form a unified data set; Multimodal knowledge graph construction module: Receives a unified dataset, merges and preprocesses it, transforms the multimodal data into a unified feature vector through a multimodal manufacturing feature fusion algorithm, combines the BERT-BiLSTM-CRF model to complete entity recognition and relationship extraction, and after knowledge fusion to eliminate conflicts and redundancies, constructs and stores a multimodal process knowledge graph, while establishing an incremental update mechanism for the graph. Intelligent reasoning and decision-making module: When the feature parameters of a new part are input, the comprehensive manufacturing similarity between the new part and historical parts in the knowledge graph is calculated by the graph-enhanced manufacturing similarity algorithm. After screening matching samples, the total processing time is calculated by the process-equipment coupled time algorithm. The total cost of candidate process routes is calculated by the multi-dimensional coupled cost algorithm. The optimal process route and cost scheme are screened by the multi-objective evaluation system, and the initial decision result is output. Dynamic optimization and adjustment module: Real-time collection of equipment operation, material consumption, working hours and quality data at the production site, comparison and analysis of deviations with benchmark data, generation of process parameter adjustment amounts through manufacturing stability dynamic optimization algorithm, reverse update of working hours and cost accounting results, and output of final optimization results; Decision output module: Integrates the final optimization results to generate a visual decision report that includes cost details, process routes, parameter adjustment basis and implementation suggestions, and supports export in multiple formats, online sharing and historical version tracking.

2. The industrial knowledge graph-driven mechanical parts cost decision support system according to claim 1, characterized in that, The multimodal data acquisition module employs standardized interfaces including RESTful API, OPCUA industrial communication protocol, and CAD software secondary development interface. The CAD software secondary development interface is compatible with SolidWorks' API, UG's OpenAPI, and AutoCAD's ObjectARX. Data verification rules specifically include: character encoding consistency verification for unstructured text data, validity verification for blank characters not exceeding 5%; format integrity verification and geometric parameter dimension chain closure verification for semi-structured drawing data; field type verification for structured data, numerical range verification based on the 3σ range of historical enterprise data, and related data encoding matching verification.

3. The industrial knowledge graph-driven mechanical parts cost decision support system according to claim 1, characterized in that, In the multimodal knowledge graph construction module, the mathematical expression of the multimodal manufacturing feature fusion algorithm is: ,in, This is a multimodal fusion feature vector. For modal type identification, Create weights for the modes. For the first The number of feature dimensions of a modality For the first Modal first Manufacturing influencing factors of each characteristic For the first Modal first The original vector of each feature, For the first Modal first The data credibility of each feature, and satisfying .

4. The industrial knowledge graph-driven mechanical parts cost decision support system according to claim 1, characterized in that, The specific steps for constructing a multimodal process knowledge graph in the graph construction module are as follows: A unified dataset is received, and standardized preprocessing is performed on the text data, semi-structured drawing data, and structured data respectively; a multimodal manufacturing feature fusion algorithm is applied to transform the preprocessed multimodal data into a unified feature vector; the BERT-BiLSTM-CRF model is used to perform entity recognition and relation extraction on the feature vectors and preprocessed data to form original entity-relation pairs; knowledge fusion is performed on the original entity-relation pairs to eliminate semantic conflicts and redundant data; the processed knowledge is transformed into entity-relation-attribute triples to construct a five-level multimodal process knowledge graph; the multimodal process knowledge graph is stored in the Neo4j graph database, and an incremental update mechanism for the graph is established.

5. The industrial knowledge graph-driven mechanical parts cost decision support system according to claim 4, characterized in that, The fine-tuning steps of the BERT-BiLSTM-CRF model are as follows: a training set is constructed based on annotated corpus in the mechanical manufacturing field, with the ratio of training set, validation set, and test set set set to 7:2:1, the number of iterations configured to 50 rounds, and the learning rate adopts a cosine annealing scheduling strategy, with the initial learning rate set to 2e-5. The priority rule for resolving semantic conflicts in knowledge fusion is: design document annotation information > PLM system storage information > MES system feedback information > manually entered information. Redundant data filtering adopts a hash value comparison method, and different attribute descriptions of the same entity are merged using a weighted average method.

6. The industrial knowledge graph-driven mechanical parts cost decision support system according to claim 1, characterized in that, In the intelligent reasoning and decision-making module, the mathematical expression for the graph-enhanced manufacturing similarity algorithm is: ,in, For new parts With historical parts Overall manufacturing similarity, For feature vector similarity weights, Fuse feature vectors for new parts. For historical parts, feature vectors are fused. It is the feature vector of the new part fusion Feature vector fusion with historical parts cosine similarity, Create a set of relationships for the core of the knowledge graph. For the first Weight of class relationships For relationship matching degree, The new parts involve the first Class creates relational instances, It is the first historical part involved Class creates relational instances, The total number of core manufacturing relationships.

7. The industrial knowledge graph-driven mechanical parts cost decision support system according to claim 1, characterized in that, In the intelligent reasoning and decision-making module, the mathematical expression of the process-equipment coupled time algorithm is: ,in, Total processing time The average baseline time for historical parts. The number of key manufacturing factors that affect working hours. The time sensitivity coefficient of the manufacturing factor. This represents the actual manufacturing factor for the new part. This represents the average manufacturing factor of historical parts. This is a function to correct equipment utilization. This represents the actual utilization rate of the equipment.

8. The industrial knowledge graph-driven mechanical parts cost decision support system according to claim 1, characterized in that, In the intelligent reasoning and decision-making module, the mathematical expression of the multi-dimensional coupling cost algorithm is: ,in, For total cost, For material base cost, This represents the real-time material loss rate. The cost per unit of labor hour Total processing time This refers to the equipment depreciation cost per unit of labor hour. This is the equipment load correction factor. This is the unit process switching cost coefficient. For the first The time required for each process This represents the total number of processes.

9. The industrial knowledge graph-driven mechanical parts cost decision support system according to claim 1, characterized in that, In the dynamic optimization adjustment module, the mathematical expression for the dynamic optimization algorithm for manufacturing stability is: ,in, This refers to the adjustment amount of process parameters. To optimize the response coefficient, To determine the actual cost of production in real time, To predict the total cost, To manufacture the stability coefficient.

10. A method for cost decision support of mechanical parts driven by industrial knowledge graph, the method being used to prepare a cost decision support system for mechanical parts driven by industrial knowledge graph as described in any one of claims 1-9, characterized in that, The specific steps of this method are as follows: S1, Multimodal Data Acquisition: Through standardized interfaces, it connects to relevant systems throughout the entire mechanical manufacturing process to collect unstructured text data, semi-structured drawing data, and structured data. After data verification and filtering out invalid information, the data is classified and stored to form a unified data set. S2, Multimodal Knowledge Graph Construction: Receive a unified dataset and perform preprocessing. Transform the multimodal data into a unified feature vector using a multimodal manufacturing feature fusion algorithm. Combine the BERT-BiLSTM-CRF model to complete entity recognition and relation extraction. After knowledge fusion to eliminate conflicts and redundancies, construct a multimodal process knowledge graph and store it in the Neo4j graph database. At the same time, establish an incremental update mechanism for the graph. S3, Intelligent Reasoning and Decision Making: Input the feature parameters of the new part, calculate the comprehensive manufacturing similarity between the new part and historical parts in the knowledge graph through the graph-enhanced manufacturing similarity algorithm, screen 3-5 valid matching samples, calculate the total processing time through the process-equipment coupled time algorithm, calculate the total cost of the candidate process route through the multi-dimensional coupled cost algorithm, and then screen the optimal process route and cost scheme through the multi-objective evaluation system, and output the initial decision result; S4, Dynamic Optimization and Adjustment: Real-time collection of equipment operation, material consumption, working hours and quality data at the production site, comparison and analysis of deviations with benchmark data, generation of process parameter adjustment through manufacturing stability dynamic optimization algorithm, reverse update of working hours and cost accounting results, and output of final optimization results; S5, Decision Output: Integrates the final optimization results to generate a visual decision report that includes cost details, process routes, parameter adjustment basis, and implementation suggestions. It supports export in multiple formats, online sharing, and historical version tracking.

Images