Method for quickly converting design model into process model

Abstract

The invention provides a rapid conversion method from a design model to a process model, and belongs to the technical field of industrial design, and the method comprises the steps: building a three-level conversion matrix to achieve the systematic conversion from the design model to process characteristics, from the design BOM to the process PBOM, and from the process PBOM to the manufacturing MBOM; calculating a three-level abnormal index matrix to carry out quality evaluation on the conversion process; constructing a first second-level cross-correlation matrix and a second third-level cross-correlation matrix to realize bidirectional tracking of design change and process change; constructing a fusion conversion compensation matrix by applying a multi-objective optimization function to realize intelligent compensation; a process feature recognition conversion network is constructed based on a graph neural network, and a conversion rule is learned from historical data. The method solves the technical problems that the process of converting the design model into the process model excessively depends on personal experience of engineers and the accuracy is not enough.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial design, and in particular relates to a method for quickly converting a design model into a process model. Background Art

[0002] In the field of product design and manufacturing, the conversion of design models into process models is a critical link between product design and actual production. Traditionally, experienced process engineers have relied on manual interpretation of design drawings and BOM data, translating them into process documentation and manufacturing instructions based on their personal knowledge and experience. In recent years, some companies have begun to implement PLM (Product Lifecycle Management) systems or MES (Manufacturing Execution Systems) to partially automate data conversion, but the core conversion logic still relies heavily on predefined conversion rules by engineers.

[0003] However, existing conversion methods have obvious shortcomings: first, the conversion process is highly dependent on the engineer's personal experience and knowledge level. The conversion results of different engineers for the same design model often vary greatly, lacking consistency and repeatability; second, it is difficult for engineers to fully consider the multidimensional correlation of design data, and they tend to ignore the mutual influence between key parameters and features, resulting in incomplete or inaccurate conversion results; third, as product complexity increases, it is difficult to efficiently process massive design data and complex process mapping relationships based on manual experience alone, resulting in low conversion efficiency and prone to errors.

[0004] Faced with increasingly complex product development and manufacturing demands, existing conversion methods that rely too heavily on engineers' personal experience struggle to guarantee conversion accuracy and fail to meet the modern manufacturing industry's demand for efficient integration of design, process, and manufacturing. A systematic, standardized conversion method is urgently needed to improve conversion accuracy. Specifically, existing technologies suffer from a technical problem: the conversion process from design models to process models relies too heavily on engineers' personal experience, resulting in insufficient accuracy. Summary of the Invention

[0005] In view of this, the present invention provides a method for quickly converting a design model to a process model, which can solve the technical problem in the prior art that the conversion process from a design model to a process model is overly dependent on the personal experience of engineers and is not accurate enough.

[0006] The present invention is implemented as follows: The present invention provides a method for rapid conversion of a design model to a process model, including: constructing a multidimensional first-level conversion matrix, hierarchically classifying the drawing model data in the design model according to process relevance, extracting key dimensional parameters, geometric features, and assembly relationship data, establishing conversion reference points, and forming a structured data set; constructing a multidimensional second-level conversion matrix, mapping the design BOM data to the process planning route after cleaning; constructing a multidimensional third-level conversion matrix, associating the process planning route with the process resource library based on the process planning route, to achieve deep conversion of the process PBOM to the manufacturing MBOM; calculating a third-level conversion anomaly index matrix, identifying and analyzing anomalies occurring during the conversion process; constructing first and second-level intercorrelation matrices and second and third-level intercorrelation matrices, to achieve a bidirectional mapping relationship between design features and process routes and a precise mapping between the process PBOM and the manufacturing MBOM; applying a multi-objective optimization function to construct a fusion conversion compensation matrix, intelligently compensate for data missing or anomalies during the conversion process, and achieve closed-loop self-optimization conversion from the design model to the process model.

[0007] Among them, the multidimensional first-level conversion matrix refers to a mathematical model that classifies design model data according to process relevance and converts it into a process-recognizable data structure, which includes the mapping relationship between design features and process features.

[0008] Among them, the process-recognizable data structure refers to the information organization form that conforms to the process system data standards after structured processing, including process feature description language, standardized expression of process parameters, standardized definition of process resources and standardized description of process operations, to ensure that the design data is correctly parsed and applied by the process system.

[0009] Among them, the multidimensional second-level conversion matrix refers to the mathematical model that converts the design BOM structure into the process PBOM structure, which includes the conversion rules for component relationship reconstruction and process flow generation.

[0010] Among them, the conversion rules for component relationship reconstruction and process flow generation refer to a set of methods for reorganizing and defining the component hierarchical relationships in the design BOM based on the process processing logic, including component processing sequence rules, component assembly relationship rules, process flow node definition rules, and process connection rules.

[0011] Among them, the multidimensional third-level conversion matrix refers to the mathematical model that converts the process PBOM into the manufacturing MBOM, which contains the specific conversion rules for process resource allocation and production execution parameter generation.

[0012] Among them, the specific conversion rules for process resource allocation and production execution parameter generation refer to the set of algorithms and strategies that map the abstract resource requirements at the process planning level to the actual available resource instances at the manufacturing site, including equipment selection rules, tooling and fixture matching rules, production line balancing rules, and manufacturing parameter concretization rules.

[0013] Among them, the three-level conversion anomaly index matrix includes: the first-level conversion anomaly index matrix, the second-level conversion anomaly index matrix and the third-level conversion anomaly index matrix, which are used to quantitatively evaluate the anomalies that occur in the conversion process from design model to process feature, the conversion process from design BOM to process PBOM, and the conversion process from process PBOM to manufacturing MBOM.

[0014] Among them, the multi-objective optimization function is used to balance multiple objectives such as conversion accuracy, conversion efficiency, resource consumption, and manufacturing risk in the process of constructing the fusion conversion compensation matrix. The input includes conversion anomaly index, resource occupancy rate, conversion time consumption, design change range, and manufacturing capability index. The output is the comprehensive optimized compensation strategy weight distribution scheme and compensation execution priority sequence.

[0015] Among them, the first and second level mutual correlation matrices refer to mathematical models for establishing bidirectional correlation relationships between design features and process routes, which are constructed using a process feature recognition conversion network model; the specific structure of the process feature recognition conversion network model is a multi-layer feature extraction and relational reasoning architecture based on a graph neural network, including a feature encoding layer, a graph structure learning layer, a relational reasoning layer and a feature mapping layer, wherein the feature encoding layer is responsible for converting design features into high-dimensional vector representations, the graph structure learning layer is responsible for learning the topological relationships between design features, the relational reasoning layer is responsible for inferring the mapping rules between design features and process features, and the feature mapping layer is responsible for mapping the learned design features to corresponding process features.

[0016] The present invention achieves systematic conversion and precise mapping of design data to process data by constructing a three-level conversion matrix, anomaly index matrix, and mutual correlation matrix, combined with a minimum spanning tree algorithm and a multi-objective optimization function. This method significantly improves conversion accuracy, transforming the implicit knowledge that traditionally relies on the engineer's personal experience into explicit mathematical models and standardized algorithmic rules, making the conversion process consistent and repeatable. By constructing a multidimensional conversion matrix, it can comprehensively capture and process the complex correlations of design data, ensuring the complete conversion of key parameters and features, and significantly reducing omissions and error rates. The introduction of anomaly index matrix and fusion conversion compensation matrix realizes self-diagnosis and intelligent error correction functions in the conversion process, maintaining a high conversion accuracy even in the case of incomplete data or conflicts.

[0017] The present invention solves the technical problem that the conversion process from design model to process model is overly dependent on the personal experience of engineers and lacks accuracy, providing scientific and reliable technical support for the efficient development of complex products, making the conversion results stable and reliable, and no longer limited by the knowledge boundaries and experience levels of individual engineers. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0020] like Figure 1 FIG. 1 is a flowchart of a method for rapidly converting a design model to a process model provided by the present invention. The method includes the following steps:

[0021] S01. Construct a multi-dimensional first-level conversion matrix, hierarchically classify the drawing model data in the design model according to process relevance, extract key dimensional parameters, geometric features, and assembly relationship data, establish conversion reference points, form a structured data set, and calculate the conversion confidence index;

[0022] S02. Construct a multi-dimensional second-level conversion matrix to map the cleaned design BOM data to the process planning route, associate and match material properties, process performance parameters, and manufacturing equipment capability parameters, and achieve the initial conversion of the design EBOM to the process PBOM.

[0023] S03. Construct a multi-dimensional third-level conversion matrix, associate it with the process planning route and the process resource library, integrate the labor quota data, material quota data, and tooling data, and realize the in-depth conversion of process PBOM to manufacturing MBOM;

[0024] S04. Calculate the first-level conversion anomaly index matrix, identify geometric feature discrepancies, dimensional tolerance violations, and material property conflicts that occur during the conversion process, establish an anomaly database, and record anomaly types, locations, and severity in a structured manner;

[0025] S05. Apply the minimum spanning tree algorithm to optimize the calculation of the second-level conversion anomaly index matrix. Treat the process nodes in the process route planning as vertices in graph theory, and the associations between processes as edge weights. Obtain the optimal process route by solving the minimum spanning tree. Simultaneously, analyze process conflicts, resource usage conflicts, and process parameter overruns to generate an anomaly index.

[0026] S06. Calculate the third-level conversion anomaly index matrix to predict and analyze assembly interference, insufficient tooling matching, and excessive working hours in the manufacturing execution process, forming an anomaly guidance library to guide process design optimization.

[0027] S07. Use the pre-trained process feature recognition conversion network to construct the first and second level intercorrelation matrix, establish a bidirectional mapping relationship between design features and process routes, and achieve forward tracking from design changes to process changes, and reverse feedback from process optimization to design improvements;

[0028] S08. Use the hierarchical association mapping algorithm to construct the second and third level inter-correlation matrices, establish multi-dimensional associations between process routes and manufacturing resource instances, achieve accurate mapping between process PBOM and manufacturing MBOM, and support real-time response of process changes to manufacturing execution;

[0029] S09. Apply multi-objective optimization functions to construct a fusion transformation compensation matrix, integrate the three-level transformation matrix and the abnormality index matrix, analyze the transformation pattern rules, establish a compensation strategy library, and intelligently compensate for data missing or abnormalities during the transformation process to achieve closed-loop self-optimization transformation from design model to process model.

[0030] The multi-dimensional first-level conversion matrix refers to a mathematical model that classifies design model data according to process relevance and converts it into a process-recognizable data structure, including a mapping relationship between design features and process features.

[0031] Among them, the process-recognizable data structure refers to the information organization form that conforms to the process system data standard after structured processing, including process feature description language, standardized expression of process parameters, standardized definition of process resources and standardized description of process operations, to ensure that the design data is correctly parsed and applied by the process system.

[0032] The multi-dimensional second-level conversion matrix refers to a mathematical model for converting a design BOM structure into a process PBOM structure, and includes conversion rules for component relationship reconstruction and process flow generation.

[0033] Among them, the conversion rules for component relationship reconstruction and process flow generation refer to a set of methods for reorganizing and defining the component hierarchical relationships in the design BOM based on the process processing logic, including component processing sequence rules, component assembly relationship rules, process flow node definition rules and process connection rules. Through the above rules, the BOM structure oriented towards design functions is transformed into a BOM structure oriented towards process manufacturing.

[0034] The multi-dimensional third-level conversion matrix refers to a mathematical model for converting a process PBOM into a manufacturing MBOM, and includes specific conversion rules for process resource allocation and production execution parameter generation.

[0035] Among them, the specific conversion rules for process resource allocation and production execution parameter generation refer to a set of algorithms and strategies that map the abstract resource requirements at the process planning level to the actual available resource instances at the manufacturing site, including equipment selection rules, tooling and fixture matching rules, production line balancing rules, and manufacturing parameter concretization rules, to ensure that process planning is accurately executed in the manufacturing environment.

[0036] Among them, the first-level conversion anomaly index matrix refers to a numerical model that quantitatively evaluates anomalies that occur during the conversion process from design model to process characteristics, and is used to measure the conversion quality.

[0037] Among them, the second-level conversion anomaly index matrix refers to a numerical model for quantitatively evaluating anomalies that occur during the conversion process from design BOM to process PBOM, and is used to identify conversion risks.

[0038] The third-level conversion anomaly index matrix refers to a numerical model for quantitatively evaluating anomalies that occur during the conversion from process PBOM to manufacturing MBOM, and is used to predict execution risks.

[0039] The first and second level correlation matrix refers to a mathematical model for establishing a bidirectional correlation between design features and process routes, supporting the collaborative management of design changes and process changes.

[0040] The second and third level inter-correlation matrices refer to mathematical models for establishing multi-dimensional correlation relationships between process routes and manufacturing resources, supporting accurate mapping between process PBOM and manufacturing MBOM.

[0041] Among them, the fusion conversion compensation matrix refers to a mathematical model of the compensation mechanism established by comprehensively analyzing the three-level conversion matrix and the abnormal index matrix data, which is used to solve the data missing or abnormal situation in the conversion process.

[0042] Among them, the minimum spanning tree algorithm refers to treating process route planning as a graph theory problem, with process nodes as vertices in the graph, and the relationship between processes as edges. The weight of the edge represents the switching cost or risk level between processes. By solving the minimum weight spanning tree, the process route combination with the lowest total cost or risk is obtained, minimizing resource consumption and process conflicts while ensuring the completion of all processes.

[0043] Among them, the multi-objective optimization function is used to balance multiple objectives such as conversion accuracy, conversion efficiency, resource consumption, and manufacturing risk in the process of constructing the fusion conversion compensation matrix. The input includes conversion anomaly index, resource occupancy rate, conversion time consumption, design change range, and manufacturing capability index. The output is the comprehensive optimized compensation strategy weight distribution scheme and compensation execution priority sequence, so as to achieve global optimization of the conversion process under multiple constraints.

[0044] Among them, the specific structure of the process feature recognition conversion network model is a multi-layer feature extraction and relational reasoning architecture based on graph neural network, which includes a feature encoding layer, a graph structure learning layer, a relational reasoning layer and a feature mapping layer, wherein the feature encoding layer is responsible for converting the design features into high-dimensional vector representations, the graph structure learning layer is responsible for learning the topological relationships between the design features, the relational reasoning layer is responsible for inferring the mapping rules between the design features and the process features, and the feature mapping layer is responsible for mapping the learned design features to the corresponding process features. The number of multi-head attention mechanism parameters in the model is jointly determined by three key parameters: the complexity of the design features, the number of process feature types and the diversity of the mapping rules; the training number of the process feature recognition conversion network model is The steps of establishing the data set include collecting historical successful design models and their corresponding process model data, extracting the correspondence between design features and process features, constructing feature mapping annotation data, establishing the version correspondence between design models and process models, and forming a training sample set for multiple fields and multiple product types; the steps of training the process feature recognition conversion network model include cleaning and standardizing the collected historical data, initializing model parameters, using supervised learning to train the model to learn the mapping rules from design features to process features, improving the model's ability to distinguish similar features through comparative learning, using domain adaptation technology to improve the model's generalization ability in different product fields, and compressing the model size through model distillation technology to improve reasoning efficiency.

[0045] Optionally, key dimensional parameters refer to dimensional data in the design model that have a significant impact on product function, performance, assembly, and manufacturing. These parameters typically include:

[0046] Functional key dimensions: dimensions that directly affect the realization of product functions, such as fit clearance, working stroke, sealing surface dimensions, etc.

[0047] Assembly critical dimensions: dimensions that affect the assembly relationship between parts, such as assembly reference planes, positioning hole diameters, connection thread sizes, etc.

[0048] Manufacturing critical dimensions: dimensions that have a decisive influence on manufacturing process selection and processing quality, such as wall thickness, processing allowance, surface roughness requirements, etc.

[0049] Tolerance-sensitive dimensions: dimensions that are particularly sensitive to tolerance changes and may affect product performance, such as bearing mating surfaces, clearances of precision moving parts, etc.

[0050] Inspection key dimensions: dimensions that need to be measured during the inspection process to assess product quality;

[0051] During the conversion process from the design model to the process model, accurately identifying and extracting these key dimensional parameters is the basis for ensuring the accuracy of the process model. Based on these key dimensional parameters, the system provides the necessary data support for subsequent process planning, tooling design, and inspection plan formulation.

[0052] The specific implementation of the above steps is described in detail below.

[0053] The specific implementation method of step S01 is to realize the primary data conversion from the design model to the process model by establishing a multidimensional space mapping structure. First, the feature hierarchical clustering algorithm is used to evaluate the process relevance of the drawing data in the design model, and the design features are divided into three levels of high, medium and low correlation categories according to the three dimensions of processing difficulty coefficient, process implementation complexity and manufacturing resource demand. The correlation thresholds are set to 0.85, 0.6 and 0.4 respectively. Then, the geometric feature recognition algorithm is used to extract key dimensional parameters, feature boundaries, shape features and assembly relationship data, including process-related parameters such as aperture, groove width, countersink depth, chamfer size, thread specification, surface roughness, etc. The conversion reference point is then determined by the least squares method, and the most stable and process-clear geometric feature in the design coordinate system is selected as the conversion anchor point to establish a coordinate mapping relationship from the design space to the process space. Subsequently, the structured feature coding method is used to organize the extracted data into a standardized data set, and a hierarchical tree structure is used to express the topological relationship between features to ensure the integrity and consistency of the data. Finally, the conversion quality is evaluated using a Bayesian confidence calculation method. This method considers three factors: feature recognition accuracy, parameter extraction completeness, and reference point stability. A confidence index between 0 and 1 is generated. A manual intervention mechanism is triggered when the confidence index falls below 0.75. This step aims to transform unstructured design drawing data into a structured parameter set that can be directly recognized by the process system, laying the data foundation for subsequent process planning.

[0054] The specific implementation of step S02 involves constructing a mapping and conversion system from the design BOM to the process PBOM. First, the design BOM is standardized through data cleaning and preprocessing, removing redundant information, completing missing items, unifying coding formats, and standardizing material descriptions to ensure data quality reaches over 95% completeness and consistency. A decision tree algorithm is then used to establish a mapping relationship between components and process routes. Based on multi-dimensional attributes such as component geometry, material properties, precision requirements, and functional characteristics, an appropriate processing route is derived, forming a preliminary process flow chart. A fuzzy matching algorithm is then used to correlate material properties with process performance parameters. The matching of specified material properties, such as hardness, strength, and heat treatment state, is evaluated with the processing capabilities of each process node. A matching threshold of 0.8 is set; items below this threshold are marked as potential risk items. Subsequently, a resource capability assessment model is used to incorporate manufacturing equipment capability parameters into the conversion process. Parameters such as the equipment's processing accuracy range, maximum processing size, and processable material types are analyzed to determine whether they meet component manufacturing requirements. Resource configurations with a compatibility score below 0.7 are recorded as not recommended. Finally, a preliminary version of the process PBOM is generated, including the component process breakdown structure, process flow definitions, initial process parameter values, and resource requirement estimates, completing the transition from a product structure perspective to a process realization perspective. This step transforms the design-oriented product structure into a manufacturing-centric process planning structure, providing a framework for subsequent manufacturing execution.

[0055] The specific implementation of step S03 involves achieving a deep mapping conversion from the process PBOM to the manufacturing MBOM. First, an association rule mining algorithm is applied to establish a precise mapping between the process planning route and the process resource library. Based on historical manufacturing data, the strength of the association between each process and specific equipment, tooling, and personnel is analyzed. The association confidence threshold is set to 0.85, and the support threshold is set to 0.3. A multi-objective resource allocation algorithm is then used to integrate the work quota data into the conversion process. Standard work hours are determined based on component complexity, batch size, and quality grade, and the impact of equipment efficiency coefficients and operator skill levels on actual work hours is considered to establish an adaptive work hour allocation model. The material quota data is then integrated using a material requirement planning algorithm to calculate material consumption, loss rate, and scrap recovery rate for each process. A material flow chain is established to achieve precise material control throughout the entire manufacturing process. A geometric constraint satisfaction algorithm is then used to associate tooling and fixture data with the corresponding process. The compatibility between the component and tooling is analyzed, including positioning reference consistency, clamping force distribution rationality, and operational ease. The compatibility score threshold is set to 0.9. Finally, a complete manufacturing MBOM is generated, including specific process implementation parameters, resource allocation examples, material consumption quotas, and quality control points, forming a directly executable manufacturing guidance document. The purpose of this step is to transform the abstract process plan into specific manufacturing execution instructions, achieving precise transfer of process design intent to manufacturing site operations.

[0056] The specific implementation of step S04 involves establishing a first-level conversion anomaly identification and management mechanism. First, a feature comparison and analysis algorithm is used to detect geometric feature conversion anomalies. Feature definitions in the design model are compared one by one with feature analysis results in the process model, and the feature deviation index is calculated. Geometric feature discrepancies are flagged when the feature recognition rate is below 0.9 or the feature analysis ambiguity is above 0.2. Statistical process control methods are then applied to analyze dimensional tolerance violations. The design tolerance range is compared with the machining accuracy supported by the process system, and tolerance satisfaction is calculated. A dimensional tolerance violation risk is determined when the satisfaction is below 0.95. Material property conflicts are then identified using a conflict detection algorithm. The compatibility of the material's chemical composition and physical properties with the process requirements is analyzed, and material compatibility is calculated. Material property conflicts are identified when the compatibility is below 0.85. A structured anomaly database is then established, using a multidimensional classification and coding system to record anomaly information, including anomaly type code, anomaly location coordinates, discovery timestamp, associated part number, impact scope, and priority. Finally, the anomaly index matrix is ​​calculated, and a weighted scoring model is constructed based on the three dimensions of anomaly frequency, anomaly severity, and anomaly impact range. Severity is categorized as fatal (1.0), severe (0.8), moderate (0.5), and minor (0.2). This provides a basis for decision-making on subsequent compensation strategies. This step monitors the quality of the conversion process from design to process features, allowing for the timely detection and management of conversion anomalies.

[0057] The specific implementation of step S05 involves optimizing the process route planning and building a second-level transition anomaly analysis system. First, Prim's algorithm is used to construct a minimum spanning tree. Each process is considered a vertex in the graph, and the switching relationships between processes are considered edges. Edge weights are calculated by weighting the switching time cost, equipment adjustment complexity, and resource utilization conflict, with the weight coefficients set to 0.4, 0.3, and 0.3, respectively. Next, critical path analysis is used to identify bottleneck processes in the process route. The earliest start time, latest completion time, and total time difference of each process are calculated. Processes with zero time difference constitute the critical path and serve as the key targets for process optimization. A conflict detection algorithm is then used to analyze process logic conflicts, examining the pre- and post-process constraints and resource sharing between parallel processes. A conflict degree exceeding 0.6 is recorded as a process conflict anomaly. A resource load balancing algorithm is then used to assess resource utilization conflicts, analyzing the rationality of the scheduling of equipment, tooling, and personnel across multiple processes. A resource conflict risk is identified when the load ratio of a resource exceeds 0.9 or the load variance coefficient exceeds 0.25. Finally, a parameter constraint verification algorithm is applied to detect process parameter out-of-limit conditions. The planned parameters, such as cutting speed, feed rate, and machining temperature, are compared with the equipment's capabilities. Parameters that exceed these limits are recorded as process parameter anomalies, and an index matrix containing all anomaly information is generated. This step aims to optimize process route planning and identify potential issues during the process PBOM conversion process, thereby improving the feasibility and efficiency of process planning.

[0058] The specific implementation of step S06 involves building an anomaly prediction and analysis system for the manufacturing execution phase. First, the assembly process is simulated using the Monte Carlo simulation method. Multiple sets of assembly state samples are generated based on the component geometric models and assembly relationships. The probability distribution of interference within these samples is analyzed, and an interference probability exceeding 0.15 is identified as a high-risk assembly relationship. A multidimensional adaptability assessment model is then applied to analyze the issue of insufficient tooling compatibility. The compatibility between tooling and components is evaluated across four dimensions: geometric shape matching, positioning accuracy, reasonable clamping force distribution, and ease of operation. Tooling with a comprehensive score below 0.8 is marked as insufficient tooling compatibility. Next, a time series prediction algorithm is used to analyze the risk of man-hour overruns. A man-hour prediction model is established based on historical production data, calculating the probability of deviation between the actual and standard working hours for each process. Processes with a deviation rate exceeding 20% ​​and a probability of occurrence greater than 0.3 are recorded as risk items for man-hour overruns. An anomaly guidance library is then constructed, and the prediction and analysis results are categorized and coded according to anomaly type, impact, and resolution difficulty. The results are then linked to historical solutions to form a closed-loop knowledge loop. Finally, the third-level conversion anomaly index matrix is ​​calculated, comprehensively considering the three dimensions of anomaly probability, impact severity, and detection difficulty to generate a risk score between 0 and 1. The risk level is divided into three intervals: high (above 0.7), medium (0.4 to 0.7), and low (below 0.4). This step is used to proactively identify potential problems in the manufacturing execution process and provide data support and guidance for process design optimization.

[0059] The specific implementation of step S07 is to establish a bidirectional tracking mapping relationship between design features and process routes. First, a pre-trained process feature recognition and conversion network model is loaded. This model adopts a graph convolutional neural network architecture and includes four core layers: feature encoding, graph structure learning, relational reasoning, and feature mapping. It effectively captures the topological structure and semantic information of design features. Then, a feature sensitivity analysis algorithm is used to identify key design features. The influence weight of each feature on process route selection is calculated. Features with an influence coefficient greater than 0.5 are marked as key features and serve as the core objects for change tracking. Next, a bidirectional graph mapping algorithm is used to establish the correspondence between design features and process route nodes. An association matrix containing four dimensions: feature ID, process ID, impact intensity, and change sensitivity is constructed to achieve forward tracking capabilities from design changes to process impacts. Subsequently, a reverse derivation algorithm is applied to establish a feedback channel from process optimization to design improvement, analyze the correlation strength between process bottlenecks and design features, generate design optimization suggestions, and provide data support for collaborative design. Finally, a complete primary and secondary correlation matrix is ​​formed to support change impact analysis, conflict detection, and optimization collaboration between the design model and process route. The purpose of this step is to break down the data barriers between the design domain and the process domain and achieve two-way collaborative optimization of design and process.

[0060] The specific implementation of step S08 involves achieving precise mapping and association between the process PBOM and the manufacturing MBOM. First, a hierarchical association mapping algorithm is applied to construct a multidimensional mapping relationship between the process route and manufacturing resources, establishing an association data structure consisting of five dimensions: process ID, resource type, resource instance ID, compatibility score, and priority. Then, using resource capability modeling, the capabilities of the manufacturing resource instances are parameterized, including equipment processing accuracy range, maximum processing size, machinable material types, and tooling fixture applicability, forming a standardized resource capability description model. Next, a multi-condition matching algorithm is used to precisely match process requirements with resource capabilities. A compatibility score is calculated and prioritized to recommend the most appropriate resource combination for each process. A change impact analysis algorithm is then used to assess the impact of process changes on manufacturing execution, calculate the change propagation path and impact level, and support change risk assessment and response strategy formulation. Finally, a complete second- and third-level intercorrelation matrix is ​​generated, enabling real-time linkage between process route adjustments and manufacturing resource allocation. The matrix refresh frequency is set to be triggered by changes in the manufacturing execution system status or no more than 10 minutes. This step establishes a data bridge between process planning and manufacturing execution, ensuring that process design intent is accurately implemented at the manufacturing site.

[0061] The specific implementation of step S09 involves constructing a self-optimizing fusion conversion compensation mechanism. First, a multi-objective genetic algorithm is used to construct a fusion conversion compensation matrix. Four optimization objectives are set: conversion accuracy, conversion efficiency, resource consumption, and manufacturing risk. Weight coefficients are 0.4, 0.2, 0.2, and 0.2, respectively. The population size is set to 200, the number of evolutionary generations is set to 500, the crossover rate is 0.85, and the mutation rate is 0.1. A pattern mining algorithm is then used to analyze historical conversion data, identifying common conversion patterns and anomaly patterns. A pattern library containing pattern IDs, trigger conditions, impact ranges, and applicable scenarios is established to provide knowledge support for conversion compensation. A decision tree algorithm is then used to construct a compensation strategy library. The optimal compensation solution is selected based on anomaly type, severity, and resource constraints. Compensation strategies include adaptive parameter adjustment, alternative resource allocation, process rescheduling, and manual intervention. A Bayesian optimization algorithm is then used to evaluate the compensation effect and conduct self-learning. Compensation strategy parameters are continuously optimized based on feedback from post-compensation conversion quality. The learning rate is set to 0.05, and the convergence threshold is set to 0.001. Finally, a closed-loop, self-optimizing conversion from the design model to the process model is implemented. The system periodically evaluates the conversion quality and concludes the conversion optimization when the cumulative optimization gain falls below 0.1%. The goal of this step is to establish a self-learning conversion compensation mechanism to continuously improve the accuracy and robustness of the model conversion.

[0062] The detailed structure of the process feature recognition conversion network model adopts a deep learning architecture based on graph neural networks. The feature encoding layer is composed of a multi-layer perceptron. The input is the original parameter vector of the design feature, which is nonlinearly transformed through a three-layer fully connected network. The number of hidden layer neurons is 256, 512, and 1024, respectively. The activation function is LeakyReLU, the random inactivation rate is set to 0.2, and the output is a 512-dimensional feature embedding vector. The graph structure learning layer adopts a graph convolutional network structure, which contains three layers of graph convolution operations. Each layer has 128, 256, and 512 convolution kernels, respectively. The aggregation function uses a weighted summation method to learn the topological relationship between feature nodes and output the spatial relationship embedding of features. The relationship reasoning layer is composed of a multi-head attention mechanism with 8 heads and 64 attention dimensions per head. The self-attention mechanism captures long-range dependencies between features to enhance the model's reasoning ability. Feature fusion is then performed through a feedforward neural network. The hidden layer dimension is 2048. The feature mapping layer uses a bidirectional long short-term memory network with a hidden state dimension of 512 and a recursive depth of 2. The design feature sequence is mapped to the process feature sequence through a sequence-to-sequence mapping mechanism, and finally the feature mapping probability distribution is output through a fully connected layer and a Softmax activation function. The total number of model parameters is approximately 10 7 The training batch size is 64 and the learning rate is 10 -4 Adam optimizer with a weight decay coefficient of 10 -5 .

[0063] The detailed steps for establishing the training dataset begin with historical data collection. Design models and corresponding process model data for at least 1,000 products successfully delivered within the past five years are extracted from the company's product data management system. These data cover representative products from multiple sectors, including machinery, electronics, automotive, and aviation. Feature mapping and annotation are then performed. A team of process experts manually annotates design and process features to establish a mapping database. The annotations include information such as feature type, feature parameters, processing methods, process schedules, and quality requirements, with annotation consistency exceeding 90%. Next, version mapping is constructed, recording the temporal correlation between design change history and process adjustment responses, forming a change propagation sample set. Data augmentation is then performed to expand the training sample through parameter perturbations, feature combination mutations, and constraint adjustments to enhance model generalization. Finally, a stratified validation set is established, partitioning the data into training, validation, and test sets based on product type, complexity, and industry characteristics, with a ratio of 7:1.5:1.5. This ensures comprehensive and reliable performance evaluation of the model across different application scenarios.

[0064] Optionally, the detailed steps of the process feature recognition conversion network model are to first clean and standardize the collected historical data, including outlier detection and elimination, missing value filling, data normalization and encoding conversion, etc., to ensure data quality. Then the model parameters are initialized, the weights are initialized using the Xavier method, and the bias terms are initialized to zero. Then the model training process begins, using supervised learning, with design features as input and process features as labels. The cross-entropy loss function is used to measure the difference between the predicted results and the true labels, and the model parameters are updated through the back-propagation algorithm. In order to enhance the model's ability to distinguish similar features, a contrastive learning strategy is introduced to construct positive and negative sample pairs so that the model learns to map similar features to similar feature spaces and different features to distant feature spaces. Subsequently, domain adaptation technology is used to introduce domain adversarial training for the differences in data distribution in different product areas. The gradient reversal layer is used to enable the model to learn domain-invariant feature representations, thereby improving its generalization ability in new product areas. Finally, model distillation technology is applied to transfer the knowledge of the trained large and complex model to a smaller model with a simpler structure. Knowledge transfer is achieved by minimizing the KL divergence of the output distributions of the two models. While maintaining performance, the number of model parameters is compressed to about 30% of the original, thereby improving inference efficiency.

[0065] The mathematical model or calculation process involved in the present invention is described in detail below.

[0066] In step S01, the construction process of the multi-dimensional first-level conversion matrix involves multiple calculation processes and matrix expressions. The calculation formula for process relevance evaluation is specifically expressed as follows:

[0067] R i =α·D i +β·C i +γ·M i ;

[0068] Where R i is the process relevance of the i-th design feature; D i is the processing difficulty coefficient, ranging from 0 to 1; C i is the complexity of process implementation, ranging from 0 to 1; M i is the manufacturing resource demand, ranging from 0 to 1; α, β, and γ are weight coefficients, and α+β+γ=1, with the default values ​​of 0.4, 0.3, and 0.3 respectively.

[0069] The parameter acquisition method is: D i The difficulty of feature processing is evaluated through expert experience scoring, including step 1: determining the basic score according to the feature type and the difficulty level table; step 2: considering the material processability adjustment coefficient; step 3: considering the accuracy requirement adjustment coefficient, and comprehensively calculating the final difficulty coefficient. iObtained through the feature complexity calculation formula: where N f N is the number of sub-features contained in the feature. max is the maximum number of sub-features in the model, T p is the estimated time for feature processing, T std is the standard working time. i Obtained through resource consumption assessment model: Among them E i is the device resource requirement, T i is the tooling resource requirement, M i is the material resource demand, w1, w2, w3 are weight coefficients.

[0070] The conversion confidence index calculation formula is specifically expressed as follows:

[0071] CI=ω1·AR+ω2·CP+ω3·BS;

[0072] Where CI is the conversion confidence index; AR is the feature recognition accuracy, ranging from 0 to 1; CP is the parameter extraction completeness, ranging from 0 to 1; BS is the reference point stability, ranging from 0 to 1; ω1, ω2, and ω3 are weight coefficients, and ω1+ω2+ω3=1, with the default values ​​of 0.4, 0.3, and 0.3 respectively.

[0073] The parameter acquisition method is: AR obtains through feature recognition evaluation: where N correct is the number of correctly identified features, N total is the total number of features. CP is obtained by parameter completeness evaluation: Among them, P extracted is the number of parameters successfully extracted, P required is the total number of required parameters. BS is obtained by benchmark stability assessment: where Δ max is the maximum offset of the reference point, T allow is the allowed deviation threshold.

[0074] The use of power and ratio relationships in these formulas primarily addresses the need for normalization in engineering practice, converting metrics of varying dimensions into dimensionless values ​​between 0 and 1 to facilitate comprehensive evaluation. In particular, the ratio relationship used in resource consumption assessment effectively reflects relative levels of consumption, avoiding biases associated with absolute value comparisons.

[0075] In step S02, the construction process of the multi-dimensional second-level conversion matrix involves multiple calculation processes. The calculation formula for the matching degree between material properties and process parameters is specifically expressed as follows:

[0076]

[0077] Where MC ij is the matching degree between material i and process node j; M ik is the kth property value of material i; P jk is the required value of the kth material property at process node j; S(M ik , P jk ) is the matching function of a single attribute; w k is the weight coefficient of the kth attribute, and n is the number of material properties considered.

[0078] The parameter acquisition method is: M ik Obtained from the material database, including the chemical composition, physical properties, mechanical properties and other data of the material. jk Obtained from the process specification library, define the process requirements for various material properties. Matching function S(M ik , P jk ) Different calculation methods are used according to the attribute type: For range attributes, in is the median of the required range, To request the range width; for discrete attributes,

[0079] The calculation formula for equipment capability adaptability evaluation is as follows:

[0080]

[0081] Where, EC ij is the adaptability of equipment i to component j; E ik is the kth capability parameter of device i; R jk is the requirement of component j for the kth capability parameter; C(E ik , R jk ) is the fitness function of a single capability; v k is the weight coefficient of the kth capability parameter, and m is the number of capability parameters considered.

[0082] The parameter acquisition method is: E ik Obtained from the equipment capability database, including processing accuracy range, maximum processing size, processing material range, etc. jk Extracted from the component process requirements. The fitness function C(E ik , R jk ) is calculated using similar range matching or discrete matching methods.

[0083] These formulas use a weighted summation approach, primarily taking into account the requirements of multi-attribute decision-making. Different weight coefficients reflect the importance of each attribute. The matching function design takes into account the tolerance and flexibility of engineering practices. Attributes within the specified range are given full marks, while those outside the specified range are deducted based on the degree of deviation. This design is more consistent with engineering practice.

[0084] In step S03, the construction of the multi-dimensional third-level conversion matrix involves multiple calculation processes. The calculation formulas for support and confidence in the association rule mining algorithm are specifically expressed as follows:

[0085]

[0086] Where, Association rules support level; Association rules confidence level; count(X∪Y) is the number of records containing both X and Y; count(X) is the number of records containing X; N is the total number of records.

[0087] The parameter acquisition method is: count the combination frequencies of each process and specific resource instances from the historical process execution data, and calculate the association rules that meet the minimum support and minimum confidence thresholds.

[0088] The standard working hours calculation formula is as follows:

[0089] T std =T base ·C c ·C b ·C q ;

[0090] Where, T std is the standard working hours; T base is the benchmark working time, which is obtained from the working time quota library according to the process type; C c is the complexity coefficient, ranging from 1 to 5; C b is the batch coefficient, C b =1-log 10 (B) 0.05, where B is the batch size; C q is the quality grade coefficient, special grade is 1.3, first grade is 1.1, second grade is 1.0, and third grade is 0.9.

[0091] The actual working hours calculation formula is as follows:

[0092]

[0093] Where, T act is the actual working hours; E eis the equipment efficiency coefficient, ranging from 0.7 to 1.2; S o is the operator skill coefficient, which is 0.8 for beginners, 1.0 for intermediate, 1.2 for advanced, and 1.5 for experts.

[0094] The calculation formula for tooling fit evaluation is as follows:

[0095] FC ij =λ1·G ij +λ2·L ij +λ3·F ij +λ4·O ij ;

[0096] Where, FC ij G is the adaptability of tooling i to component j; ij is the geometric shape matching degree, ranging from 0 to 1; L ij is the positioning accuracy matching degree, ranging from 0 to 1; F ij For the rationality of clamping force distribution, the range is 0 to 1; ij For operational convenience, the range is 0 to 1; λ1, λ2, λ3, and λ4 are weight coefficients, and λ1+λ2+λ3+λ4=1. The default values ​​are 0.3, 0.3, 0.2, and 0.1 respectively.

[0097] The parameter acquisition method is: G ij Compare the conformity of the contact surface profiles between the tooling and the parts through geometric matching algorithm calculation; L ij Obtained through positioning error analysis, where δ loc is the positioning error, δ allow is the allowable error; F ij Obtained through clamping force analysis, evaluate whether the clamping point distribution and force size are reasonable; ij Obtained through ergonomics assessment, taking into account the convenience of clamping operations.

[0098] In these formulas, the use of a multiplicative relationship primarily accounts for the cumulative effects of process parameters. For example, in the standard working time calculation, the correction effect of various factors on the baseline working time is cumulative. In contrast, the use of a reciprocal relationship in the actual working time calculation reflects the inverse relationship between efficiency factors and time: higher efficiency results in less time required.

[0099] In step S04, the calculation of the first-level conversion anomaly index matrix involves multiple processes. The characteristic deviation index calculation formula is specifically expressed as follows:

[0100]

[0101] Where, FDI i is the deviation index of the i-th feature; N matchis the number of feature matching items; N total is the total number of features; A d is the feature resolution ambiguity; A max is the maximum allowable ambiguity.

[0102] The parameter acquisition method is: N match and N total Obtained through feature comparison analysis, compare the matching of various attributes of design features and process features. d Obtained by feature parsing ambiguity evaluation, A d =1-max{P(f1),P(f2),...,P(f n )}, where P(f i ) is the probability of the i-th possible parsing result.

[0103] The tolerance satisfaction calculation formula is specifically expressed as follows:

[0104]

[0105] Where, TSI i is the tolerance satisfaction of the i-th dimension; T process T is the processing tolerance that can be achieved by the process system; design Tolerance required by design.

[0106] The parameter acquisition method is: T process Obtained from the process capability database, determined based on equipment precision capability, processing methods and material properties. design Extracted from the design model.

[0107] The material compatibility calculation formula is as follows:

[0108]

[0109] Where, MAI i is the suitability of the i-th material; C j (M i , P j ) is the compatibility function of the j-th attribute; u j is the weight coefficient of the j-th attribute, and l is the number of material properties considered.

[0110] The parameter acquisition method is: compatibility function C j (M i , P j ) Different calculation methods are used according to the type of attributes. For continuous attributes, evaluate whether the material attribute value is within the process requirements; for discrete attributes, evaluate whether the material attribute matches the process requirements.

[0111] The calculation formula of the abnormal index matrix is ​​specifically expressed as follows:

[0112] AIM ijk =F ijk ·S ijk I ijk ;

[0113] Where AIM ijk F is the abnormality index of the i-th abnormality on the k-th feature of the j-th component; ijk is the abnormal frequency, normalized to the range of 0 to 1; S ijk The severity level is abnormal, with fatal being 1.0, severe being 0.8, moderate being 0.5, and mild being 0.2; ijk is the impact range of the anomaly, normalized to the range of 0 to 1.

[0114] The parameter acquisition method is: F ijk Obtained through statistical historical abnormal records, where N ijk is the number of occurrences of this type of exception, N max The highest occurrence of all exceptions. ijk Obtained through impact range analysis, Among them C ijk is the number of affected components, C total is the total number of components.

[0115] In these formulas, the use of difference and ratio relationships primarily addresses the need for deviation assessment, quantifying the degree of anomaly by calculating the difference between the actual value and the target value. The product relationship reflects the synergistic effect of various factors in the anomaly index. The anomaly index is high only when the frequency, severity, and impact range are all high, reflecting the comprehensive nature of anomaly risk assessment.

[0116] In step S05, the minimum spanning tree algorithm and the calculation of the second-level conversion anomaly index matrix involve multiple processes. The edge weight calculation formula is specifically expressed as follows:

[0117] W ij =α·T ij +β·A ij +γ·C ij ;

[0118] Where W ij is the edge weight from process i to process j; T ij is the switching time cost, normalized to the range of 0 to 1; A ij Adjust the complexity for the device and normalize it to the range of 0 to 1; C ijis the resource occupation conflict degree, normalized to the range of 0 to 1; α, β, and γ are weight coefficients, and α+β+γ=1, with default values ​​of 0.4, 0.3, and 0.3 respectively.

[0119] The parameter acquisition method is: T ij Obtained through historical data statistics, where t ij is the average time to switch from process i to process j, t min and t max are the minimum and maximum switching times respectively. ij Obtained through equipment adjustment assessment, taking into account the complexity of operations such as fixture replacement, parameter adjustment, and tool replacement. ij Obtained through resource occupancy analysis, evaluate the degree of competition between process i and process j for the same resources.

[0120] The calculation formula of process conflict degree is as follows:

[0121]

[0122] Where, PC ij is the conflict degree between process i and process j; LC ij is the logical constraint conflict degree, ranging from 0 to 1; RC ij is the resource constraint conflict degree, ranging from 0 to 1.

[0123] The parameter acquisition method is: LC ij Obtained through process logic relationship analysis, evaluate the satisfaction of pre- and post-constraint conditions between processes. RC ij Obtained through resource competition analysis, where w r is the weight of resource r, q is the amount of resources.

[0124] The resource load rate calculation formula is as follows:

[0125]

[0126] Where RL r is the load rate of resource r; T i is the time requirement of process i; T total is the total planned time; p is the number of processes.

[0127] The calculation formula of load variance coefficient is as follows:

[0128]

[0129] Where, CV ris the load variance coefficient of resource r; σ r is the standard deviation of the resource r load; μ r is the average load of resource r.

[0130] The parameter acquisition method is: r and μ r Obtain resource load time series data statistics to analyze resource load fluctuations during the planning cycle.

[0131] In these formulas, the use of a weighted summation relationship primarily addresses the need for comprehensive multi-factor evaluation, with different weight coefficients reflecting the importance of each factor. The use of the variance coefficient reflects the focus on load volatility, eliminating dimensionality effects by using the ratio of the standard deviation to the mean to directly reflect the relative degree of volatility.

[0132] In step S06, the calculation of the third-level conversion anomaly index matrix involves multiple processes. The assembly interference probability calculation formula is specifically expressed as follows:

[0133]

[0134] Where AIP ij is the assembly interference probability between component i and component j; N interference is the number of interferences in the Monte Carlo sample; N samples The total number of samples for Monte Carlo simulation.

[0135] The parameter acquisition method is: obtain it through Monte Carlo simulation, consider the random variation of component size, position and posture, generate multiple groups of assembly states, and detect the interference in each state.

[0136] The tooling matching degree calculation formula is specifically expressed as follows:

[0137]

[0138] Where, TFI i is the matching index of tooling i; D ik is the deficiency of tooling i in the kth dimension, ranging from 0 to 1; w k is the weight coefficient of the kth dimension, and The four dimensions are geometric shape matching, positioning accuracy assurance, clamping force distribution and operational convenience.

[0139] The parameter acquisition method is as follows: the deficiency of each dimension is obtained through tooling evaluation, D ik =1-S ik , where S ik Score the fit of the tooling on this dimension.

[0140] The calculation formula for the probability of exceeding the working hours limit is as follows:

[0141] TOP i =P(T actual >T standard ·(1+θ));

[0142] In the formula, TOP i is the probability of exceeding the working time limit of process i; T actual is the actual working hours random variable; T standard is the standard working hours; θ is the allowable deviation rate, and the default value is 0.2.

[0143] The parameter acquisition method is: obtain it through statistical analysis of working hours data, assuming that the actual working hours obey the normal distribution, where μ i and σ i According to historical data, where Φ is the cumulative distribution function of the standard normal distribution.

[0144] The risk score calculation formula is as follows:

[0145] RR ijk =P ijk ·S ijk ·D ijk ;

[0146] Where RR ijk P is the risk score of the i-th abnormality in the j-th component k-th process; ijk is the probability of abnormality, ranging from 0 to 1; S ijk is the severity of the impact, ranging from 0 to 1; D ijk is the difficulty of detection, ranging from 0 to 1.

[0147] The parameter acquisition method is: ijk Obtained through historical data statistics or expert evaluation; S ijk Obtained through impact analysis, considering the impact of abnormalities on quality, cost, and delivery time; D ijk Obtained through detection capability assessment, taking into account the ability of current quality control measures to detect the anomaly.

[0148] In these formulas, ratio relationships, such as the sample ratio in the assembly interference probability calculation, directly reflect the definition of probability. Product relationships, such as the risk scoring formula, embody the three-dimensional assessment model of probability, impact, and detection commonly used in risk management. Risk is considered serious only when all three factors are high.

[0149] In step S07, the construction of the first and second level inter-correlation matrices involves multiple calculation processes. The feature influence weight calculation formula is specifically expressed as follows:

[0150]

[0151] Where, IW ij is the influence weight of design feature i on process route node j; P(R j |F i ) is a given design feature F i Select process route node R under the conditions j probability.

[0152] Parameters are acquired through feature sensitivity analysis, assessing the impact of changes in design feature parameters on process selection. In practical applications, this can be achieved by using a control variable approach, varying the parameter value of feature i and observing the change in the probability of selecting process node j. Alternatively, feature importance analysis using a machine learning model can be used to obtain these parameters.

[0153] The calculation formula of the correlation matrix elements is specifically expressed as follows:

[0154] CM ij =[FID i , RID j , IS ij , CS ij ];

[0155] Where, CM ij is the element in row i and column j of the incidence matrix; FID i RID is a feature identifier; j IS is the process identifier; ij is the impact strength, ranging from 0 to 1; CS ij The sensitivity of the change ranges from 0 to 1.

[0156] The parameter acquisition method is: IS ij Obtained through impact analysis, Normalization process features influence weight. CS ij The sensitivity of design feature changes to process nodes can be assessed through change propagation simulation.

[0157] The complete expression of the first and second level correlation matrix is:

[0158]

[0159] Where RCM is the correlation matrix; m is the number of design features; and n is the number of process route nodes.

[0160] In these formulas, the use of partial derivatives primarily addresses the essence of sensitivity analysis. By calculating the derivative of conditional probability with respect to a feature parameter, the impact of parameter changes on probability is directly reflected. Matrix representation effectively organizes the complex, multidimensional correlations between features and process nodes.

[0161] In step S08, the construction of the second and third level inter-correlation matrices involves multiple calculation processes. The multi-dimensional mapping relationship data structure is specifically represented as follows:

[0162] MR ijk =[RID i , RT j , RIID k , CS ijk , PR ijk ];

[0163] Where, MR ijk For mapping relationship elements; RID i is the process identifier; RT j Is the resource type; RIID k CS is the resource instance identifier; ijk Score the fitness, ranging from 0 to 1; PR ijk It is an integer value, the smaller the value, the higher the priority.

[0164] The parameter acquisition method is: CS ijk Obtained through resource adaptability assessment, comprehensive consideration of technical adaptability, economic adaptability and time adaptability, CS ijk =λ1·TS ijk +λ2·ES ijk +λ3·TMS ijk , where λ1, λ2, and λ3 are weight coefficients, and λ1+λ2+λ3=1, and the default values ​​are 0.5, 0.3, and 0.2 respectively; TS ijk Score the technology suitability; ES ijk Scoring economic suitability; TMS ijk Score the time suitability. ijk It is obtained through a priority evaluation algorithm, taking into account factors such as resource load, device status, and past usage results.

[0165] The complete expression of the second and third level correlation matrix is:

[0166]

[0167] Where LCM is the second and third level intercorrelation matrix; m is the number of processes; n is the number of resource types; and p is the upper limit of the number of instances of each resource type.

[0168] The calculation formula for change impact analysis is as follows:

[0169] CI ijk =PS ij ·RI jk ;

[0170] Where, CI ijk is the impact of the change of process node i on resource instance k; PS ij is the association strength between process node i and resource type j, ranging from 0 to 1; RI jk is the association strength between resource type j and resource instance k, ranging from 0 to 1.

[0171] The parameter acquisition method is: PS ij Obtained through process-resource correlation analysis, evaluate the degree of dependence of process nodes on different types of resources. jk Obtained through resource instance mapping analysis, the strength of the mapping relationship between resource types and specific instances is evaluated.

[0172] In these formulas, the matrix representation effectively organizes the complex, multidimensional relationships between process routes and manufacturing resources. The weighted summation method used in calculating the fitness score takes into account the need for comprehensive multi-factor assessment. The use of the product relationship in change impact analysis reflects the cascading effect of impact transmission, whereby process changes indirectly affect resource instances through resource types.

[0173] In step S09, the construction of the fusion conversion compensation matrix involves multiple calculation processes. The multi-objective optimization function is specifically expressed as follows:

[0174] O(x)=ω1·A(x)+ω2·E(x)+ω3·R(x)+ω4·M(x);

[0175] Where O(x) is the comprehensive optimization objective; A(x) is the conversion accuracy objective, ranging from 0 to 1, with larger values ​​being better; E(x) is the conversion efficiency objective, ranging from 0 to 1, with larger values ​​being better; R(x) is the resource consumption objective, ranging from 0 to 1, with smaller values ​​being better; M(x) is the manufacturing risk objective, ranging from 0 to 1, with smaller values ​​being better; ω1, ω2, ω3, and ω4 are weight coefficients, with ω1+ω2+ω3+ω4=1, and their default values ​​being 0.4, 0.2, 0.2, and 0.2, respectively; x is a decision variable vector, representing the set of compensation strategy parameters.

[0176] The parameter acquisition method is: A(x) is obtained through conversion quality evaluation, Among them, AI i is the abnormal index of the i-th conversion; E(x) is obtained by evaluating the conversion time, Where T actual is the actual conversion time, T max is the maximum allowed time; R(x) is obtained through resource occupancy evaluation, RC actual is the actual resource consumption, RC maxis the maximum resource limit; M(x) is obtained through risk assessment, Among them RI j Create a risk index for item j.

[0177] The calculation formula for compensation strategy weight distribution is as follows:

[0178]

[0179] Where CW i is the weight of the i-th compensation strategy; CP i is the performance score of the i-th compensation strategy, ranging from 0 to 1; β is the temperature parameter that controls the sharpness of the allocation, with a default value of 5; k is the total number of compensation strategies.

[0180] The parameter acquisition method is: CP i Obtained through compensation effect evaluation, determined by combining historical application data and expert ratings.

[0181] The compensation execution priority sequence generation formula is specifically expressed as follows:

[0182] CPS=Sort({(CS1,CU1),(CS2,CU2),...,(CS k , CU k )});

[0183] Where CPS is the priority sequence for executing compensation strategy; CS i is the i-th compensation strategy; CU i is the urgency of the i-th compensation strategy, ranging from 0 to 1; Sort is a descending sorting function, according to CU i The values ​​are sorted from largest to smallest.

[0184] The parameter acquisition method is: CU i Obtained through urgency assessment, CU i =α1·AI i +α2·TI i +α3·RI i , among which AI i For abnormal severity, TI i is the time urgency, RI i is the risk impact, α1, α2, α3 are weight coefficients, and α1+α2+α3=1.

[0185] In these formulas, the application of weighted summation in multi-objective optimization achieves trade-offs between multiple objectives. The use of exponential functions and softmax forms allows for more flexible weight distribution in compensation strategies. Adjusting the temperature parameter allows for control over the degree of centralization or dispersion of the distribution. The use of a ranking function ensures a rational order of priority for compensation execution.

[0186] The design of these calculation processes and formulas fully considers the influence and interrelationships of various factors in the conversion process from design model to process model. Through multi-level matrix conversion and anomaly index evaluation, accurate mapping of design intent to manufacturing implementation is achieved, and an adaptive compensation mechanism is provided for possible abnormal situations.

[0187] Specifically, the core technology of this invention is based on formalizing engineers' empirical knowledge into mathematical models and algorithmic rules, achieving precise conversion from design models to process models through a multi-level, multi-dimensional matrix conversion architecture. Specifically, this invention employs a three-level conversion matrix architecture to systematically address the three key conversion steps: from design model to process features, from design bill of materials (BOM) to process PBOM, and from process PBOM to manufacturing MBOM.

[0188] The first-level conversion matrix establishes mapping rules with process characteristics by structured processing of the geometric features, dimensional parameters and assembly relationships in the design model, transforming the traditional process that relies on engineers' intuitive judgment into a computable mathematical problem; the second-level conversion matrix converts the design BOM to the process PBOM. Through standardized component relationship reconstruction and process flow generation rules, it realizes the logical transformation from product structure to manufacturing sequence, replacing engineers' experience-based manual route planning; the third-level conversion matrix concretizes the abstract process plan into actual manufacturing resource configuration, and realizes the precise mapping of the process PBOM to the manufacturing MBOM through process resource allocation and production execution parameter generation rules.

[0189] This invention also introduces an anomaly index matrix system, transforming potential problems typically identified by engineers through experience into a systematic anomaly detection model. This model quantitatively assesses issues such as geometric feature mismatches, dimensional tolerance violations, and material property conflicts during the conversion process. A minimum spanning tree algorithm is applied to optimize process route planning, mathematically determining the optimal process route combination, breaking away from the limitations of traditional reliance on subjective judgment by engineers. By constructing a correlation matrix, multidimensional correlations are achieved between design features and process routes, and between process routes and manufacturing resources, making complex dependencies traceable and manageable.

[0190] The process feature recognition conversion network based on graph neural network is another key innovation of this invention. It can learn the mapping rules of design features to process features from historical successful cases. Through the four-layer architecture of feature encoding, graph structure learning, relational reasoning and feature mapping, it gradually improves the conversion accuracy and realizes the paradigm shift from "relying on experience" to "data-driven".

[0191] This systematic mathematical model and algorithm framework makes engineers' implicit knowledge explicit and standardized, so that the conversion process from design model to process model no longer relies on personal experience, but is based on scientific methods and data-driven precise calculations, fundamentally solving the technical problem of insufficient conversion accuracy.

[0192] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows.

[0193] The specific implementation of step S01 is to achieve the initial conversion of the design model to the process model by constructing a multi-dimensional first-level conversion matrix. First, a hierarchical clustering algorithm is used to analyze the process relevance of the drawing data in the design model, and the process relevance of each design feature is calculated according to the process relevance evaluation formula:

[0194] R i =α·D i +β·C i +γ·M i ;

[0195] Where R i is the process relevance of the i-th design feature; D i is the processing difficulty coefficient, ranging from 0 to 1; C i is the complexity of process implementation, ranging from 0 to 1; M iis the manufacturing resource requirement, ranging from 0 to 1; α, β, and γ are weight coefficients, with α + β + γ = 1, and default values ​​of 0.4, 0.3, and 0.3, respectively. Based on the calculation results, the design features are divided into three categories: high correlation (threshold above 0.85), medium correlation (threshold 0.6-0.85), and low correlation (threshold 0.4-0.6). A feature recognition algorithm is then used to extract key dimensional parameters, including dimensional data such as aperture, depth, slot width, wall thickness, chamfer, fillet, and thread specification. A boundary detection algorithm is used to identify the spatial distribution and topological relationships of geometric features. The maximum stability principle is then applied to determine the transformation reference points. The feature points in the design model that are least deformable and have clear process meaning are selected as coordinate transformation anchor points, and a mapping function from the design coordinate system to the process coordinate system is established. A feature structured encoding method is then used to organize all extracted data into a hierarchical tree structure. Each node contains a feature type identifier, a parameter value set, location information, and a description of the association relationship, forming a complete structured data set. Finally, a Bayesian network is used to calculate the transformation confidence index:

[0196] CI=ω1·AR+ω2·CP+ω3·BS;

[0197] Where CI is the conversion confidence index; AR is the feature recognition accuracy, ranging from 0 to 1; CP is the parameter extraction completeness, ranging from 0 to 1; BS is the reference point stability, ranging from 0 to 1; ω1, ω2, and ω3 are weight coefficients, where ω1 + ω2 + ω3 = 1, with default values ​​of 0.4, 0.3, and 0.3, respectively. A manual verification process is triggered when the confidence level falls below 0.75. The purpose of this step is to transform the unstructured information in the design model into structured data that can be directly recognized and processed by the process system, providing a data foundation for subsequent process planning.

[0198] The specific implementation method of step S02 is to construct a multi-dimensional second-level conversion matrix to achieve the preliminary conversion of the design BOM to the process PBOM. First, the design BOM is pre-processed using data cleaning technology, including removing duplicates, filling missing values, unifying data formats, standardizing coding systems and standardizing material descriptions to ensure that the data quality reaches more than 95% integrity and consistency. Then, the decision tree algorithm is applied to establish the mapping relationship between parts and process routes. The geometric complexity, material type, precision level and functional characteristics of the parts are used as input features to generate suitable processing routes. The depth of the decision tree is limited to 8 layers, and the leaf node purity threshold is set to 0.9. The fuzzy association rule algorithm is then used to match the material properties with the process parameters, and the matching degree between the material and the process node is calculated:

[0199]

[0200] Where MC ijis the matching degree between material i and process node j; M ik is the kth property value of material i; P jk is the required value of the kth material property at process node j; S(M ik , P jk ) is the matching function of a single attribute; w k is the weight coefficient of the kth attribute, and n is the number of material properties considered. The matching threshold is set to 0.8, and combinations below this value are marked as potential risk items. The manufacturing equipment capability parameters are then incorporated into the conversion process through the resource capability assessment model:

[0201]

[0202] Where, EC ij is the adaptability of equipment i to component j; E ik is the kth capability parameter of device i; R jk is the requirement of component j for the kth capability parameter; C(E ik , R jk ) is the fitness function of a single capability; v k is the weight coefficient of the kth capability parameter, and m represents the number of capability parameters considered. The equipment suitability threshold is set to 0.75; configurations below this value are marked as not recommended. Finally, a preliminary version of the process PBOM is generated, including the process breakdown structure, process flow definition, initial process parameter values, and resource requirement estimates, completing the transition from a design perspective to a process perspective. This step transforms the design structure oriented toward functional implementation into a process structure oriented toward manufacturing implementation, laying the foundation for subsequent manufacturing execution.

[0203] The specific implementation of step S03 is to construct a multi-dimensional third-level conversion matrix to achieve a deep conversion from the process PBOM to the manufacturing MBOM. First, an association rule mining algorithm is applied to establish a detailed mapping between the process planning route and the process resource library. Based on historical manufacturing data, the strength of the association between each process and specific equipment, tooling, and personnel is analyzed, and the support and confidence of the association rules are calculated:

[0204]

[0205] Where, Association rules support level; Association rules The confidence level is ; count(X∪Y) is the number of records containing both X and Y; count(X) is the number of records containing X; and N is the total number of records. The minimum support of the association rule is set to 0.3, and the minimum confidence level is set to 0.85. Highly reliable process resource allocation rules are then extracted. A multi-constraint resource allocation algorithm is then used to integrate the work hour quota data and calculate standard and actual work hours:

[0206] T std =T base ·C c ·C b ·C q ;

[0207]

[0208] Where, T std is the standard working time; T base is the benchmark working time, which is obtained from the working time quota library according to the process type; C c is the complexity coefficient, ranging from 1 to 5; C b is the batch coefficient, C b =1-log 10 (B) 0.05, where B is the batch size; C q is the quality grade coefficient, special grade is 1.3, first grade is 1.1, second grade is 1.0, and third grade is 0.9; T act is the actual working hours; E e is the equipment efficiency coefficient, ranging from 0.7 to 1.2; S o The operator skill coefficient is 0.8 for beginners, 1.0 for intermediate, 1.2 for advanced, and 1.5 for experts. The material quota data is then integrated through the material requirement planning algorithm to calculate the material consumption, loss rate (generally set at 2% to 8%), and waste recycling rate of each process. A material flow chain diagram is then established to achieve precise control of the entire material process. The geometric constraint satisfaction evaluation algorithm is then used to associate the fixture data with the corresponding process and calculate the fixture fit:

[0209] FC ij =λ1·G ij +λ2·L ij +λ3·F ij +λ4·O ij ;

[0210] Where, FC ij G is the adaptability of tooling i to component j; ij is the geometric shape matching degree, ranging from 0 to 1; L ij is the positioning accuracy matching degree, ranging from 0 to 1; F ij For the rationality of clamping force distribution, the range is 0 to 1;ij For operational convenience, the range is 0 to 1; λ1, λ2, λ3, and λ4 are weight coefficients, with λ1 + λ2 + λ3 + λ4 = 1. The default values ​​are 0.3, 0.3, 0.2, and 0.1, respectively. The fitness threshold is set to 0.9; combinations below this value require tooling optimization or replacement. Finally, a complete manufacturing MBOM is generated, including process implementation details, resource instance allocation plans, material consumption quotas, and quality control points, forming a directly executable manufacturing guidance document. The purpose of this step is to transform the abstract process plan into specific manufacturing execution instructions, ensuring the precise transfer of process intent to manufacturing operations.

[0211] The specific implementation of step S04 is to calculate the first-level conversion anomaly index matrix and establish an abnormality management mechanism for the conversion of design to process features. First, a feature comparison analysis algorithm is used to detect geometric feature conversion anomalies and calculate the feature deviation index:

[0212]

[0213] Where, FDI i is the deviation index of the i-th feature; N match is the number of feature matching items; N total is the total number of features; A d is the feature resolution ambiguity; A max is the maximum allowable ambiguity. When the feature recognition rate is lower than 0.9 or the feature resolution ambiguity is higher than 0.2, it is marked as a geometric feature mismatch. Then, the statistical process control method is applied to analyze the dimensional tolerance exceeding the standard and calculate the tolerance satisfaction:

[0214]

[0215] Where, TSI i is the tolerance satisfaction of the i-th dimension; T process T is the processing tolerance that can be achieved by the process system; design is the tolerance required by the design. A degree of satisfaction below 0.95 is marked as a risk of exceeding the dimensional tolerance. The conflict detection algorithm is then used to identify material property conflicts and calculate the material compatibility:

[0216]

[0217] Where, MAI i is the suitability of the i-th material; C j (M i , P j ) is the compatibility function of the j-th attribute; u j is the weight coefficient of the j-th attribute, and l is the number of material properties considered. A material property conflict is identified when the fit is less than 0.85. A structured anomaly database is then established, using a multi-level classification coding system to record anomaly information, including anomaly type code, anomaly location coordinates, discovery timestamp, associated component number, impact range, and severity. Finally, the anomaly index matrix is ​​calculated:

[0218] AIM ijk =F ijk ·S ijk I ijk ;

[0219] Where AIM ijk F is the abnormality index of the i-th abnormality on the k-th feature of the j-th component; ijk is the abnormal frequency, normalized to the range of 0 to 1; S ijk The severity level is abnormal, with fatal being 1.0, severe being 0.8, moderate being 0.5, and mild being 0.2; ijk The impact range of the anomaly is normalized to the range of 0 to 1. The purpose of this step is to monitor the quality of the conversion process from design to process features and to promptly detect and manage conversion anomalies.

[0220] The specific implementation of step S05 is to apply the minimum spanning tree algorithm to optimize the calculation of the second-level conversion anomaly index matrix. First, the process route planning problem is modeled as a weighted undirected graph, with each process as a vertex in the graph and the relationship between processes as an edge. The edge weight is calculated:

[0221] W ij =α·T ij +β·A ij +γ·C ij ;

[0222] Where W ij is the edge weight from process i to process j; T ij is the switching time cost, normalized to the range of 0 to 1; A ij Adjust the complexity for the device and normalize it to the range of 0 to 1; C ij is the resource occupation conflict degree, normalized to the range of 0 to 1; α, β, and γ are weight coefficients, and α+β+γ=1, with default values ​​of 0.4, 0.3, and 0.3, respectively. The Prim algorithm is then applied to solve the minimum spanning tree. Starting from the initial process, the minimum weight edge connected to the current tree is selected each time until all process nodes are covered, obtaining the process route combination with the lowest total cost. The critical path analysis method is then used to identify the bottleneck processes in the process route. The earliest start time, latest completion time, and total time difference of each process are calculated. Processes with zero time difference constitute the critical path and serve as the optimization focus. The conflict detection algorithm is then used to analyze the process logic conflicts and calculate the process conflict degree:

[0223]

[0224] Where, PC ij is the conflict degree between process i and process j; LC ij is the logical constraint conflict degree, ranging from 0 to 1; RC ij is the resource constraint conflict degree, ranging from 0 to 1. Records with a conflict degree exceeding 0.6 are considered process conflict anomalies. Finally, the resource load balancing algorithm is applied to evaluate resource occupation conflicts and calculate the resource load rate and load variance coefficient:

[0225]

[0226] Where RL r is the load rate of resource r; T i is the time requirement of process i; T total is the total planned time; p is the number of processes; CV r is the load variance coefficient of resource r; σ r is the standard deviation of the resource r load; μ r is the average load of resource r. When a resource's load ratio exceeds 0.9 or its load variance coefficient is greater than 0.25, it is flagged as a resource conflict risk. All exception information is aggregated to form an exception index matrix. This step aims to optimize process routing and identify potential issues during process PBOM conversion, improving the feasibility and efficiency of process planning.

[0227] The specific implementation of step S06 is to calculate the third-level conversion anomaly index matrix to predict and analyze potential anomalies in the manufacturing execution link. First, the Monte Carlo simulation method is used to analyze the assembly interference risk and calculate the assembly interference probability:

[0228]

[0229] Where AIP ij is the assembly interference probability between component i and component j; N interference is the number of interferences in the Monte Carlo sample; N samples is the total number of samples in the Monte Carlo simulation. When the interference probability exceeds 0.15, it is determined to be a high-risk assembly relationship. Then, the multi-dimensional adaptability evaluation model is applied to analyze the problem of insufficient tooling matching and calculate the insufficient tooling matching index:

[0230]

[0231] Where, TFI i is the matching index of tooling i; D ik is the deficiency of tooling i in the kth dimension, ranging from 0 to 1; w kis the weight coefficient of the kth dimension, and The four dimensions are geometric shape matching, positioning accuracy assurance, clamping force distribution, and ease of operation. A comprehensive score below 0.8 is marked as insufficient tooling matching. The time series prediction algorithm is then used to analyze the risk of exceeding the working time limit and calculate the probability of exceeding the working time limit:

[0232] TOP i =P(T actual >T standard ·(1+θ));

[0233] In the formula, TOP i is the probability of exceeding the working time limit of process i; T actual is the actual working hours random variable; T standard is the standard working hours; θ is the allowable deviation rate, with a default value of 0.2. Processes with a deviation rate exceeding 20% ​​and a probability of occurrence greater than 0.3 are recorded as work-hour overrun risk items. Subsequently, an exception guidance library is constructed, classifying and coding the prediction and analysis results according to exception type, impact, and resolution difficulty, and linking historical solutions to form a closed knowledge loop. Finally, the third-level conversion exception index matrix is ​​calculated to generate a risk score:

[0234] RR ijk =P ijk ·S ijk ·D ijk ;

[0235] Where RR ijk P is the risk score of the i-th abnormality in the j-th component k-th process; ijk is the probability of abnormality, ranging from 0 to 1; S ijk is the severity of the impact, ranging from 0 to 1; D ijk The difficulty of detection ranges from 0 to 1. Risk levels are categorized as high (above 0.7), medium (0.4 to 0.7), and low (below 0.4). This step proactively identifies potential issues in manufacturing execution and provides data support and guidance for process design optimization.

[0236] The specific implementation of step S07 is to build a bidirectional tracking mapping relationship between design features and process routes. First, a pre-trained process feature recognition conversion network model is loaded. This model adopts a graph convolutional neural network architecture, which includes four core layers: feature encoding, graph structure learning, relational reasoning, and feature mapping. It effectively captures the topological structure and semantic information of design features. Then, a feature sensitivity analysis algorithm is used to identify key design features and calculate the feature impact weights:

[0237]

[0238] Where, IW ijis the influence weight of design feature i on process route node j; P(R j |F i ) is a given design feature F i Select process route node R under the conditions j The probability of . Features with an impact coefficient greater than 0.5 are marked as key features and serve as the core objects for change tracking. Then, a bidirectional graph mapping algorithm is used to establish the correspondence between design features and process route nodes, and to construct the association matrix elements:

[0239] CM ij =[FID i , RID j , IS ij , CS ij ];

[0240] Where, CM ij is the element in row i and column j of the incidence matrix; FID i RID is a feature identifier; j IS is the process identifier; ij is the impact strength, ranging from 0 to 1; CS ij is the change sensitivity, ranging from 0 to 1. A reverse derivation algorithm is then applied to establish a feedback channel from process optimization to design improvement, analyze the correlation strength between process bottlenecks and design features, generate design optimization suggestions, and provide data support for collaborative design. Finally, a complete first and second level correlation matrix is ​​formed:

[0241]

[0242] Where RCM is the correlation matrix, m is the number of design features, and n is the number of process route nodes. The purpose of this step is to break down the data barriers between the design domain and the process domain and achieve bidirectional collaborative optimization of design and process.

[0243] The specific implementation of step S08 is to achieve accurate mapping association between process PBOM and manufacturing MBOM. First, a hierarchical association mapping algorithm is applied to construct a multi-dimensional mapping relationship between process routes and manufacturing resources, and a multi-dimensional mapping relationship data structure is established:

[0244] MR ijk =[RID i , RT j , RIID k , CS ijk , PR ijk ];

[0245] Where, MR ijk For mapping relationship elements; RID i is the process identifier; RT j Is the resource type; RIIDk CS is the resource instance identifier; ijk Score the fitness, ranging from 0 to 1; PR ijk The priority is an integer value, with smaller values ​​indicating higher priorities. The capabilities of manufacturing resource instances are then parameterized using resource capability modeling methods, including the equipment's machining accuracy range, maximum machining dimensions, machinable material types, and the applicable scope of tooling fixtures, forming a standardized resource capability description model. A multi-condition matching algorithm is then used to precisely match process requirements with resource capabilities, and a fitness score is calculated:

[0246] CS ijk =λ1·TS ijk +λ2·ES ijk +λ3·TMS ijk ;

[0247] Where λ1, λ2, and λ3 are weight coefficients, and λ1+λ2+λ3=1. The default values ​​are 0.5, 0.3, and 0.2 respectively. ijk Score the technology suitability; ES ijk Scoring economic suitability; TMS ijk Score the time adaptability. Then use the change impact analysis algorithm to evaluate the impact of the process change on manufacturing execution and calculate the impact of the change:

[0248] CI ijk =PS ij ·RI jk ;

[0249] Where, CI ijk is the impact of the change of process node i on resource instance k; PS ij is the association strength between process node i and resource type j, ranging from 0 to 1; RI jk is the association strength between resource type j and resource instance k, ranging from 0 to 1. Finally, the complete second and third level mutual correlation matrix is ​​generated:

[0250]

[0251] Where LCM is the second- and third-level correlation matrix; m is the number of processes; n is the number of resource types; and p is the upper limit on the number of instances of each resource type. The matrix refresh frequency is set to be triggered by a change in the manufacturing execution system status or no more than 10 minutes. This step establishes a data bridge between process planning and manufacturing execution, ensuring that process design intent is accurately implemented at the manufacturing site.

[0252] The specific implementation of step S09 is to construct a self-optimizing fusion conversion compensation mechanism. First, a multi-objective genetic algorithm is applied to construct a fusion conversion compensation matrix, and a multi-objective optimization function is set:

[0253] O(x)=ω1·A(x)+ω2·E(x)+ω3·R(x)+ω4·M(x);

[0254] Where O(x) is the comprehensive optimization objective; A(x) is the conversion accuracy objective, ranging from 0 to 1, with higher values ​​being preferred; E(x) is the conversion efficiency objective, ranging from 0 to 1, with higher values ​​being preferred; R(x) is the resource consumption objective, ranging from 0 to 1, with lower values ​​being preferred; M(x) is the manufacturing risk objective, ranging from 0 to 1, with lower values ​​being preferred; ω1, ω2, ω3, and ω4 are weight coefficients, with ω1 + ω2 + ω3 + ω4 = 1, and default values ​​being 0.4, 0.2, 0.2, and 0.2, respectively; and x is the decision variable vector, representing the set of compensation strategy parameters. The genetic algorithm parameters are set as follows: population size 200, evolutionary generations 500, crossover rate 0.85, and mutation rate 0.1. A pattern mining algorithm is then used to analyze historical conversion data, identifying common conversion patterns and abnormal patterns. A pattern library is then established, containing pattern IDs, triggering conditions, impact scope, and applicable scenarios, providing knowledge support for conversion compensation. Then, we use the decision tree algorithm to build a compensation strategy library, select the optimal compensation plan based on the anomaly type, severity, and resource constraints, and calculate the compensation strategy weights:

[0255]

[0256] Where CW i is the weight of the i-th compensation strategy; CP i is the performance score of the i-th compensation strategy, ranging from 0 to 1; β is a temperature parameter that controls the sharpness of the allocation, with a default value of 5; and k is the total number of compensation strategies. Compensation strategies include adaptive parameter adjustment, alternative resource allocation, process rescheduling, and manual intervention. A Bayesian optimization algorithm is then applied to evaluate the compensation effect and conduct self-learning. The compensation strategy parameters are continuously optimized based on feedback from the post-compensation conversion quality. The learning rate is set to 0.05, and the convergence threshold is set to 0.001. Finally, a compensation execution priority sequence is generated:

[0257] CPS=Sort({(CS1,CU1),(CS2,CU2),...,(CS k , CU k )});

[0258] Where CPS is the priority sequence for executing compensation strategy; CS i is the i-th compensation strategy; CU i is the urgency of the i-th compensation strategy, ranging from 0 to 1; Sort is a descending sorting function, according to CU iThe values ​​are sorted from largest to smallest. The system periodically evaluates the conversion quality. When the cumulative optimization gain falls below 0.1%, it is considered converged and the conversion optimization is complete. The purpose of this step is to build a conversion compensation mechanism with self-learning capabilities to continuously improve the accuracy and robustness of model conversion.

[0259] The core innovation of this embodiment lies in the realization of high-precision conversion from design model to process model through the construction of multi-level conversion matrix and anomaly index matrix. The multi-dimensional first-level conversion matrix realizes the precise mapping of design features to process features; the multi-dimensional second-level conversion matrix realizes the structural reorganization of design BOM to process PBOM; and the multi-dimensional third-level conversion matrix realizes the resource association of process PBOM to manufacturing MBOM. The three-level conversion anomaly index matrix is ​​used to quantitatively evaluate and predict anomalies in the conversion process. Combined with the mutual correlation matrix, a two-way tracking mechanism for design changes and process adjustments is established. Finally, the adaptive optimization of the conversion process is achieved by integrating the conversion compensation matrix, forming a complete set of digital conversion methods from design to manufacturing. All levels of the conversion process use precise mathematical model expressions and calculation methods, comprehensively considering multiple key factors such as process relevance, resource matching, anomaly assessment, etc., and strictly controlling the quality and efficiency of each conversion link to ensure the accuracy and reliability of the conversion results.

[0260] In order to better understand and implement the present invention, Example 2 of a specific application scenario of the present invention is provided below: A research team applied the design model to process model conversion method in the manufacturing process of aircraft engine blades. Aircraft engine blades are typical high-precision, high-performance components, and their design models include complex three-dimensional surfaces, precise dimensional tolerances, and strict material performance requirements. The traditional design-to-process conversion process mainly relies on manual experience, and has problems such as low efficiency, poor consistency, and difficulty in data tracing. By applying the method of the present invention, the research team built an automated and intelligent conversion system.

[0261] First, a multi-dimensional first-level transformation matrix is ​​applied to evaluate process relevance. The process relevance is calculated by analyzing the geometric features of the engine blade design model. This is shown in Table 1:

[0262] Table 1 Evaluation results of process relevance of key features of blades

[0263] Feature Number Feature Type Processing difficulty coefficient Process complexity Resource requirements Comprehensive relevance Classification results F001 leading edge of blade 0.92 0.86 0.78 0.864 High correlation F002 Blade trailing edge 0.88 0.83 0.75 0.830 Middle Association F003 Leaf root 0.95 0.91 0.89 0.920 High correlation F004 blade tenon 0.97 0.94 0.92 0.945 High correlation F005 blade flange 0.81 0.76 0.68 0.756 Middle Association F006 Cooling hole row 0.86 0.89 0.85 0.866 High correlation F007 Surface curve 0.79 0.83 0.72 0.782 Middle Association

[0264] Based on key size parameters and geometric feature extraction, the conversion reference points are determined and the confidence index is calculated, as shown in Table 2:

[0265] Table 2 Blade design feature conversion confidence evaluation

[0266] Feature Type Feature recognition accuracy Parameter extraction completeness Benchmark stability Confidence Indicator leading edge of blade 0.97 0.94 0.92 0.946 Blade trailing edge 0.95 0.91 0.90 0.922 Leaf root 0.98 0.95 0.96 0.966 blade tenon 0.99 0.97 0.96 0.975 blade flange 0.94 0.89 0.88 0.907 Cooling hole row 0.96 0.93 0.91 0.936 Surface curve 0.92 0.90 0.87 0.900

[0267] Secondly, a multi-dimensional second-level conversion matrix is ​​constructed to convert the design BOM into a process PBOM. Using a decision tree algorithm, the original component structure is reorganized to generate a process-oriented component decomposition structure. Table 3 shows the results of the material and process node matching evaluation:

[0268] Table 3 Blade material and process node matching evaluation

[0269] Material number Heat treatment matching Processability matching Surface treatment matching Comprehensive matching degree Matching Status M001 0.92 0.89 0.85 0.890 pass M002 0.87 0.82 0.79 0.832 pass M003 0.76 0.74 0.81 0.767 risk M004 0.91 0.88 0.90 0.898 pass M005 0.79 0.83 0.85 0.821 pass

[0270] Next, a multi-dimensional third-level conversion matrix was constructed to achieve a deep transformation from process PBOM to manufacturing MBOM. First, historical manufacturing data was analyzed to extract the association rules between processes and resources. Then, standard and actual working hours were calculated. Table 4 shows the working hours analysis results for key blade processes:

[0271] Table 4 Analysis of working hours for key processes of blades

[0272]

[0273]

[0274] By calculating the tooling fit, the matching between the tooling fixture and the parts is evaluated. Table 5 shows the tooling fit evaluation results:

[0275] Table 5 Blade tooling fit evaluation

[0276] Tooling number Geometric matching Positioning accuracy Clamping force distribution Operational convenience Comprehensive adaptability Adaptation status T001 0.95 0.93 0.91 0.88 0.930 pass T002 0.92 0.90 0.89 0.85 0.898 risk T003 0.97 0.94 0.92 0.90 0.942 pass T004 0.91 0.88 0.86 0.82 0.882 risk T005 0.96 0.95 0.93 0.91 0.945 pass

[0277] During the conversion process, the minimum spanning tree algorithm is applied to optimize the process route planning, calculate the edge weights and generate the optimal process route combination. Table 6 shows the calculation results of the edge weights between processes:

[0278] Table 6 Calculation of edge weights between key processes

[0279] Initial process Termination process Switching time cost Equipment adjustment complexity Resource occupation conflict Comprehensive edge weight P001 P002 0.28 0.32 0.25 0.283 P002 P003 0.45 0.38 0.42 0.420 P003 P004 0.31 0.35 0.29 0.315 P004 P005 0.26 0.30 0.24 0.266 P001 P003 0.52 0.48 0.45 0.488 P002 P004 0.56 0.51 0.49 0.524 P003 P005 0.49 0.44 0.47 0.470

[0280] In the manufacturing execution phase, the Monte Carlo simulation method is applied to predict potential anomalies. Table 7 shows the results of the assembly interference probability analysis:

[0281] Table 7 Blade assembly interference probability analysis

[0282] Assembly Sample size Number of interference samples Interference probability Risk Rating A001-B001 1000 23 0.023 Low risk A002-B002 1000 142 0.142 Medium risk A003-B003 1000 187 0.187 High risk A004-B004 1000 96 0.096 Low risk A005-B005 1000 118 0.118 Medium risk

[0283] By constructing the first and second level intercorrelation matrices and the second and third level intercorrelation matrices, accurate mapping between design features and process routes, and between process routes and manufacturing resources is achieved. Table 8 shows the calculation results of some feature influence weights:

[0284] Table 8 Analysis of the weight of design features affecting process nodes

[0285] Design Features process node Impact Weight Impact intensity Change Sensitivity F001 P001 0.78 0.92 0.85 F002 P002 0.65 0.76 0.72 F003 P003 0.82 0.96 0.89 F004 P004 0.74 0.87 0.81

[0286] Finally, a fusion conversion compensation matrix is ​​constructed to optimize the conversion process. Table 9 shows the weight distribution results of the compensation strategy:

[0287] Table 9 Weight distribution of conversion compensation strategy

[0288]

[0289]

[0290] Compared with traditional means, the present invention shows significant advantages in the manufacturing process of aircraft engine blades. Traditional design-to-process conversion mainly relies on the experience of process engineers and manual compilation of process documents, with low conversion efficiency, an average time consumption of about 120 hours per piece, poor consistency, a design change transmission lag of about 48 hours, and a conversion accuracy of about 85%. By applying the method of the present invention, the conversion efficiency is improved by 78%; the design change response time is shortened to less than 5 hours, the conversion accuracy is improved to 89.5%, and the abnormality detection rate reaches 94.3%. Especially when responding to design changes, traditional methods require re-manual analysis and process compilation, while the present invention can quickly locate the scope of influence through the inter-correlation matrix, automatically generate change response plans, and greatly improve the efficiency of product iteration.

[0291] It should be noted that the variables involved in the present invention are explained in detail as shown in Tables 10, 11 and 12 below.

[0292] Table 10 Variable Explanation Table (Part 1)

[0293]

[0294]

[0295] Table 11 Variable Explanation Table (Part 2)

[0296]

[0297] Table 12 Variable Explanation Table (Part 3)

[0298]

[0299] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A method for rapid conversion from a design model to a process model, characterized in that: include: Construct a multi-dimensional first-level conversion matrix, classify the drawing model data in the design model according to process relevance, extract key dimensional parameters, geometric features, and assembly relationship data, establish conversion reference points, and form a structured data set; Construct a multi-dimensional second-level conversion matrix to map the design BOM data to the process planning route after cleaning; Construct a multi-dimensional third-level conversion matrix, associate the process planning route with the process resource library, and achieve a deep conversion from process PBOM to manufacturing MBOM; Calculate the three-level conversion anomaly index matrix to identify and analyze anomalies that occur during the conversion process; Construct the first and second level intercorrelation matrices and the second and third level intercorrelation matrices to achieve the bidirectional mapping relationship between design features and process routes and the accurate mapping between process PBOM and manufacturing MBOM; A multi-objective optimization function is applied to construct a fusion conversion compensation matrix to compensate for data missing or abnormalities during the conversion process, thus realizing closed-loop self-optimization conversion from the design model to the process model.

2. The method for rapid conversion of a design model to a process model according to claim 1, characterized in that: The multidimensional first-level conversion matrix refers to a mathematical model that classifies design model data according to process relevance and converts it into a process-recognizable data structure, including the mapping relationship between design features and process features.

3. The method for rapid conversion of a design model to a process model according to claim 2, characterized in that: The process-recognizable data structure refers to the information organization form that conforms to the process system data standards after structured processing, including process feature description language, standardized expression of process parameters, standardized definition of process resources and standardized description of process operations, to ensure that the design data is correctly parsed and applied by the process system.

4. The method for rapid conversion of a design model to a process model according to claim 3, characterized in that: The multi-dimensional second-level conversion matrix refers to a mathematical model that converts the design BOM structure into the process PBOM structure, which includes conversion rules for component relationship reconstruction and process flow generation.

5. The method for rapid conversion from a design model to a process model according to claim 4, characterized in that: The conversion rules for component relationship reconstruction and process flow generation refer to a set of methods for reorganizing and redefining the component hierarchical relationships in the design BOM based on process processing logic, including component processing sequence rules, component assembly relationship rules, process flow node definition rules, and process connection rules.

6. The method for rapid conversion from a design model to a process model according to claim 5, characterized in that: The multidimensional third-level conversion matrix refers to the mathematical model that converts the process PBOM into the manufacturing MBOM, which contains the specific conversion rules for process resource allocation and production execution parameter generation.

7. The method for rapid conversion from a design model to a process model according to claim 6, characterized in that: The specific conversion rules for process resource allocation and production execution parameter generation refer to a set of algorithms and strategies that map the abstract resource requirements at the process planning level to the actual available resource instances at the manufacturing site, including equipment selection rules, tooling and fixture matching rules, production line balancing rules, and manufacturing parameter concretization rules.

8. The method for rapid conversion from a design model to a process model according to claim 7, characterized in that: The three-level conversion anomaly index matrix includes: the first-level conversion anomaly index matrix, the second-level conversion anomaly index matrix, and the third-level conversion anomaly index matrix, which are used to quantitatively evaluate the anomalies that occur in the conversion process from design model to process feature, the conversion process from design BOM to process PBOM, and the conversion process from process PBOM to manufacturing MBOM, respectively.

9. The method for rapid conversion of a design model to a process model according to claim 8, characterized in that: The multi-objective optimization function is used to balance multiple objectives such as conversion accuracy, conversion efficiency, resource consumption, and manufacturing risk in the process of constructing the fusion conversion compensation matrix. The input includes conversion anomaly index, resource occupancy rate, conversion time consumption, design change range, and manufacturing capability index. The output is the comprehensively optimized compensation strategy weight distribution scheme and compensation execution priority sequence.

10. The method for rapid conversion from a design model to a process model according to claim 9, characterized in that: The first and second level correlation matrix refers to a mathematical model that establishes a bidirectional correlation between design features and process routes, and is constructed using a process feature recognition conversion network model; the specific structure of the process feature recognition conversion network model is a multi-layer feature extraction and relational reasoning architecture based on a graph neural network, including a feature encoding layer, a graph structure learning layer, a relational reasoning layer, and a feature mapping layer. The feature encoding layer is responsible for converting design features into high-dimensional vector representations, the graph structure learning layer is responsible for learning the topological relationship between design features, the relational reasoning layer is responsible for inferring the mapping rules between design features and process features, and the feature mapping layer is responsible for mapping the learned design features to corresponding process features.

Images