Multi-dimensional demand profile driven adaptive manufacturing process mix recommendation method and system

Abstract

The application relates to the technical field of manufacturing processes, and specifically discloses a multi-dimensional demand portrait driven adaptive manufacturing process mixed recommendation method and system, which comprises the following steps: firstly, a multi-dimensional design demand portrait label system covering basic attributes, resource constraints, professional requirements and environmental demands is constructed; secondly, the labels are divided into continuous variables and classification variables according to data types, clustering analysis is adopted to complete clustering and output results; thirdly, the label weight is redistributed by using an entropy weight method, a design demand feature vector is constructed, and process data is preprocessed to generate a process feature vector; fourthly, the two types of feature vectors are hierarchically fused, label weights are dynamically allocated through an attention mechanism, and an attention weighted feature vector is obtained; and finally, a mixed recommendation algorithm is adopted to complete matching, and an optimal process scheme adaptive to the design demand is output; the application effectively improves the accuracy and adaptability of process recommendation, shortens the process selection cycle, and provides practical protection for enterprises to quickly respond to market demand and consolidate the competitive advantage in the industry.

Description

Technical Field

[0001] This invention relates to the field of manufacturing process technology, and in particular to a method and system for recommending hybrid manufacturing processes driven by multi-dimensional demand profiles. Background Technology

[0002] In the new era of accelerated evolution towards intelligent, green, and customized manufacturing, the dynamic and complex nature of design requirements and the stringent environmental requirements are becoming increasingly prominent. Traditional process solution decisions rely excessively on the experience of designers and lack a systematic analysis of the requirements of the design task itself. This results in process solutions that fail to fully align with product design goals and resource constraints, thus hindering enterprises' ability to respond quickly to new markets and new orders and their potential for sustainable development.

[0003] User profiling technology, as a key method for accurately identifying user characteristics, uncovering potential needs, and building virtual prototypes, transforms complex and abstract group characteristics into a concrete and quantifiable tag system through multi-source data fusion and tag-based modeling, providing support for subsequent personalized services. This method has demonstrated powerful user understanding and behavior prediction capabilities in fields such as e-commerce, social media, and healthcare.

[0004] Furthermore, with the development of intelligent manufacturing technology, the amount of process data accumulated by enterprises is growing explosively. Against this backdrop, enterprises struggle to quickly sift through massive amounts of process data to find process solutions that match current manufacturing needs, are compatible with resource constraints, and meet low-carbon requirements, exacerbating the problem of information overload. Recommendation algorithms, as one effective way to address information overload, can uncover core correlations from massive amounts of process data, providing efficient and tailored decision support for specific needs. This method does not rely on the subjective experience of designers; instead, it establishes a deep correlation between needs and processes based on feature matching of process design requirement profiles and group pattern mining, thereby accurately recommending process solutions that meet technical indicators, resource constraints, and low-carbon requirements for different demand groups. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for hybrid recommendation of adaptive manufacturing processes driven by multi-dimensional demand profiles. It proposes a research main line of characterizing design demand features, identifying demand groups through clustering, and recommending hybrid process solutions. This effectively improves the accuracy and adaptability of process recommendations, shortens the process selection cycle, and provides a practical guarantee for enterprises to quickly respond to market demands and consolidate their competitive advantages in the industry.

[0006] On the one hand, it provides a hybrid recommendation method for adapting manufacturing processes driven by multi-dimensional demand profiles, including: A multi-dimensional design requirement profile labeling system is constructed based on the basic attributes, resource constraints, professional requirements, and environmental needs of the process to be recommended. Based on the attribute characteristics of the multidimensional design requirements profile tags for manufacturing processes, the tags for basic attributes, resource constraints, professional requirements, and environmental requirements are divided into continuous variables and categorical variables according to data type; cluster analysis is used to cluster the continuous variables and categorical variables to obtain the clustering results. The design requirement labels are reassigned to the clustering results using the entropy weight method, and a feature vector is constructed to obtain the design requirement feature vector. At the same time, the process scheme data is preprocessed and the processed process features are combined into a process feature vector. The design requirement feature vector and the process feature vector are hierarchically fused and the weight of the design requirement label is dynamically allocated using an attention mechanism to obtain the attention-weighted design requirement feature vector. A hybrid recommendation algorithm is used to perform hybrid recommendation matching on the attention-weighted design requirement feature vector to obtain the optimal process solution that fits the design requirements.

[0007] On the other hand, it provides a hybrid recommendation system for adapting manufacturing processes driven by multi-dimensional demand profiles, including: The module constructs a multi-dimensional design requirement profile tagging system based on the basic attributes, resource constraints, professional requirements, and environmental needs of the process to be recommended. The clustering analysis module, based on the attribute characteristics of the multi-dimensional design requirements profile tags of the manufacturing process, divides the tags of basic attributes, resource constraints, professional requirements, and environmental requirements into continuous variables and categorical variables according to data type; and uses clustering analysis methods to cluster the continuous variables and categorical variables to obtain the clustering results. The hybrid recommendation module uses entropy weighting to reassign the weights of design requirement labels to the clustering results and constructs feature vectors to obtain design requirement feature vectors. Simultaneously, it preprocesses the process scheme data and combines the processed process features into a process feature vector. The design requirement feature vector and the process feature vector are then hierarchically fused, and an attention mechanism is used to dynamically assign weights to the design requirement labels to obtain attention-weighted design requirement feature vectors. Finally, a hybrid recommendation algorithm is used to perform hybrid recommendation matching on the attention-weighted design requirement feature vectors to obtain the optimal process scheme that best matches the design requirements.

[0008] Furthermore, an electronic device is also provided, including: Memory, used for non-transitory storage of computer-readable instructions; and Processor, for executing the computer-readable instructions, When the computer-readable instructions are executed by the processor, they perform the method described in the first aspect above.

[0009] In another aspect, a storage medium is also provided for non-transitory storage of computer-readable instructions, wherein when the non-transitory computer-readable instructions are executed by a computer, the method described in the first aspect is performed.

[0010] The above technical solution has the following advantages or beneficial effects: This invention addresses the mixed characteristics of categorized and continuous variables in design requirement profiles by proposing a clustering method that integrates Gower distance and K-prototypes. This method effectively identifies the core characteristics and adaptation patterns of design requirement groups, providing a group characteristic basis for accurate matching of subsequent requirements and solutions. Through this clustering method, enterprises can quickly categorize new order requirements and, when allocating production resources, develop targeted solutions based on the characteristics of different requirement groups, reducing resource mismatch losses.

[0011] This invention addresses the shortcomings of traditional recommendation algorithms, such as their difficulty in capturing the complex relationship between design requirements and process solutions, and their low recommendation accuracy. It proposes an attention-driven demand-solution hybrid recommendation algorithm. This algorithm dynamically highlights key demand tags (such as "priority" for urgent tasks and "carbon emission indicators" for low-carbon requirements) by introducing an attention mechanism. Simultaneously, it combines deep learning, collaborative filtering, and content matching to achieve comprehensive recommendations, ensuring efficient and accurate matching between process solutions and demand features. Based on the core attributes and constraints of design requirements, this algorithm can quickly recommend suitable process solutions for enterprises, avoiding production delays and cost waste caused by inappropriate solutions. It helps enterprises respond quickly to different types of orders, improve order delivery efficiency and customer satisfaction, and enhance their core competitiveness. Attached Figure Description

[0012] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0013] Figure 1 This is a schematic diagram of the overall process of the multi-dimensional demand profile-driven adaptive manufacturing process hybrid recommendation method in this embodiment. Figure 1 ; Figure 2 This is a schematic diagram of the overall process of the multi-dimensional demand profile-driven adaptive manufacturing process hybrid recommendation method in this embodiment. Figure 2 ; Figure 3 This is a diagram of the multi-dimensional design requirement profile tagging system constructed in the multi-dimensional requirement profile-driven adaptive manufacturing process hybrid recommendation method of this embodiment. Figure 4 This diagram illustrates the source of design requirement profile tag data in the multi-dimensional requirement profile-driven adaptive manufacturing process hybrid recommendation method of this embodiment. Figure 5This is a flowchart of the clustering analysis process that integrates Gower distance and K-prototypes algorithm in the multi-dimensional demand profile-driven hybrid recommendation method for adaptive manufacturing processes in this embodiment. Figure 6 This is a flowchart of the hybrid recommendation algorithm in the multi-dimensional demand profile-driven adaptive manufacturing process hybrid recommendation method of this embodiment; Figure 7 This is a data graph of the classification variable labels for the design requirement profile in the multi-dimensional requirement profile-driven hybrid recommendation method for adaptive manufacturing processes in this embodiment; wherein, Figure 7 (a) Complexity, (b) Maturity, (c) Priority, (d) Standardization, (e) Equipment Resources, (f) Auxiliary Materials, (g) Personnel Skills, (h) Site Environment, (i) Testing Resources, (j) Material Processing Performance, (k) Geometric Feature Adaptability, and (l) Low-Carbon Process Requirements. Figure 8 This is a trend chart of the profile coefficients for different cluster numbers in the multi-dimensional demand profile-driven hybrid recommendation method for adaptive manufacturing processes in this embodiment. Figure 9 This is a comparison chart of cluster parameter performance results in the hybrid recommendation method for adaptive manufacturing processes driven by multi-dimensional demand profiles in this embodiment; Figure 10 This is a comparison chart of MAE results in the multi-dimensional demand profile-driven adaptive manufacturing process hybrid recommendation method of this embodiment; Figure 11 This is a comparison chart of the accuracy results of the multi-dimensional demand profile-driven adaptive manufacturing process hybrid recommendation method in this embodiment. Figure 12 This is a comparison chart of recall results in the multi-dimensional demand profile-driven hybrid recommendation method for matching manufacturing processes in this embodiment. Figure 13 This is a comparison chart of F1 scores in the multi-dimensional demand profile-driven hybrid recommendation method for manufacturing processes in this embodiment. Detailed Implementation

[0014] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0015] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the invention. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0016] In this embodiment of the invention, "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, in the description of this invention, "multiple" refers to two or more.

[0017] Furthermore, to facilitate a clear description of the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0018] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0019] Example 1 Against the backdrop of deep integration with intelligent manufacturing, the manufacturing industry faces significant challenges in improving efficiency and achieving low-carbon transformation. As the market evolves towards multi-variety, small-batch, and short-cycle production, manufacturing enterprises urgently need to gain a competitive edge by rapidly responding to order changes. Process design, as the initial decision-making stage in the manufacturing process, directly determines the operational efficiency of the entire chain, including subsequent processing, assembly, and logistics. The generation of process solutions relies heavily on the systematic capture and in-depth analysis of design requirements during the manufacturing process. Therefore, how to efficiently, timely, and accurately match and output process solutions that meet these requirements has become an urgent demand for manufacturing enterprises to enhance their core competitiveness. Currently, process design requirements originate from diverse product orders and manufacturing tasks, inherently possessing characteristics such as multi-source, dynamic, and heterogeneous nature. Traditional human-centered design models struggle to systematically capture and characterize these requirements, leading to insufficient adaptability of process solutions and delayed order response. To address this, this invention shifts to a demand-centered analysis approach, supporting accurate process solution recommendations through deep coupling of design requirements and process solutions. This method provides support for manufacturing enterprises to overcome process design pain points and strengthen their rapid response capabilities under the new circumstances.

[0020] Accurate recommendation of process design schemes mainly relies on two important prerequisites: first, how to systematically deconstruct the process design requirements in the manufacturing process and build a multi-dimensional and structured process design requirement profile system to lay the foundation for accurate characterization of requirement features and discovery of group patterns; second, how to design reliable adaptive process recommendation algorithms to fully explore the complex relationship between requirement features and process schemes, and support efficient and accurate process scheme recommendations.

[0021] User profiling technology, as a key method for accurately identifying user characteristics, uncovering potential needs, and constructing virtual prototypes, transforms complex and abstract group characteristics into a concrete and quantifiable tag system through multi-source data fusion and tag-based modeling, providing support for subsequent personalized services. This method has demonstrated powerful user understanding and behavior prediction capabilities in fields such as e-commerce, social media, and healthcare. For example, some existing technologies target e-commerce products, constructing user profiles using user demographic information, user ratings, and content information. By maximizing posterior probability, they learn the potential factors of users and items in both domains to achieve reasonable product recommendations. Others mine user interests from web browsing behavior, generating domain-interest-based user profiles, which are then applied to personalized recommendations on social media, solving the problem of accurately depicting user profiles in niche areas due to limited user numbers and insufficient information. Still others accurately describe elderly users through four dimensions: demographics, society, consumption, and health, and construct user profile models by embedding tags, using these embedded user profile models to recommend medical services.

[0022] In recent years, user profiling technology has gradually extended into the manufacturing sector. For example, some existing technologies target production equipment, comprehensively characterizing its quality and safety status through a tagging system, and mining hidden data value based on this to achieve equipment accident and failure correlation, safety status early warning, and quality and safety prediction. Others target users, separating and recovering underlying user experience data from online reviews, proposing a faceted conceptual model to clarify key factors of user experience and support UX-centric product design activities. Still others analyze the behavioral characteristics, preferences, and needs of a company's electric vehicle users, quantifying the profiling features during user participation in services, and proposing incentive pricing strategies for electric vehicle users participating in automatic power generation control services based on user profiles.

[0023] It is not difficult to find that user profiling studies in the manufacturing field mostly target non-demand entities such as equipment, or analyze consumer behavior and preferences, making it difficult to depict the relationship between demand and solutions, thus limiting the accuracy of process solution recommendations. Furthermore, under the background of green manufacturing, environmental constraints such as low carbon and energy consumption have become rigid requirements for process design. Existing research often uses low carbon indicators as evaluation criteria for process solutions, without incorporating them into the profiling dimension of process design needs. This leads to a disconnect between demand characterization and low carbon goals, making it difficult to support the recommendation of low-carbon adapted process solutions. Therefore, how to construct a multi-dimensional process design demand profiling system that integrates constraints such as low carbon and energy consumption for the processing and manufacturing process, and systematically explore the key attributes and core demands of the demand group, becomes a key entry point to support deep coupling between demand and process, and improve the efficiency and adaptability of process design.

[0024] Furthermore, with the development of intelligent manufacturing technology, the amount of process data accumulated by enterprises has exploded. Against this backdrop, enterprises struggle to quickly sift through massive amounts of process data to find process solutions that match current processing and manufacturing needs, are compatible with resource constraints, and meet low-carbon requirements, exacerbating the problem of information overload. Recommendation algorithms, as one effective way to address information overload, can mine core correlation patterns from massive amounts of process data, providing efficient and adaptable decision support for specific needs. This method does not rely on the subjective experience of designers but rather establishes a deep correlation between needs and processes based on feature matching and group pattern mining of process design needs profiles, thereby accurately recommending process solutions that meet technical indicators, resource constraints, and low-carbon requirements for different groups of needs. Currently, common recommendation algorithms are mainly divided into four categories: content-based recommendation algorithms, collaborative filtering recommendation algorithms, learning model-based recommendation algorithms, and hybrid recommendation algorithms. Content-based recommendation algorithms can fully utilize the inherent characteristics of items to recommend new items with high similarity to users' preferred features, and the recommendation results have good interpretability. Among existing recommendation algorithms, some techniques propose content recommendation algorithms that integrate semantic information, comprehensively utilizing word frequency and semantic information between feature words to represent the content information of items from different perspectives for recommendation. However, this algorithm mainly relies on historical data for modeling user preferences, and its recommendation performance is poor when there are new users or significant changes in user preferences. Collaborative filtering recommendation algorithms can discover users' potential interests without requiring extensive manual annotation of item features. Other algorithms, to address the temporal and dynamic effects of user-item interactions, propose a collaborative filtering model that incorporates temporal effects, utilizing matrix factorization methods in collaborative filtering to achieve dynamic recommendations. However, this method suffers from data sparsity and cold-start problems, making it difficult to handle the characteristics of rapidly updated and highly sparse process data.

[0025] Furthermore, to handle large-scale data and delve deeper into the complex relationships between users and items, researchers have proposed recommendation algorithms based on learning models. For example, some existing techniques propose a recommendation method based on structural component processing technology knowledge using a large language model, while others propose a personalized recommendation method for manufacturing services based on a federated learning framework, which recommends manufacturing services by analyzing manufacturing service rating data. However, this method has a complex training process and requires significant computational resources. Therefore, to overcome the limitations of single recommendation algorithms, researchers have proposed hybrid recommendation algorithms. For instance, existing techniques propose a hybrid developer recommendation algorithm that combines explicit and implicit features, integrating supervised learning and collaborative filtering to complete task-specific developer recommendations. This algorithm combines the advantages of multiple recommendation algorithms, improving the accuracy and robustness of recommendations.

[0026] Given the multi-source, heterogeneous, massive, and dynamically changing characteristics of design requirement profiles in the manufacturing process, and the high-dimensional, complex, and frequently updated nature of process data, traditional single recommendation models struggle to achieve accurate matching between process solutions and design requirements. Furthermore, existing recommendation methods fail to fully integrate the commonalities and individual differences among design requirements, limiting the accuracy and generalization ability of the recommendation results. Therefore, how to leverage hybrid recommendation algorithms to better adapt to the needs of process solution recommendation scenarios based on design requirement profile clustering, and to provide higher-quality and more personalized manufacturing process recommendation services, has become a critical issue that urgently needs to be addressed.

[0027] To address the aforementioned issues, this invention proposes a multi-dimensional demand profile-driven hybrid recommendation method and system for manufacturing processes, achieving deep coupling between demand and process. Specific details include: (1) constructing a multi-dimensional design demand profile tagging system covering basic attributes, resource constraints, professional requirements, and environmental needs to accurately depict the key attributes and core features of design demands; (2) proposing a clustering algorithm integrating Gower distance and K-prototypes to effectively process categorical and continuous mixed-type feature data in demand profiles, deeply identifying the core demands and adaptation patterns of different demand groups; (3) proposing an attention mechanism-driven hybrid recommendation algorithm that dynamically captures the complex nonlinear relationship between design demand features and processes based on deep learning, integrating content recommendation and collaborative filtering mechanisms to achieve accurate and efficient process solution adaptation recommendation. Finally, the method is validated using the processing of wind turbine gearbox housings as an example. This method effectively improves the accuracy and adaptability of process recommendations, shortens the process selection cycle, and provides a practical guarantee for enterprises to quickly respond to market demands and consolidate their competitive advantages in the industry.

[0028] This embodiment provides a hybrid recommendation method for manufacturing processes driven by multi-dimensional demand profiles. First, the system analyzes the processing and manufacturing scenarios, constructs a multi-dimensional user profile of process design requirements, and accurately extracts the core demand features of different processing tasks. Second, it integrates Gower distance and K-prototypes clustering algorithms to accurately classify processing tasks with similar demand patterns, fully exploring the commonalities among demand groups. Finally, it proposes a hybrid recommendation algorithm for processing processes driven by an attention mechanism, which combines the demand features of processing tasks to accurately recommend suitable process solutions.

[0029] like Figure 1-2 As shown, the steps of the multi-dimensional demand profile-driven adaptive manufacturing process hybrid recommendation method in this embodiment include: S01: Construct a multi-dimensional design requirement profile tagging system based on the basic attributes, resource constraints, professional requirements, and environmental needs of the process to be recommended.

[0030] To comprehensively depict the process design requirements in the manufacturing process, this invention constructs a multi-dimensional design requirement profile based on the constraints and objectives of the manufacturing task. This profile accurately represents the task's functional requirements, resource conditions, and low-carbon orientation, providing data support for subsequent requirement clustering analysis and recommended adaptive process solutions. Specifically: First, a multi-dimensional design requirement profile labeling system is constructed. To systematically represent the core characteristics and constraints of process design requirements, following the process design logic of "requirement-constraint-goal," this invention constructs a design requirement profile labeling system covering four dimensions: basic attributes of the process to be recommended, resource constraints, professional requirements, and environmental requirements. This system abstracts and quantifies the objective characteristics and subjective goals of the requirements.

[0031] The multi-dimensional design requirement profile tagging system includes both static basic information about the requirements (basic attributes) and the practical constraints of process implementation (resource constraints). It also focuses on the core technical objectives of the process solution (professional requirements) and emerging requirements in the context of low-carbon manufacturing (environmental requirements). This comprehensive system captures the key characteristic dimensions of processing design requirements within the manufacturing scenario, providing structured and implementable requirement feature support for subsequent precise matching of process solutions. The multi-dimensional design requirement profile tagging system is as follows: Figure 3 As shown.

[0032] Basic attributes describe fundamental information about the design task, including static data such as complexity, maturity, priority, standardization level, and order quantity, used to construct the initial outline of the requirements profile. Complexity is categorized into four types: simple parts, moderately complex parts, complex parts, and ultra-complex parts. Maturity is divided into three categories based on design innovation level and technology iteration stage: innovative design, adaptive design, and mature design. Priority is ranked according to time and importance, categorized into regular orders, urgent tasks, high-value customer orders, and strategic R&D. Standardization level covers non-standard parts, enterprise standard parts, and industry standard parts. Order quantity is represented by actual data. These basic attribute dimensions characterize the strategic positioning and production context of the design task, serving as the top-level input for process route planning.

[0033] Resource constraints focus on the various resource conditions required to achieve the design task, covering dimensions such as equipment resources, auxiliary materials, personnel skills, site environment, and testing resources. Equipment resources include three categories: general-purpose machine tools, CNC machine tools, and machining centers; auxiliary materials are divided into special cutting fluids, general cutting fluids, and environmentally friendly auxiliary materials; personnel skills are divided into senior technicians, intermediate technicians, junior technicians, and general workers; site environments include three categories: clean rooms, explosion-proof rooms, and general workshops; testing resources include high-precision testing, routine testing, and manual testing. Resource constraints depict the resource limitations and needs for achieving design requirements, helping to understand the characteristics of resource allocation for design requirements.

[0034] Professional requirements, focusing on indicators such as accuracy, surface quality, material machinability, and geometric feature adaptability, precisely describe the technical objectives that the design task must achieve. Accuracy is categorized into five types based on tolerance values: ultra-high accuracy, high accuracy, relatively high accuracy, medium accuracy, and low accuracy. Surface quality is categorized into three types based on surface roughness: high, medium, and low. Material machinability is categorized into easy-to-machinable, difficult-to-machinable, and special-machinable types based on material properties. Geometric feature adaptability is categorized into three types based on the specific requirements of the part structure for the process: high adaptability, medium adaptability, and low adaptability. These professional requirements define the technical standards and performance thresholds that the process scheme must meet, serving as the core technical input driving process decisions.

[0035] Environmental requirements focus on the green orientation of the design task in terms of carbon emission targets, energy consumption limits, and low-carbon process requirements. Carbon emission targets and energy consumption limits are based on actual numerical limits and are divided into high, medium, and low levels. Low-carbon process requirements are classified into high, medium, and low categories according to the company's relevant low-carbon manufacturing standards and specifications. By systematically depicting the task's external compliance requirements and intrinsic value orientation for green and low-carbon manufacturing, the environmental requirements guide the optimization of process solutions towards environmentally friendly and resource-saving directions.

[0036] Furthermore, to ensure the integrity, accuracy, and timeliness of the design requirement profile tagging data, this invention adopts a multi-source heterogeneous data fusion strategy, constructing a data collection system from the perspectives of the enterprise's internal core business systems, customer requirements and enterprise standard documents, and external data. Internal core business systems include Product Lifecycle Management (PLM) systems, Enterprise Resource Planning (ERP) and Manufacturing Execution System (MES), and the enterprise's internal manufacturing resource database, primarily acquiring data through API interfaces or database exports; customer requirements and enterprise standard documents are parsed using natural language processing technology; external data is obtained by crawling publicly available information and then filtering it by experts, with specific sources and methods as follows... Figure 4 As shown.

[0037] Based on the characteristics of the design requirement profile tagging system, the data sources are divided as follows: (1) Product Lifecycle Management System. This system carries the core data of product design. The basic attributes of the design task (such as complexity, maturity, and standardization) are obtained automatically through the analysis of the product structure within the system: complexity is initially calculated based on the hierarchy of the Bill of Materials (BOM) and the number of features in the 3D model, and then calibrated in conjunction with expert confirmation in the design approval process; maturity is determined based on the product version number and change records, such as the newly released V1.0 or the adaptively improved V2.1; standardization is determined by querying the system's component library and matching whether it is a standard part number or has a standard drawing number. At the same time, the 3D model and engineering drawings managed by the system provide data support for the professional requirements dimension: the accuracy level is automatically identified by analyzing the tolerance code and value of the drawing, the surface quality is determined based on the roughness symbol Ra value, and the geometric feature adaptability is evaluated by analyzing the structural features of the model, such as thin walls and deep holes. In addition, based on the material grade in the PLM bill of materials, it can be linked to an external material knowledge base and automatically classified into "easy to process" and "difficult to process" types. All data is securely extracted through the standardized API interface provided by the PLM system. The transmission process uses HTTPS encryption combined with token authentication to ensure the uniqueness, consistency and access security of the data source.

[0038] (2) Enterprise Resource Planning and Manufacturing Execution System. ERP and MES systems record detailed information on orders and production plans, which is crucial for obtaining order quantity, priority, and auxiliary materials. Order quantity is directly derived from the "Order Quantity" field in ERP sales orders or production orders; priority is determined by a combination of customer classification in ERP (such as strategic customer marking), order delivery dates, and emergency insertion flags in the MES production plan. Auxiliary material requirements can be obtained from the cutting fluid type field specified in the ERP Manufacturing Bill of Materials (MBOM). Data acquisition uses a combination of real-time interface and batch export. Dynamic data related to production scheduling is obtained in real time through the MES API, while static data such as order attributes can be directly exported from the ERP database.

[0039] (3) Internal Manufacturing Resource Database. The internal manufacturing resource database integrates the company's equipment, personnel, and site records, primarily serving the purpose of retrieval of resource constraint labels. Equipment resource constraints are determined by querying machine tool models, precision grades, and processing ranges in the equipment database and matching them with process route suggestions; personnel skill requirements are mapped based on historical records of skill levels required for processing similar parts; and site environment requirements are extracted from environmental conditions defined in equipment records or special process specifications. This database provides a query interface through the company's internal network service to ensure the accuracy and synchronization of resource status information.

[0040] (4) Customer Requirements and Enterprise Standard Documents. These documents contain external requirements and internal compliance requirements, providing environmental and testing resource requirements data for requirement profiling. Natural language processing (NLP) is used to analyze customer requirement documents, extracting unstructured requirements such as carbon emission indicators, energy consumption limits, and customer-specified testing requirements. Simultaneously, based on enterprise-issued production technology standards and other normative documents, the classification of different low-carbon process requirements is clarified. The acquisition of this type of unstructured data is primarily achieved through the standardized services of the enterprise content management platform and document management system.

[0041] (5) External Data. This data source is used to supplement the values ​​of some tags. The core is publicly available external data, including government-issued industry carbon emission quota standards, industry analysis reports, competitor process design case studies, and other publicly available data related to carbon emissions. Data acquisition adopts the method of "web crawling + expert review". Web crawling technology is used to selectively crawl publicly available information from industry reporting platforms, government open databases, and patent search systems. The crawled raw data is further reviewed and screened by experts in the field, and finally, reasonable information that is highly related to the design requirement profile tags is retained.

[0042] After completing the collection of multi-source raw data, the raw dataset is prone to problems such as missing data, numerical noise, and abnormal invalid data due to factors such as data collection delays and manual input errors. To ensure data quality, this invention preprocesses and cleans the multi-source raw data. Specifically, for missing numerical data, mean imputation is used for repair; for discrete data with abnormal fluctuations, nearest neighbor imputation is used to reduce noise interference, and obviously illogical invalid values ​​are directly removed. After standardization preprocessing, the initially messy raw data can be transformed into a standardized analysis dataset with a regular structure, reliable quality, and effective features, effectively ensuring the accuracy of subsequent algorithm analysis and process recommendation results. Since the above data cleaning and preprocessing methods adopt mature and common technologies in this field...

[0043] S02: Based on the attribute characteristics of the multi-dimensional design requirements profile tags for manufacturing processes, the tags for basic attributes, resource constraints, professional requirements, and environmental requirements are divided into continuous variables and categorical variables according to data type; cluster analysis is used to cluster the continuous variables and categorical variables to obtain the clustering results.

[0044] Cluster analysis is used to cluster continuous and categorical variables to obtain clustering results. Specifically, after quantifying the weight of each label using the entropy weight method, the weight of each label is multiplied by the local distance of different samples with the same label and the result is weighted and summed to obtain the Gower distance. The K-prototypes algorithm is used for clustering iteration. During the iteration, the Gower distance is used to measure the similarity between the sample and the initial cluster center, and the sample is assigned to the cluster with the smallest distance. Then, the cluster centers are updated. The above process is repeated until the change in cluster centers is less than the threshold or the maximum number of iterations is reached, and the final clustering result is output.

[0045] Before performing cluster analysis, the weights of each label for both continuous and categorical variables are quantified using the entropy weight method to obtain the weights of each label.

[0046] To address the mixed characteristics of categorical variables (e.g., complexity, maturity) and continuous variables (e.g., order quantity, precision) in multidimensional design requirement profiles, this invention proposes a clustering analysis method integrating Gower distance and the K-prototypes algorithm. First, entropy weighting is used to quantify the contribution of each label to the clustering results, improving the objectivity and rationality of weight allocation. Second, Gower distance is used to uniformly measure the sample similarity of mixed-type variables, and the distances of different types of variables are integrated through weighted aggregation, overcoming the limitation of traditional Euclidean distance being only applicable to continuous data. Finally, the K-prototypes algorithm is used to perform clustering iterations. This algorithm integrates the mean clustering mechanism of K-means for continuous variables and the mode clustering ability of K-modes for categorical variables, achieving accurate grouping of mixed features by dynamically updating the cluster prototypes (mean for continuous variables, mode for categorical variables). This method ensures the consistency between the clustering results and data distribution characteristics, and improves the business interpretability of cluster features by introducing knowledge from the process design domain. The specific process is as follows: Figure 5 As shown.

[0047] Specifically, Step 1: Classify the types of feature variables. Based on the attribute characteristics of the multi-dimensional design requirements profile tags for manufacturing processes, the four categories of tags—basic attributes, resource constraints, professional requirements, and environmental requirements—are classified into continuous variables and categorical variables according to data type.

[0048] Continuous variables include order quantity, precision, surface quality, carbon emission indicators, and energy consumption restrictions. Categorical variables include complexity, maturity, priority, standardization level, equipment resources, auxiliary materials, personnel skills, site environment, testing resources, material processing performance, geometric feature adaptability, and low-carbon process requirements.

[0049] To facilitate subsequent analysis, ordinal coding is used to process data with inherent hierarchical relationships in the design requirement profile classification variables, including complexity, maturity, and priority. For example, in complexity, simple components, medium-complex components, complex components, and ultra-complex components have a clear hierarchical order, coded as 1, 2, 3, and 4 respectively, reflecting their order from low to high. One-hot coding is used to encode variables without order constraints in the design requirement profile classification variables, including equipment resources and testing resources. One-hot coding mainly involves setting an N-bit status register to encode N states. Each state has an independent register bit, and at any given time, only one bit is valid, i.e., the corresponding bit is 1, and the other bits are 0. For example, the "Equipment Resource" sub-attribute in the resource constraint attribute includes three values: conventional machine tool, CNC machine tool, and machining center. The vector encoding is a 3-dimensional vector. When the attribute value is "conventional machine tool," the vector encoding is {1,0,0}; when the attribute value is "CNC machine tool," the vector encoding is {0,1,0}; and when the attribute value is "machining center," the vector encoding is {0,0,1}. The encoding of each category variable in the design requirement profile is shown in Table 1. Table 1. Classification Variables for Design Requirement Profile

[0050] Step 2: Quantify label weights using the entropy weight method. To quantify the contribution of each label to the clustering results, the entropy weight method is used to determine the weights, enhancing the objectivity of the clustering. First, continuous variables and ordinal coded categorical variables are standardized using Z-score, as shown in the following formula: (1) in, Let $\frac{ ... Let j be the value of the j-th label of the i-th sample. Let be the sample standard deviation of all samples for the j-th label.

[0051] The categorical variables with one-hot encoding have been transformed into 0-1 binary variables through standardization, so no further standardization is needed; the 0-1 values ​​can be directly retained. Then, a standardized decision matrix is ​​constructed. R =( r ij ) m*n After normalization, the proportion matrix is ​​obtained. The formula is as follows: (2) in, This represents the relative importance of the i-th sample on the j-th label.

[0052] The entropy value of the j-th label is obtained by solving the problem. And through information utility value Obtain the final weight The specific formula is as follows: (3) (4) Where k is the correction coefficient. A lower entropy value for a label indicates stronger discriminative power and higher weight.

[0053] Step 3: Measure sample similarity based on Gower distance. Gower distance overcomes the limitations of Euclidean distance by weightedly fusing the differences between continuous and categorical variables. Define samples. and distance The formula is as follows: (5) in, Let the weight of the j-th label be . For the sample on the j-th label and For local distance, the standardized absolute distance is used for continuous variables and ordinal coded categorical variables, and the matching degree distance is used for one-hot coded variables (0 for the same category and 1 for different categories). The formula is as follows: (6) (7) in, The value of the j-th label of the i-th sample is the sample value. The value taken on the j-th continuous variable / sequential coded categorical variable. Indicates sample The value taken on the j-th continuous variable / sequential coded categorical variable. and These represent the global maximum and minimum values ​​of the j-th variable, respectively.

[0054] Step 4: Use the K-prototypes algorithm for clustering iteration. During the iteration process, use the Gower distance to measure the similarity between the sample and the initial cluster center to achieve clustering of mixed feature data.

[0055] Initialize cluster centers by randomly selecting k samples as initial cluster centers (prototype). P 1, P 2, ..., P k Each prototype contains the mean of a continuous variable and the mode of a categorical variable. For each sample... Calculate its Gower distance d with all prototypes. , P t ),Will Assign to the cluster with the smallest distance The formula is as follows: (8) in, d gower ( x i , P t ) is the sample x i With prototype P t The Gower distance, where k is the preset number of clusters.

[0056] Furthermore, the clustering prototype is updated, and the centers are recalculated for each cluster t. The updates for continuous and categorical variables are shown below: (9) (10) Where, n t Let be the total number of samples within the t-th cluster. Let represent the continuous variable portion of the t-th cluster prototype, where cluster_t is the set of all samples contained in the t-th cluster. Let be the categorical variable part of the t-th cluster prototype, and Mode be the mode function, which returns the most frequent value in the set. For sample x i The categorical variable part.

[0057] Repeat steps 3-4 until the change in cluster centers is less than the threshold ε or the maximum number of iterations is reached, and then output the final clustering result.

[0058] Step 5: Evaluate the clustering results. The silhouette coefficient method is used to measure the clustering effect of the samples, and the formula is as follows: (11) in, a ( x i ) is the sample x i The average Gower distance to other samples in the same cluster. b ( x i ) is the sample x i The average Gower distance to the nearest heterocluster s ( x i )∈[-1,1], the closer the value is to 1, the better the clustering effect.

[0059] Finally, the characteristics of each cluster are analyzed, the mean and mode distribution of the labels of each cluster are analyzed, the clusters are named in combination with the knowledge of process design, and the data characteristics and patterns of design requirements are presented intuitively through graphical means (such as heat maps), transforming abstract data such as design requirement profile labels and clustering results into visual information.

[0060] S03: The design requirement labels are reassigned using the entropy weighting method to construct a feature vector, resulting in a design requirement feature vector. Simultaneously, the process scheme data is preprocessed, and the processed process features are combined into a process feature vector. The design requirement feature vector and the process feature vector are hierarchically fused, and the weights of the design requirement labels are dynamically assigned using an attention mechanism to obtain an attention-weighted design requirement feature vector. A hybrid recommendation algorithm is used to perform hybrid recommendation matching on the attention-weighted design requirement feature vector to obtain the optimal process scheme that matches the design requirements.

[0061] The design requirement feature vector and the process feature vector are hierarchically fused and the weights of the design requirement labels are dynamically assigned using an attention mechanism to obtain an attention-weighted design requirement feature vector. Specifically, the design requirement feature vector and the process feature vector are first concatenated and merged using primary fusion to obtain a joint feature vector; then, the joint feature vector is nonlinearly mapped using deep fusion to obtain a deep fusion feature; using the deep fusion feature as the query, the attention weights of each design requirement label are calculated based on the attention mechanism; finally, the original design requirement features are weighted and summed using the attention weights to obtain the attention-weighted design requirement feature vector.

[0062] Based on the established multi-dimensional design requirement profile tagging system, this invention proposes an attention-driven hybrid recommendation algorithm for processing technology. This algorithm achieves the recommendation of suitable processing solutions through a process including design requirement feature extraction, process feature extraction, attention-based feature interaction and fusion, hybrid recommendation algorithm, and result output. Specifically, as follows... Figure 6 As shown in the diagram. The design requirement feature extraction primarily refines label weights based on clustering results; the process feature extraction focuses on representing the vector information of the process scheme; feature interaction and fusion based on the attention mechanism can deeply explore the complex relationship between design requirements and process attributes; and the hybrid recommendation algorithm combines the advantages of deep learning, collaborative filtering, and content recommendation to ultimately output the optimal process scheme that meets the design requirements. This method achieves precise process scheme recommendation by deeply coupling design requirement profiles and process features, combining the common features and differences of requirements within the same cluster. The specific process is as follows: (1) Extraction of design requirements features Based on the different cluster groups obtained by the clustering algorithm, the weights of the design requirement labels are further refined. For each cluster, the weight of each label is recalculated using the entropy weight method for the sample data within that cluster. For example, if the maturity level in a cluster is generally high, then the label's distinguishability within that cluster is relatively low, and its label weight will decrease accordingly; if the low-carbon process requirement label shows significant differences between different design requirements within the cluster, i.e., strong distinguishability, then the weight will increase. In this way, the importance of the design requirement characteristics within each cluster can be more accurately reflected.

[0063] Based on the refined label weights, a feature vector is constructed for each design requirement, resulting in a design requirement feature vector. Each label value is multiplied by its corresponding weight, and then the weighted values ​​of all labels are combined into a vector, which serves as the feature representation of the design requirement. For example, requirement The feature vector can be represented as ,in It is the first Tag weight, It is a demand In the The values ​​that can be taken from each label.

[0064] (2) Extraction of process features Collect enterprise process plan data, including process type, process parameters, equipment type, and information such as process carbon emissions, processing time, and processing costs. For continuous data (such as process parameters and carbon emissions), use the same standardization method as the continuous variables in the design requirement profile to process it, ensuring it has the same scale. For categorized process features (such as equipment type and process type), use one-hot encoding to convert them into numerical form for easier subsequent calculations. Further, combine the processed process features into a process feature vector P. For example, the feature vector of a process plan can be represented as... ,That This process scheme is in the [number]th [year]. The values ​​that can be taken on each feature.

[0065] (3) Feature interaction and fusion based on attention mechanism To deeply explore the complex nonlinear relationship between design requirements and process technology, an attention mechanism is introduced for feature interaction. First, the design requirement feature vector and process feature vector are hierarchically fused, divided into two stages: primary fusion and deep fusion. Primary fusion uses a feature concatenation method to combine the design requirement feature vector and the process feature vector... and for joint feature vectors F It preserves the explicit associations of the original features; deep fusion uses a multilayer perceptron to process the joint feature vector. FA nonlinear mapping is performed to capture the implicit interaction between design requirements and process technology, as shown in the following formula: (12) (13) In the formula, U represents the design requirement feature vector, P represents the process feature vector, and ⊕ represents the vector concatenation operation. w 1, w 2 are the weight matrices of the hidden layer and the output layer, respectively. b 1, b 2 represents the bias terms for the hidden layer and the output layer, respectively. F is the ReLU activation function. deep This is a feature of deep fusion.

[0066] Furthermore, an attention mechanism is introduced to dynamically allocate the weights of design requirement labels, highlighting labels that significantly impact process recommendation while weakening secondary features. The deeply fused feature vector is used as the query, and attention is calculated between it and the original design requirement feature vector to dynamically learn the importance of each design requirement label for the current process recommendation task. Specifically: (14) In the formula, a j This represents the attention weight of the j-th design requirement label; u j This is represented as the key vector representation of the j-th label in the design requirements; score ( F deep , u j ) represents the scoring function used to calculate the query vector F. deep and key vector u j The higher the score, the more relevant the design requirement feature is to the current process recommendation context.

[0067] The original features of the design requirements are weighted and summed using attention weights to obtain the attention-weighted feature vector of the design requirements. U att It contains information recommended for the current process scenario.

[0068] (15) In the formula, U att The design requirement feature vector after attention weighting.

[0069] Furthermore, a hybrid recommendation algorithm is employed to perform hybrid recommendation matching on the attention-weighted design requirement feature vector to obtain the optimal process solution that best matches the design requirements. Specifically, a deep learning model is used to match the attention-weighted design requirement vector and the process feature vector to obtain the deep learning matching degree; a collaborative filtering method is used to calculate the historical scores of the process solution for the design requirements within the cluster to obtain the average preference score; and based on the explicit matching between the design requirement profile tags and the process features, a weighted cosine similarity is used to calculate the content matching degree. The deep learning matching degree, average preference score, and content matching degree are weighted and fused to obtain the final recommendation score. The recommendation scores are sorted in descending order, and the top-N solutions with the highest scores are selected to obtain the process solution that best meets the design requirements.

[0070] (4) Hybrid Recommendation Algorithm This stage employs a hybrid recommendation strategy, calculating the demand-process matching degree from different dimensions and merging them to generate the final recommendation score. Specifically: Deep learning matching. A bidirectional long short-term memory network is used to construct a subnetwork for extracting design requirement features, and the input is an attention-weighted design requirement vector. U att Output high-dimensional latent vectors H u Furthermore, a process feature extraction subnetwork is constructed using a convolutional neural network. Local correlations of process features are extracted through 1D convolutional layers, outputting a process latent vector. H p As shown in equation (16): (16) In the formula, σ represents the activation function, and p is the process feature vector. w c , b c These represent the weights and biases of the convolutional layer.

[0071] Furthermore, the latent vector of design requirements is calculated using cosine similarity. H u With process implicit vectors H p Matching degree S deep To measure the potential relationship between the two: (17) Item-based collaborative filtering. Item-based collaborative filtering methods utilize the collective behavior of the cluster to make recommendations, identifying other historical design solutions highly similar to the process scheme matching the current design requirement, and pushing these highly similar solutions as recommendations to the current design requirement. This is done for the cluster C to which the design requirement belongs. kThe average preference score of the design requirements within the cluster for the process scheme is calculated using the following formula: (18) In the formula, The historical score (1-5 points, combined with the adoption rate and satisfaction) of process scheme p is given to design requirement i. For clusters Group preference score for process scheme p.

[0072] Content-based recommendation. Based on explicit matching of design requirement profile tags and process features, a weighted cosine similarity is used to obtain the content matching degree, with the formula: (19) (20) in, To determine the weighting coefficients for design requirement profile tags, Cosine similarity between design requirement characteristics and process characteristics; , The design requirements profile and process solution are respectively represented in the first... The feature vectors in the label dimension, where ||| represents the Euclidean norm of the vector.

[0073] The final recommendation score is obtained by weighting and merging the matching scores as described above, using the following formula: (twenty one) In the formula, , , To integrate weights, satisfy + + =1.

[0074] (5) Recommendation results output The matching degree of all candidate process solutions is calculated and sorted in descending order of score. The top-N solutions with the highest scores are selected to form a recommendation list, providing process designers with choices that meet their design requirements, thereby achieving a hybrid recommendation of suitable process solutions.

[0075] The following case analysis is based on the above methods: The wind turbine pitch gearbox is installed inside the hub and controls the rotor speed by adjusting the blade angle in real time, thereby controlling the absorbed mechanical energy. This improves wind energy utilization while preventing excessive wind force from impacting the wind turbine. This invention uses the gearbox housing of a 2.5MW wind turbine as an example to verify the effectiveness of the proposed multi-dimensional demand profile-driven adaptive manufacturing process hybrid recommendation method. The housing material is cast iron, and its weight is 470kg.

[0076] (1) Data settings This case study involves sending API data requests to CRRC Shandong Wind Power Co., Ltd.'s product lifecycle management system, enterprise resource planning and manufacturing execution system, and internal manufacturing resource database. It structures customer requirements and company standard documents, uses web scraping technology to crawl some of the company's process cards, and utilizes a wind turbine case library previously compiled and developed by the research team as data sources. This yields basic attributes, resource constraints, professional requirements, and environmental requirement tags for design requirement profiles, resulting in the collection of 368 design requirement datasets and their corresponding process design scheme datasets. The statistical data of the categorical variables in the design requirement profiles are as follows: Figure 7 As shown, Figure 7 (a) Complexity, (b) Maturity, (c) Priority, (d) Standardization, (e) Equipment resources, (f) Auxiliary materials, (g) Personnel skills, (h) Site environment, (i) Testing resources, (j) Material processing performance, (k) Geometric feature adaptability, and (l) Low-carbon process requirements.

[0077] The data for continuous variables are shown in Table 2: Table 2. Continuous variable label data for the design requirements profile.

[0078] (2) Analysis of Dataset Clustering Results Based on the aforementioned historical dataset, the experimental environment shown in Table 3 was used to perform clustering simulation analysis using the Python programming language. The clustering analysis method based on Gower distance and the K-prototypes algorithm proposed in this invention was employed to analyze all historical design requirements. Using the silhouette coefficient as the evaluation index, the number of clusters k was set to 2-10, and silhouette coefficient results for different numbers of clusters were obtained, as shown in the figure. Figure 8 As shown.

[0079] Table 3 Experimental Environment for Program Execution

[0080] Based on the characteristics of each cluster, this invention categorizes design requirements into three types: standard requirements, advanced breakthrough requirements, and innovative low-carbon requirements. Standard requirements are characterized by a preference for mature and stable design patterns, with low carbon not being a primary focus, and relatively basic overall technical requirements. These requirements are at a standard level in terms of complexity, maturity, and standardization, and are mostly for mature products in mass production. They have clear and standard requirements for equipment, personnel, and facilities, requiring no special or high-precision resources. These requirements are generally low to medium in terms of precision and surface quality, and have lower requirements for carbon emissions, energy consumption, and low-carbon processes, prioritizing production efficiency and delivery time. Advanced breakthrough requirements, on the other hand, are characterized by higher overall technical requirements and the inclusion of low carbon as a clear consideration. These requirements are mostly for adaptive designs or high-value customer orders, representing the exploratory phase of a company's green transformation. They have higher requirements for equipment resources, personnel skills, and testing resources, and are beginning to experiment with environmentally friendly auxiliary materials and other green resources. These types of demands require high precision and surface quality, while also imposing moderate requirements on carbon emission standards, energy consumption limits, and low-carbon processes. These demands serve as a bridge between conventional and innovative approaches, offering both challenges and growth potential. They require designers with potential to incorporate low-carbon concepts while ensuring quality. Innovative low-carbon demands are characterized by stringent professional technical indicators and low-carbon requirements. These demands primarily target high-value clients and strategic R&D, involving ultra-complex parts or entirely new designs with low standardization. They have high resource requirements, mainly using environmentally friendly auxiliary materials and high-precision testing equipment. These demands pursue high levels of precision, surface quality, material processing performance, and geometric adaptability, and impose high requirements on carbon emissions, energy consumption limits, and low-carbon processes. They represent the technological benchmark and future direction of enterprises and are the core vehicle for promoting low-carbon transformation.

[0081] For the new input design requirements, we analyzed them and obtained their profile tag data, as shown in Table 4: Table 4 Newly Input Design Requirement Profile Tag Data

[0082] Furthermore, the cluster to which the new design requirement belongs was determined using the prototype matching method based on Gower distance, and the results are shown in Table 5: Table 5 New Design Requirements Data and Distance from Gower to Each Cluster Center

[0083] As shown in Table 5, the new design requirement data has the smallest Gower distance with cluster 2, indicating the highest similarity and belonging to the advanced breakthrough requirement category. Cluster classification and categorization based on design requirement profiles can quickly identify the key characteristics of design requirements and the common demands of corresponding requirement groups, laying a data foundation for subsequent targeted screening, optimization of process solutions, and the recommendation of reasonable and efficient solutions.

[0084] (3) Analysis of the results of the recommended scheme 1) Analysis of Recommendation Results Based on the collected processing technology dataset, and using the obtained carbon emissions, costs, and processing time as evaluation indicators, the mapping relationship between design requirements and processes was analyzed through feature interaction and fusion using an attention mechanism. With fusion weights of λ1=0.3, λ2=0.4, and λ3=0.3, a hybrid recommendation algorithm was used to obtain the final recommendation results. The similarity scores and recommendation results are shown in Table 6. The final scores were calculated using the hybrid recommendation algorithm to obtain the top 5 process design schemes that meet the design requirements. The recommendation results for each design scheme are 42, 32, 102, 89, and 109, respectively.

[0085] Table 6. Recommended Process Results

[0086] In Table 6, Sdeep represents the matching degree obtained based on deep learning matching; Score represents the average preference score of cluster design requirements for process solutions obtained based on collaborative filtering of items; and Scontent represents the content matching degree obtained based on content recommendation.

[0087] 2) Recommendation algorithm performance evaluation Furthermore, the mean absolute error (MAE), precision, recall, and the comprehensive metric F1 score are used to evaluate the performance of the process solution recommendation algorithm based on the design requirement profile. The mean absolute error is the average of the absolute deviations between the recommended solution and the core matching index of the design requirement. It can intuitively reflect the degree of fit between the recommended solution and the actual requirements. The smaller the MAE value, the higher the matching degree of the algorithm to the requirements and the better the performance. Precision represents the proportion of solutions in the recommended solution list that match the design requirements. That is, the ratio of the solutions in the recommended solution list that meet the requirements to all solutions in the recommended solution list, as shown in Equation (22). Recall represents the probability of recommending all solutions that meet the design requirements. That is, the ratio of the solutions in the recommended solution list that meet the design requirements to the total number of solutions that meet the design requirements, as shown in Equation (23). The F1 score is used to comprehensively evaluate precision and recall, as shown in Equation (24). Among them, the larger the values ​​of recall, precision, and F1 score, the better the recommendation effect.

[0088] (twenty two) (twenty three) (twenty four) (25) In the formula, N represents the number of samples; y represents the predicted result value; T(u) represents the actual result value; T(u) represents the recommended solution that meets the design requirements; R(u) represents the total number of solutions that meet the design requirements obtained according to the recommendation algorithm.

[0089] To verify the effectiveness of the clustering method proposed in this invention for the overall recommendation results, the evaluation results were further analyzed under recommendation scheme N=5 and different cluster types k. The results are as follows: Figure 9 Show.

[0090] pass Figure 9 As the number of clusters k increases from 2 to 10, the MAE initially decreases slightly, then shows an overall upward trend, indicating that the prediction error is better when k is smaller. Precision reaches its peak of 0.388 at k=3, then fluctuates and decreases, suggesting that appropriate clustering helps improve recommendation accuracy. Recall remains relatively stable, between 0.526 and 0.590. The F1 score is optimal at k=3, then gradually decreases. Overall, when the number of clusters k=3, the MAE is lowest, while Precision, Recall, and F1 scores all reach their peak, resulting in optimal comprehensive performance. Too many or too few clusters lead to performance degradation. This result is consistent with the optimal number of clusters obtained through Gower distance and K-prototypes algorithm analysis, verifying the effectiveness of the proposed clustering method. As the k value continues to increase, all indicators show varying degrees of decline, indicating that excessive clustering can actually lead to a decline in recommendation performance. Therefore, the clustering method proposed in this invention can construct well-discriminate process scheme categories, providing a high-quality clustering foundation for recommendation systems.

[0091] To verify the recommendation performance of the proposed hybrid recommendation algorithm, it was compared and analyzed with individual content-based recommendation, user-based collaborative filtering, item-based collaborative filtering, and FunkSVD matrix factorization-based recommendation algorithms. The recommendation results considered TOP-N, with N=5, 10, 15, 20, and 25, a total of five cases. The evaluation results for each recommendation algorithm are as follows: Figure 10-13 As shown.

[0092] Comparative analysis of the results shows that, under different lengths of recommended suggestions, the recommendation algorithm proposed in this invention is superior to other algorithms in terms of the accuracy and precision of the recommendation results. Figure 10For the MAE result comparison analysis, it can be seen from the figure that the hybrid recommendation algorithm proposed in this invention achieves the lowest MAE, indicating that its prediction error is lower than other algorithms. For example, when N=5, its MAE is reduced by a maximum of 5.69% compared to other algorithms, demonstrating superior accuracy and stability. Since the growth rate of the number of suitable solutions meeting design requirements in the recommendation results is lower than the growth rate of all recommended process solutions, therefore... Figure 11 The accuracy rate shows a decreasing trend from high to low. Analysis reveals that the recommendation algorithm proposed in this invention significantly improves accuracy compared to algorithms based on users, content, items, and matrix factorization, indicating that it can more effectively capture design requirement characteristics and provide more precise and suitable process solution recommendations for design tasks. Figure 12 The recall results in Figure 11 The opposite trend is because as the total number of recommended process solutions increases, the number of compatible solutions that meet the design requirements covered by the algorithm also increases simultaneously. Although the growth rate of the total number of recommendations is faster than the growth rate of effective solutions, the absolute value of successfully discovered compatible solutions is increasing, which leads to an increase in recall rate.

[0093] The hybrid algorithm proposed in this invention, by accurately modeling design requirement characteristics and fusing requirement cluster technology preference identification with comprehensive matching degree, expands the recommendation coverage while more efficiently capturing potential suitable solutions. Therefore, its recall rate not only increases with the total number of recommendations but also outperforms traditional algorithms based on users, content, items, and matrix factorization. The F1 score is an evaluation metric that combines precision and recall. Figure 13 All recommendation algorithms show a pattern of high F1 scores followed by low F1 scores. The recommendation algorithm of this invention outperforms other recommendation algorithms in terms of overall F1 score.

[0094] To achieve rapid generation and accurate response of manufacturing process solutions, and to address the challenges of diverse processing task requirements, complex constraints, high environmental demands, and difficulties in efficiently matching them with manufacturing resources, this invention, following the logical framework of "accurate characterization of design requirements - mining of clustering patterns - hybrid recommendation of process solutions," has conducted research on a method for constructing design requirement profiles and recommending suitable solutions for the manufacturing process. The main results are as follows: (1) Breaking through the limitations of traditional single-dimensional demand analysis, this approach extracts the commonalities and differences in design requirements, constructs a multi-dimensional labeling system for design requirements covering basic attributes, resource constraints, professional requirements, and environmental requirements. It integrates demand attributes, resource matching, performance targets, and low-carbon requirements into a unified framework, transforming design requirements from abstract order descriptions into a concrete and quantifiable set of labels. This labeling system clearly defines the core attributes and constraints of different requirements, providing precise feature support for subsequent demand clustering and solution matching.

[0095] (2) To address the mixed characteristics of categorization and continuous variables in design requirement profiles, a clustering method integrating Gower distance and K-prototypes is proposed. This method effectively identifies the core characteristics and adaptation patterns of design requirement groups, providing a basis for accurate matching of requirements and solutions. Through this clustering method, enterprises can quickly categorize new order requirements and formulate targeted solutions based on the characteristics of different requirement groups when allocating production resources, thereby reducing resource mismatch losses.

[0096] (3) To address the problems of traditional recommendation algorithms' difficulty in capturing the complex relationship between design requirements and process solutions, and their low recommendation accuracy, an attention-driven demand-solution hybrid recommendation algorithm is proposed. This algorithm introduces an attention mechanism to dynamically highlight key demand tags (such as "priority" for urgent tasks and "carbon emission indicators" for low-carbon requirements). Simultaneously, it combines deep learning with collaborative filtering and content matching to achieve comprehensive recommendations, ensuring efficient and accurate matching between process solutions and demand characteristics. This algorithm can quickly recommend suitable process solutions for enterprises based on the core attributes and constraints of design requirements, avoiding production delays and cost waste caused by inappropriate solutions. It helps enterprises respond quickly to different types of orders, improve order delivery efficiency and customer satisfaction, and enhance their core competitiveness.

[0097] Example 2 This embodiment provides a multi-dimensional demand profile-driven adaptive manufacturing process hybrid recommendation system, including: The module constructs a multi-dimensional design requirement profile tagging system based on the basic attributes, resource constraints, professional requirements, and environmental needs of the process to be recommended. The clustering analysis module, based on the attribute characteristics of the multi-dimensional design requirements profile tags of the manufacturing process, divides the tags of basic attributes, resource constraints, professional requirements, and environmental requirements into continuous variables and categorical variables according to data type; and uses clustering analysis methods to cluster the continuous variables and categorical variables to obtain the clustering results. The hybrid recommendation module uses entropy weighting to reassign the weights of design requirement labels to the clustering results and constructs feature vectors to obtain design requirement feature vectors. Simultaneously, it preprocesses the process scheme data and combines the processed process features into a process feature vector. The design requirement feature vector and the process feature vector are then hierarchically fused, and an attention mechanism is used to dynamically assign weights to the design requirement labels to obtain attention-weighted design requirement feature vectors. Finally, a hybrid recommendation algorithm is used to perform hybrid recommendation matching on the attention-weighted design requirement feature vectors to obtain the optimal process scheme that best matches the design requirements.

[0098] Example 3 This embodiment also provides an electronic device, including: one or more processors, one or more memories, and one or more computer programs; wherein, the processor is connected to the memory, and the one or more computer programs are stored in the memory. When the electronic device is running, the processor executes the one or more computer programs stored in the memory to cause the electronic device to perform the method described in Embodiment 1.

[0099] Example 4 This embodiment also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the method described in Embodiment 1.

[0100] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-dimensional requirement profile driven adaptive manufacturing process mix recommendation method, characterized in that, include: A multi-dimensional design requirement profile labeling system is constructed based on the basic attributes, resource constraints, professional requirements, and environmental needs of the process to be recommended. Based on the attribute characteristics of the multidimensional design requirements profile tags for manufacturing processes, the tags for basic attributes, resource constraints, professional requirements, and environmental requirements are divided into continuous variables and categorical variables according to data type. Cluster analysis was used to cluster continuous and categorical variables to obtain clustering results. The design requirement labels are reassigned to the clustering results using the entropy weight method, and a feature vector is constructed to obtain the design requirement feature vector. At the same time, the process scheme data is preprocessed and the processed process features are combined into a process feature vector. The design requirement feature vector and the process feature vector are hierarchically fused and the weight of the design requirement label is dynamically allocated using an attention mechanism to obtain the attention-weighted design requirement feature vector. A hybrid recommendation algorithm is used to perform hybrid recommendation matching on the attention-weighted design requirement feature vector to obtain the optimal process solution that fits the design requirements.

2. The multi-dimensional demand-portfolio-driven, tailored manufacturing process mix recommendation method of claim 1, wherein: The method employs cluster analysis to cluster continuous and categorical variables, obtaining clustering results. Specifically, after quantifying the weight of each label using the entropy weight method, the weight of each label is multiplied by the local distance of different samples with the same label and then summed in weights to obtain the Gower distance. The K-prototypes algorithm is used for iterative clustering. During the iteration process, the Gower distance is used to measure the similarity between the sample and the initial cluster center, and the sample is assigned to the cluster with the smallest distance. Then, the cluster centers are updated. The above process is repeated until the change in cluster centers is less than a threshold or the maximum number of iterations is reached, and the final clustering result is output.

3. The multi-dimensional demand-portfolio-driven, tailored manufacturing process mix recommendation method of claim 1, wherein: The step involves hierarchically fusing the design requirement feature vector and the process feature vector, and dynamically allocating the weights of the design requirement labels using an attention mechanism to obtain an attention-weighted design requirement feature vector. Specifically, this involves: firstly, using primary fusion to concatenate and merge the design requirement feature vector and the process feature vector to obtain a joint feature vector; and then using deep fusion to perform nonlinear mapping on the joint feature vector to obtain a deep fused feature. Using deep fusion features as queries, attention weights for each design requirement label are calculated based on an attention mechanism. Finally, the original design requirement features are weighted and summed using the attention weights to obtain the attention-weighted design requirement feature vector.

4. The method of claim 1, wherein: The process involves using a hybrid recommendation algorithm to perform hybrid recommendation matching on the attention-weighted design requirement feature vector to obtain the optimal process solution that best matches the design requirements. Specifically, this involves using a deep learning model to match the attention-weighted design requirement vector and the process feature vector to obtain the deep learning matching degree. The collaborative filtering method was used to calculate the historical scores of the design requirements within the cluster on the process scheme, and the average preference score was obtained. Based on the explicit matching of design requirement profile tags and process features, the content matching degree is calculated using weighted cosine similarity. The final recommendation score is obtained by weighted fusion of deep learning matching degree, average preference score and content matching degree. The recommended scores are sorted in descending order, and the top-N solutions with the highest scores are selected to obtain the optimal process solution that best meets the design requirements.

5. The method of claim 1, wherein: Before performing cluster analysis, the weights of each label for both continuous and categorical variables are quantified using the entropy weight method to obtain the weights of each label.

6. The multi-dimensional demand-portfolio-driven, custom manufacturing process mix recommendation method of claim 1, wherein: The continuous variables include order quantity, accuracy, surface quality, carbon emission indicators, and energy consumption limits.

7. The method of claim 1, wherein: The classification variables include complexity, maturity, priority, degree of standardization, equipment resources, auxiliary materials, personnel skills, site environment, testing resources, material processing performance, geometric feature adaptability, and low-carbon process requirements.

8. A multi-dimensional requirement profile driven adaptive manufacturing process mix recommendation system characterized in that, include: The module constructs a multi-dimensional design requirement profile tagging system based on the basic attributes, resource constraints, professional requirements, and environmental needs of the process to be recommended. The clustering analysis module, based on the attribute characteristics of the multi-dimensional design requirements profile tags of the manufacturing process, divides the tags of basic attributes, resource constraints, professional requirements, and environmental requirements into continuous variables and categorical variables according to data type; and uses clustering analysis methods to cluster the continuous variables and categorical variables to obtain the clustering results. The hybrid recommendation module uses the entropy weight method to reallocate the weights of the design requirement labels in the clustering results and constructs a feature vector to obtain the design requirement feature vector; at the same time, it preprocesses the process scheme data and combines the processed process features into a process feature vector. The design requirement feature vector and the process feature vector are hierarchically fused and the weight of the design requirement label is dynamically allocated using an attention mechanism to obtain an attention-weighted design requirement feature vector. A hybrid recommendation algorithm is then used to perform hybrid recommendation matching on the attention-weighted design requirement feature vector to obtain the optimal process scheme that matches the design requirements.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps in the multi-dimensional demand profile-driven adaptive manufacturing process hybrid recommendation method as described in any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the multi-dimensional demand profile-driven adaptive manufacturing process hybrid recommendation method as described in any one of claims 1 to 7.

Images