Artificial intelligence-based industrial design review method and system

Abstract

The application provides an industrial design review method and system based on artificial intelligence, which comprises the following steps: performing design intention semantic analysis on industrial design drawings to generate design intention semantic model data, and forming a unified review semantic framework by constructing a multi-dimensional constraint knowledge graph; generating design elements and constraint correlation mapping data based on the unified review semantic framework, and generating design intention differentiation result data by identifying intentional design features and negligent design defects through design intention differentiation processing; generating optimization scheme evaluation report data based on the design intention differentiation result data and the multi-dimensional constraint knowledge graph, and performing final review decision to obtain final design review result data; and generating industrial design comprehensive review data based on the final design review result data and risk early warning and improvement suggestion generation processing. By constructing a unified review semantic framework, the design intention, manufacturing constraints and user requirements are integrated, and the comprehensiveness and accuracy of the industrial design review are improved.

Description

Technical Field

[0001] This invention proposes an industrial design review method and system based on artificial intelligence, belonging to the field of artificial intelligence-assisted design technology. Background Technology

[0002] In the field of industrial design, traditional review methods face numerous challenges. Current AI review systems for industrial design suffer from fundamental flaws, with severely fragmented review dimensions. CAD plugins often focus solely on geometric interference or DFM (Design for Manufacturing) rule checks, while user experience tools independently perform tasks such as eye-tracking and gesture simulation. The lack of a unified semantic framework to integrate design intent, manufacturing constraints, and user needs makes comprehensive and accurate reviews difficult.

[0003] The feedback is also extremely static, mostly presented as an "error list," such as indicating "wall thickness < 1mm," but it cannot provide editable optimization suggestions or alternative solutions. At the same time, the system cannot distinguish between "intentionally thin walls" for aesthetic reasons and design errors such as "neglecting thin walls," resulting in a high false alarm rate.

[0004] Furthermore, existing AI functionalities are limited, only capable of classification and detection, lacking the ability to generate new solutions that conform to the original design language. Moreover, most publicly available solutions remain at the level of "AI automatically annotating drawings" and "DFM rule engines," without any deep integration of Large Language Models (LLM), 3D Generative Diffusion Models (3DDiffusion), and manufacturing knowledge graphs to construct an active design review agent with a closed loop of "understanding-questioning-generation-verification." This makes it difficult to become a "design collaborator with engineering empathy and creative capabilities," and thus fails to meet the demands of efficient and high-quality review in modern industrial design. Summary of the Invention

[0005] This invention provides an artificial intelligence-based industrial design review method and system to solve the problems mentioned in the background section above: The present invention proposes an artificial intelligence-based industrial design review method, which includes: performing semantic analysis of industrial design drawings to generate semantic model data of design intent, and forming a unified review semantic framework by constructing a multi-dimensional constraint knowledge graph; generating design element and constraint association mapping data based on the unified review semantic framework, and identifying intentional design features and neglected design defects through design intent differentiation processing to generate design intent differentiation result data; generating optimization scheme evaluation report data based on design intent differentiation result data and multi-dimensional constraint knowledge graph, and making a final review decision to obtain final design review result data; and generating comprehensive industrial design review data based on final design review result data and risk warning and improvement suggestions generation processing.

[0006] The industrial design review system based on artificial intelligence proposed in this invention includes: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method described in any one of the above.

[0007] The beneficial effects of this invention are as follows: By constructing a unified review semantic framework that integrates design intent, manufacturing constraints, and user needs, the comprehensiveness and accuracy of industrial design reviews are improved, breaking the traditional fragmented review dimensions. The use of design intent for differentiated processing reduces false alarm rates, accurately distinguishing between intentional design features and overlooked design defects, avoiding wasted design resources due to misjudgments. The generation of manufacturable optimization solutions using a large language model and a 3D generative diffusion model enhances the innovation and feasibility of designs, providing designers with more high-quality options. It reduces the time cost of repeated manual design modifications, improving development efficiency. Simultaneously, AR / VR immersive feedback allows designers to interactively verify optimization solutions, avoiding later manufacturing problems caused by unreasonable solutions. This invention not only achieves a paradigm shift from passive compliance checks to proactive collaborative optimization but also significantly improves design quality, providing strong support for next-generation AI-driven product innovation and promoting the intelligent and efficient development of industrial design. Attached Figure Description

[0008] Figure 1 The method steps provided in the embodiments of the present invention are illustrated.

[0009] Figure 2 The method steps are shown in another embodiment of the present invention. Detailed Implementation

[0010] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0011] like Figure 1As shown, the present invention proposes an artificial intelligence-based industrial design review method, which includes: performing semantic analysis of industrial design drawings to generate semantic model data of design intent, and forming a unified review semantic framework by constructing a multi-dimensional constraint knowledge graph; generating design element and constraint association mapping data based on the unified review semantic framework, and identifying intentional design features and neglected design defects through design intent differentiation processing to generate design intent differentiation result data; generating optimization scheme evaluation report data based on design intent differentiation result data and multi-dimensional constraint knowledge graph, and making a final review decision to obtain final design review result data; and generating comprehensive industrial design review data based on final design review result data and risk warning and improvement suggestions generation processing.

[0012] Preferably, generating design intent semantic model data includes: collecting multi-format industrial design drawings, performing standardization processing, and generating a standardized industrial design drawing dataset; based on the standardized industrial design drawing dataset, using a semantic parsing model with a bidirectional attention mechanism, performing deep semantic parsing of the information in the drawings to generate design intent semantic model data.

[0013] Preferably, constructing a multidimensional constraint knowledge graph includes: extracting core constraint elements from design intent semantic model data using knowledge graph construction tools to generate a multidimensional constraint element dataset; cleaning and structuring the multidimensional constraint element dataset to generate a structured constraint element dataset; using a graph neural network to construct a node and edge association model based on the structured constraint element dataset, quantitatively modeling the hierarchical subordinate relationships and horizontal cross-influence relationships between constraint elements to generate a constraint element association matrix; and combining the constraint element association matrix with preset conflict resolution logic rules to construct a multidimensional constraint knowledge graph.

[0014] Preferably, generating the constraint element association matrix includes: extracting the unique identifier, attribute features, and application scenario labels of each constraint element based on the structured constraint element dataset to construct a constraint element node set; for the constraint element node set, using the hierarchical modeling module of a graph neural network to mine the subordinate relationships of elements within the same constraint category to generate a hierarchical association weight matrix; based on the constraint element node set and the hierarchical association weight matrix, using the cross-association module of the GNN to analyze the mutual influence between elements of different constraint categories, calculate the cross-influence coefficient, and generate a horizontal association weight matrix; and fusing the hierarchical association weight matrix and the horizontal association weight matrix to construct a complete node and edge association model, transforming the association quantification results into a standardized matrix form to generate the constraint element association matrix.

[0015] Another embodiment of the present invention, such as Figure 2 As shown, an artificial intelligence-based industrial design review method includes: S1. Perform semantic analysis of industrial design drawings to generate semantic model data of design intent; construct a multi-dimensional constraint knowledge graph containing manufacturing constraints, user needs and aesthetic rules based on the semantic model data of design intent, thereby forming a unified review semantic framework. S2. Based on the unified review semantic framework, conduct design element correlation analysis and use the three-dimensional generation diffusion model to perform initial compliance checks on industrial design drawings to obtain preliminary design compliance data and potential problem area identification data; perform design intent differentiation processing on the potential problem area identification data to identify intentional design features (such as aesthetic thin walls) and negligent design defects (such as manufacturing risk thin walls), and generate design intent differentiation result data. S3. Based on the design intent, differentiate the result data, combine the multidimensional constraint knowledge graph, and use the Large Language Model (LLM) to generate manufacturable optimization solution data that conforms to the original design language and satisfies manufacturing constraints; perform multi-solution comparison and evaluation on the manufacturable optimization solution data, and generate optimization solution evaluation report data. S4. Utilize the optimization scheme evaluation report data to perform AR / VR immersive feedback processing on industrial design drawings, generating immersive feedback scene data. Within this immersive feedback scene data, designers can interactively view, edit, and verify optimization schemes, generating interactive design verification data. Finally, conduct a review and decision-making process on the interactive design verification data to obtain the final design review result data. S5. Based on the final design review results data, perform design quality scoring to generate design quality score data; based on the design quality score data, perform risk warning and improvement suggestion generation to generate comprehensive industrial design review data, realizing the upgrade of the review paradigm from passive compliance inspection to proactive collaborative optimization.

[0016] The working principle and effects of the above technical solution are as follows: By semantic parsing of design intent and construction of a multi-dimensional constraint knowledge graph, design requirements and review standards are accurately aligned, significantly improving review accuracy and effectively avoiding the problem of the review being out of touch with the original design intent; Parallel compliance checks are achieved using a 3D generation diffusion model, significantly improving review efficiency and reducing the omissions and time consumption of manual checks; Design intent differentiation processing can accurately identify aesthetic designs and oversights, preventing high-quality ideas from being misjudged and rejected, and protecting the originality of the design; Optimized solutions adapted to the original design language are generated by a large language model, enhancing the collaboration between design and manufacturing and reducing the cost of later modifications; AR / VR immersive feedback allows designers to intuitively verify the solution, improving modification efficiency and verification accuracy, and reducing the communication costs of repeated modifications; Multi-role collaborative review and scientific decision-making mechanisms enhance the comprehensiveness of the review and avoid decision-making errors caused by the bias of a single role; Ultimately, a proactive collaborative optimization review paradigm is achieved, reducing design risks in the production stage, ensuring design quality, improving design iteration efficiency, and facilitating the implementation of high-quality industrial design results.

[0017] In one embodiment of the present invention, S1 includes: S11. Collect industrial design drawings in multiple formats (including CAD vector graphics, rendered images, BIM model files, etc.), and generate a standardized industrial design drawing dataset through drawing format standardization processing (e.g., format conversion, redundant information removal, key layer extraction). S12. Based on the standardized industrial design drawing dataset, a semantic parsing model with a two-way attention mechanism is used to perform deep semantic parsing of the design intent of the information in the drawings (such as structural features, functional annotations, parameter constraints, etc.) to generate fine-grained design intent semantic model data. The fine-grained design intent semantic model data includes sub-modules such as core functional requirements, aesthetic design orientation, and technical parameter boundaries. S13. Based on the semantic model data of design intent, extract the core constraint elements through the knowledge graph construction tool. The core constraint elements include manufacturing constraints (e.g., material processing limits, process feasibility thresholds), user needs (e.g., usage scenario adaptation, interaction experience requirements), and aesthetic rules (e.g., proportion coordination principles, color matching specifications), and generate a multi-dimensional constraint element dataset. S14. Utilize graph neural networks (GNNs) to model the relationships of the multidimensional constraint element dataset and construct a multidimensional constraint knowledge graph. The multidimensional constraint knowledge graph includes the hierarchical structure of constraint elements, cross-influence rules, and conflict resolution logic, ultimately forming a unified review semantic framework (achieving semantic alignment between design intent and review standards).

[0018] The working principle and effects of the above technical solution are as follows: By standardizing the processing of multi-format drawings, the compatibility problem of different types of design drawings is effectively solved, avoiding subsequent processing delays or errors caused by format chaos and redundant information interference, and significantly improving the smoothness and standardization of drawing processing; by using a semantic parsing model with a two-way attention mechanism, the design intent behind the drawings can be deeply mined, generating fine-grained semantic model data, greatly improving the accuracy of understanding the design intent, and avoiding the problem of the review being out of touch with the core design requirements; by extracting multi-dimensional constraint elements and constructing a knowledge graph, the scattered manufacturing, user, and aesthetic constraints are integrated and associated, enhancing the comprehensiveness and relevance of the review standards, and reducing the one-sidedness of the review caused by a single constraint consideration; by using GNN modeling to realize the hierarchicalization and conflict resolution of constraint elements, a unified review semantic framework is formed, effectively avoiding semantic deviations and conflicts between different review standards, ensuring the consistency of the review basis, laying a solid foundation for subsequent review stages, and reducing cross-stage communication costs and comprehension errors.

[0019] In one embodiment of the present invention, step S14 includes: S141. Clean and structure the multidimensional constraint feature dataset (e.g., remove duplicate constraint terms, add feature attribute labels, and standardize data format) to generate a structured constraint feature dataset. S142. Based on the structured constraint element dataset, a node and edge association model is constructed using a graph neural network (GNN) to quantitatively model the hierarchical subordinate relationship and horizontal cross-influence relationship between constraint elements, and generate a constraint element association matrix. S143. Combine the constraint element relationship matrix, embed the preset conflict resolution logic rules, and construct a preliminary multi-dimensional constraint knowledge graph; S144. Perform semantic alignment verification between the preliminary multidimensional constraint knowledge graph and the design intent semantic model data. Correct the graph association deviation through iterative optimization, and finally form a unified review semantic framework (to achieve semantic alignment between design intent and review standards).

[0020] The working principle and effects of the above technical solution are as follows: By cleaning and structuring the multi-dimensional constraint element dataset, duplicate and redundant items are effectively eliminated, and the data format is standardized, significantly improving the accuracy and regularity of constraint data and avoiding interference from messy data to subsequent modeling, thus laying a solid data foundation for relationship mining; by using GNN to construct node and edge association models, the hierarchical subordinate and horizontal cross relationships of constraint elements are accurately quantified, clearly presenting the inherent relationships between each constraint, enhancing the depth and comprehensiveness of constraint relationship modeling, and avoiding omissions in review criteria due to unclear relationship analysis; by embedding conflict resolution logic rules to construct a preliminary graph, potential contradictions between different constraints can be resolved in advance, reducing decision-making dilemmas caused by constraint conflicts in subsequent reviews; through alignment verification and iterative optimization with the design intent semantic model, graph association deviations are accurately corrected, and the final unified review semantic framework effectively achieves accurate alignment between design intent and review criteria, avoiding deviations in review from core requirements due to semantic understanding biases, ensuring consistency and standardization in subsequent review stages, and reducing cross-stage communication and understanding costs.

[0021] In one embodiment of the present invention, step S142 includes: Based on the structured constraint element dataset, the unique identifier, attribute features and application scenario tags of each constraint element are extracted to construct a constraint element node set, which includes sub-node clusters such as manufacturing constraint nodes, user requirement nodes, and aesthetic rule nodes. For the set of constraint element nodes, the hierarchical modeling module of graph neural network (GNN) is used to mine the subordinate relationship of elements within the same constraint category (e.g., material constraints, aluminum alloy processing constraints and aluminum alloy bending limits) and generate a hierarchical association weight matrix. Based on the constraint element node set and hierarchical association weight matrix, the cross-association module of GNN is used to analyze the mutual influence between different constraint categories of elements (such as aesthetic thin-wall design, material thickness constraints and processing technology feasibility), calculate the cross-influence coefficient, and generate the horizontal association weight matrix. By integrating the hierarchical association weight matrix and the horizontal association weight matrix, a complete node and edge association model is constructed. The quantitative results of the association relationship are transformed into a standardized matrix form, generating a constraint element association relationship matrix.

[0022] The working principle and effects of the above technical solution are as follows: By extracting the unique identifier, attribute features, and application scenario tags of constraint elements to construct a node set, and further subdividing it into sub-node clusters such as manufacturing constraints and user requirements, the scattered constraint elements are effectively sorted out, avoiding deviations in subsequent association analysis caused by chaotic element classification and unclear identification, thus laying a regular foundation for relationship modeling; by using the GNN hierarchical modeling module to mine the subordinate relationships of constraints of the same category, a hierarchical association weight matrix is ​​generated, which can clearly present the hierarchical logic between constraints (such as the progressive relationship from material constraints to specific processing constraints), greatly improving the accuracy of subordinate relationship identification and avoiding confusion in the execution of review standards due to hierarchical ambiguity; through GN The N-level cross-association module analyzes the mutual influence of constraints of different categories, calculates cross-influence coefficients to generate a horizontal association weight matrix, breaks the limitations of single-category constraint analysis, accurately captures the association between cross-category constraints (such as the association between aesthetic design and process feasibility), enhances the comprehensiveness of constraint relationship analysis, and avoids missing key cross-influence factors. The fusion of the two weight matrices generates a standardized constraint element association matrix, which quantifies the association and unifies the format, significantly reducing the difficulty and error of subsequent knowledge graph construction. It can not only ensure the integrity and accuracy of constraint relationships, but also provide reliable support for subsequent conflict resolution and semantic alignment, and improve the efficiency of the overall review framework construction.

[0023] In one embodiment of the present invention, S2 includes: S21. Based on the unified review semantic framework, a multimodal element association algorithm is adopted to perform bidirectional association analysis on the structural elements, material elements, process elements and constraint items in the multidimensional constraint knowledge graph in industrial design drawings, and generate design element and constraint association mapping data. S22. Input the design elements and constraint association mapping data into the pre-trained 3D generation diffusion model. The model performs parallel initial compliance checks on the geometric dimensions compliance, material compatibility, and process feasibility of the drawings, generating preliminary design compliance data (including a list of compliance items and details of non-compliance items) and potential problem area identification data (including problem area coordinates and associated constraint types). S23. Based on the potential problem area identification data, extract the structural feature parameters of the problem area (such as wall thickness, surface curvature, assembly gap) and perform correlation matching with the design intent semantic model data to construct a problem area and design intent mapping matrix; S24. Using a lightweight classification model, design intent differentiation processing is performed on the problem area and design intent mapping matrix. By comparing the preset intentional design feature library (e.g., aesthetic thin walls, personalized shaped grooves) with the neglected design defect library (e.g., insufficient strength rounded corners, assembly interference structures), intentional design features and neglected design defects are accurately identified, and design intent differentiation result data is generated. The design intent differentiation result data includes feature type labels, confidence scores, and related design intent descriptions.

[0024] The working principle and effects of the above technical solution are as follows: Through bidirectional analysis of the multimodal element association algorithm, it can accurately connect the structural, material, and process elements and multidimensional constraints in industrial design drawings, avoiding omissions or misjudgments in the review due to association deviations, and significantly improving the accuracy and comprehensiveness of element association; by using a pre-trained 3D generation diffusion model to achieve parallel compliance checks, it greatly improves review efficiency, reduces the time-consuming and omission-prone nature of traditional manual point-by-point checks, and can clearly identify potential problem areas and associated constraints, providing clear direction for subsequent processing; by extracting the structural features of the problem area and matching them with the semantic model of the design intent, the problem analysis is closely aligned with the original design intention, avoiding blind judgments detached from the design context; by using a lightweight classification model to accurately distinguish between intentional design features and oversight defects, it effectively avoids high-quality aesthetic ideas being misjudged as defects and rejected, protecting the originality of the design, and reducing the risk of later production due to defect omissions. The generated distinction results with confidence and intent descriptions can not only provide accurate basis for subsequent optimization, but also reduce the subjective judgment error of reviewers, improving the objectivity and reliability of the review.

[0025] In one embodiment of the present invention, step S21 includes: Based on a unified review semantic framework, core features of constraint items (such as constraint type, threshold range, and applicable scenarios) are extracted from the multidimensional constraint knowledge graph, and a standardized constraint item feature set is generated through feature encoding. Multimodal element extraction is performed on industrial design drawings to obtain the original feature data of structural elements (such as geometric topology and dimensional parameters), material elements (such as material type and performance parameters), and process elements (such as preset processing technology and assembly process), thereby generating a multimodal design element original dataset. The original dataset of multimodal design elements is cleaned and standardized (e.g., outlier removal, unified dimensions, feature dimensionality reduction) to generate a normalized multimodal design element feature set; The standardized constraint feature set and the normalized multimodal design element feature set are input into the multimodal element association algorithm. First, a forward matching is performed from design elements to constraints to calculate the feature similarity score. Then, a reverse verification is performed from constraints to design elements to supplement any missing association items. By integrating positive matching results and negative verification results, and labeling the association confidence level, design element and constraint association mapping data containing the correspondence between elements and constraints, the basis for association, and the confidence level score is generated.

[0026] The working principle and effects of the above technical solution are as follows: By extracting the core features of constraint items and encoding them to generate a standardized feature set, and simultaneously extracting, cleaning, and standardizing the multimodal elements of the design drawings, the problem of inconsistent formats and messy features between constraint items and design elements is effectively solved, significantly improving the regularity of the two types of data and clearing data obstacles for subsequent correlation analysis; the multimodal elements comprehensively cover the dimensions of structure, materials, and processes, avoiding the bias in review caused by single-element analysis and enhancing the completeness of element extraction; the bidirectional correlation mode of forward matching and reverse verification is adopted, which can accurately calculate feature similarity through forward matching and supplement missing correlation items through reverse verification, greatly improving the accuracy of the correlation between elements and constraints and avoiding the omission or mismatch problems that are easy to occur in one-way matching; after the fusion results are labeled with the correlation confidence, the generated correlation mapping data is accompanied by clear basis and confidence score, reducing the judgment bias caused by ambiguity in subsequent reviews, providing an accurate correlation basis for compliance checks, reducing the workload of reviewers in sorting out element constraint relationships, and improving the efficiency of the overall review process.

[0027] In one embodiment of the present invention, S3 includes: S31. Based on the design intent, differentiate the result data, filter out the constraint association information corresponding to the neglected design defects that need to be optimized, and combine the manufacturing constraint rules in the multi-dimensional constraint knowledge graph with the successful case library to generate a defect optimization requirement list data. S32. Input the defect optimization requirement list data, the original design intent semantic model data, and the multi-dimensional constraint knowledge graph into the fine-tuned large language model (LLM). Through prompting engineering guidance, the model generates manufacturable optimization scheme data that conforms to the original design language style (e.g., shape lines, color system), retains the intentional design features, and meets manufacturing constraints. The optimization scheme data includes multi-dimensional sub-schemes such as structural improvement schemes, process adjustment schemes, and material replacement schemes. S33. Construct a multi-dimensional solution evaluation index system, which includes manufacturing feasibility, cost control level, design style consistency, and user experience retention. Use the analytic hierarchy process (AHP) to quantify and assign weights to the manufacturable optimization solution data. S34. Based on the quantitative scoring results, conduct a multi-solution comparative evaluation, analyze the advantages and disadvantages, implementation difficulties and potential risks of each solution, and generate an optimization solution evaluation report data. The optimization solution evaluation report data includes solution ranking, key indicator comparison charts, and risk warning explanations.

[0028] The working principle and effects of the above technical solution are as follows: By accurately screening the constraint-related information corresponding to overlooked design defects, and combining manufacturing constraint rules with successful cases to generate an optimization requirement list, the optimization direction is effectively clarified, avoiding the wasted effort caused by blind optimization, and laying a solid foundation for subsequent solution generation; with the help of a finely tuned large language model, optimization solutions are generated under the guidance of prompting engineering. This not only strictly matches the original design language style and fully preserves intentional design features such as aesthetics, but also accurately meets manufacturing constraint requirements, avoiding the problem of optimization solutions being out of touch with the original design intention or failing to meet production conditions. At the same time, multi-dimensional sub-solutions cover structure, process, and other aspects. The design incorporates advanced techniques and materials, enhancing the flexibility and comprehensiveness of optimization. A multi-dimensional evaluation index system is constructed, and the analytic hierarchy process (AHP) is used for quantitative scoring, breaking the limitations of single-dimensional evaluation and significantly improving the objectivity and scientific rigor of the scheme evaluation, avoiding evaluation biases caused by subjective judgment. Through comparative analysis of multiple schemes, an evaluation report with rankings, indicator charts, and risk warnings is generated, making the advantages, disadvantages, difficulties, and risks of each scheme readily apparent. This significantly reduces the decision-making difficulty for designers and minimizes the subsequent modification costs and time losses caused by selecting the wrong scheme. It ensures the feasibility and rationality of the optimized scheme while improving the efficiency and quality of the design optimization process.

[0029] In one embodiment of the present invention, step S4 includes: S41. Based on the data from the optimization scheme evaluation report, extract the visual identifiers of the problem area, the 3D model of the optimization scheme, and the related information of the constraint basis. Use the AR / VR scene building engine to build an immersive feedback scene and generate immersive feedback scene data that integrates design drawings, optimization schemes, and knowledge graph visualization. S42. In the immersive feedback scenario data, the designer operates through an interactive terminal (such as a VR controller or AR glasses). The operation includes locating the problem area, switching the optimization scheme, and fine-tuning the design parameters. The system calls the compliance check model in real time to dynamically verify the edited scheme and generate interactive design verification data. The interactive design verification data includes editing operation records, real-time compliance check results, and designer modification opinions. S43. Based on interactive design verification data, launch the multi-role collaborative review module, invite manufacturing engineers, user representatives, and aesthetic designers to access the immersive scene through hierarchical access permissions, submit their respective review opinions, and generate a summary data of multi-role review opinions. S44. Utilize the opinion fusion algorithm to resolve conflicts and extract consensus from the aggregated data of multi-role review opinions. Combine this with the compliance verification results in the interactive design verification data to make the final review decision and obtain the final design review result data. The final design review result data includes the approval conclusion, modification suggestions, and reasons for rejection.

[0030] The working principle and effects of the above technical solution are as follows: An immersive scene integrating design drawings, optimized solutions, and knowledge graphs is built using an AR / VR scene construction engine, making problem areas, optimized solutions, and constraints intuitively visible. This effectively solves the problems of unclear understanding of solutions and ambiguous constraint relationships in traditional 2D reviews, significantly improving designers' efficiency in understanding optimized solutions. Designers can operate and fine-tune solutions in real time using interactive terminals and obtain dynamic compliance verification, enabling them to promptly identify compliance issues after modifications, avoiding the hassle of rework, reducing the time cost of repeated verification, and improving the accuracy of design verification. Multi-role collaborative reviews are initiated and permissions are supported. Tiered access allows the needs of different roles, such as manufacturing, users, and aesthetics, to be fully integrated into the review process, avoiding omissions or biased judgments caused by single-role reviews, and enhancing the comprehensiveness and objectivity of the review. Utilizing opinion fusion algorithms to resolve conflicts and extract consensus, combined with compliance verification results, makes the final decision, effectively avoiding decision-making deadlocks caused by disagreements among multiple roles, and improving the efficiency and scientific nature of review decisions. The final review results, with conclusions, suggestions, and reasons, provide designers with clear directions for improvement and ensure the traceability of review results, promoting the upgrade of reviews from one-way feedback to multi-role collaborative optimization, and improving the overall quality of design reviews.

[0031] In one embodiment of the present invention, S44 includes: The aggregated review opinions from multiple roles are preprocessed in a structured manner (e.g., text segmentation, semantic encoding, and normalization of scoring indicators) to extract the core viewpoints, evaluation dimensions, and bias labels of the opinions, thereby generating a normalized review opinion dataset. The normalized review opinion dataset is input into the opinion fusion algorithm. First, the points of conflict (such as the disagreement between manufacturing feasibility and aesthetic design) and points of consensus are identified through similarity clustering. Then, the conflict resolution calculation is performed based on role weight (such as manufacturing engineers having a higher weight for process feasibility), generating conflict resolution results and a set of consensus viewpoints. Extract real-time compliance verification results from interactive design verification data, correlate and integrate them with conflict resolution results and consensus viewpoint sets, mark the matching relationship between compliance and review opinions, and generate comprehensive review basis data; Based on comprehensive review data, a review decision rule base is constructed (e.g., consensus across all roles + compliance approval → direct approval; insufficient consensus but compliance meets standards → supplementary modification suggestions; and non-compliance meets standards → rejection with explanation of reasons), and the final review decision is executed. The review and decision results are structured and organized, supplemented with decision-making basis and summary of role opinions, and the final design review result data is generated.

[0032] The working principle and effects of the above technical solution are as follows: By structurally preprocessing the opinions of multiple roles, scattered text opinions and scoring data are effectively sorted out, core viewpoints and evaluation tendencies are extracted, and interference from messy information in subsequent analysis is avoided, significantly improving the regularity of opinion data. Utilizing the similarity clustering and role weight resolution mechanism of the opinion fusion algorithm, conflict points in different dimensions such as manufacturing and aesthetics can be accurately identified, while consensus is extracted, effectively avoiding decision-making deadlocks caused by disagreements among multiple roles, and significantly improving the efficiency and rationality of conflict resolution. The integration of real-time compliance verification results with the resolved opinions ensures that review decisions are supported by both subjective opinions and objective compliance data, avoiding the one-sidedness of relying solely on subjective judgment and enhancing the scientific nature of decision-making. A clear review decision rule base is constructed, making the final judgment systematic and avoiding arbitrariness in decision-making, improving the consistency of decisions in different design review scenarios. Finally, the decision results are structured and supplemented with a summary of supporting evidence, making the review results clear and traceable. This provides designers with clear reasons for their decisions, reduces the cost of subsequent interpretation and communication of review results, and ensures the closed-loop effectiveness of the review process.

[0033] In one embodiment of the present invention, step S5 includes: S51. Based on the final design review results data and combined with the preset design quality scoring system, the design quality scoring system includes four primary indicators and 12 secondary sub-indicators: compliance, manufacturability, aesthetic performance, and user adaptability. The fuzzy comprehensive evaluation method is used to perform quantitative scoring of design quality and generate design quality scoring data, which includes the scores of each indicator, the total score, and the grade rating. S52. Conduct risk source analysis on indicators in the design quality score data that are below the threshold, and identify potential risk points by combining the risk case library in the multidimensional constraint knowledge graph. The risk points include production delays, cost overruns and user complaints, and generate design risk warning data. S53. Based on the modification suggestions in the design risk warning data and the final design review result data, use a large language model to generate targeted improvement suggestions, which include technical improvement paths, resource allocation schemes, and cycle planning suggestions, and generate design improvement suggestion data; S54. Integrate design quality scoring data, design risk warning data, and design improvement suggestion data, organize and visualize them to generate comprehensive industrial design review data, which includes review conclusions, scoring reports, risk warnings, and improvement plans.

[0034] The working principle and effects of the above technical solution are as follows: A pre-set multi-dimensional quality scoring system, combined with fuzzy comprehensive evaluation, is used to quantitatively score design quality, covering compliance, manufacturing, aesthetics, and user adaptation. This effectively avoids the bias and arbitrariness of traditional subjective scoring, significantly improving the objectivity and accuracy of design quality assessment. Risk tracing is conducted for low-scoring indicators, and a risk case library is used to identify potential risks such as production delays and cost overruns. This provides early warning of design flaws, preventing these risks from erupting during the production stage and reducing the cost and time wasted on rework and rectification later. Targeted improvement suggestions, including technical paths and resource allocation, are generated using a large language model, making the designer's optimization direction clearer and avoiding ineffective investment caused by blind improvements, thus enhancing the operability of improvement measures. The integration of various data for structured organization and visualization makes the comprehensive review data intuitive and easy to understand. This provides a comprehensive and reliable basis for design decisions, reduces the cost of interpreting and communicating review results for different roles, facilitates the rapid implementation of review conclusions, promotes continuous improvement in design quality, and ultimately extends the value of review from result judgment to proactive quality improvement.

[0035] One embodiment of the present invention provides an artificial intelligence-based industrial design review system, comprising: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method described in any one of the above.

[0036] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An industrial design review method based on artificial intelligence, characterized in that, The method includes: Industrial design drawings are subjected to semantic analysis of design intent to generate semantic model data of design intent, and a unified review semantic framework is formed by constructing a multi-dimensional constraint knowledge graph. Based on a unified review semantic framework, design element and constraint association mapping data are generated, and design intent differentiation results data are generated by identifying intentional design features and neglected design defects through design intent differentiation processing. Based on the design intent, the result data and the multidimensional constraint knowledge graph are distinguished to generate the optimization scheme evaluation report data, and the final review decision is made to obtain the final design review result data; Based on the final design review results data and risk warnings and improvement suggestions, comprehensive industrial design review data is generated.

2. The industrial design review method based on artificial intelligence according to claim 1, characterized in that, The process of generating design intent semantic model data includes: collecting multi-format industrial design drawings, performing standardization processing, and generating a standardized industrial design drawing dataset; based on the standardized industrial design drawing dataset, using a semantic parsing model with a bidirectional attention mechanism, performing deep semantic parsing of the information in the drawings to generate design intent semantic model data.

3. The industrial design review method based on artificial intelligence according to claim 2, characterized in that, The construction of a multidimensional constraint knowledge graph includes: extracting core constraint elements from the semantic model data of the design intent using knowledge graph construction tools to generate a multidimensional constraint element dataset; cleaning and structuring the multidimensional constraint element dataset to generate a structured constraint element dataset; using the structured constraint element dataset, constructing a node and edge association model using graph neural networks to quantitatively model the hierarchical subordinate relationships and horizontal cross-influence relationships between constraint elements, generating a constraint element association matrix; and combining the constraint element association matrix with preset conflict resolution logic rules to construct a multidimensional constraint knowledge graph.

4. The industrial design review method based on artificial intelligence according to claim 3, characterized in that, A unified review semantic framework is formed; the multidimensional constraint knowledge graph and the design intent semantic model data are semantically aligned and verified, and the graph association deviation is corrected through iterative optimization, ultimately forming a unified review semantic framework.

5. The industrial design review method based on artificial intelligence according to claim 1, characterized in that, The design intent differentiation process identifies intentional design features and neglected design defects, generating design intent differentiation result data. This process includes: inputting design element and constraint association mapping data into a pre-trained 3D generative diffusion model; performing parallel initial compliance checks through the model to generate preliminary design compliance data and potential problem area identification data; based on the potential problem area identification data, extracting structural feature parameters of the problem areas and performing correlation matching with the design intent semantic model data to construct a problem area and design intent mapping matrix; and using a lightweight classification model to perform design intent differentiation processing on the problem areas and design intent mapping matrix to identify intentional design features and neglected design defects, generating design intent differentiation result data.

6. The industrial design review method based on artificial intelligence according to claim 1, characterized in that, The final design review result data includes: distinguishing the result data from the design intent and generating manufacturable optimization scheme data from the multi-dimensional constraint knowledge graph, and generating optimization scheme evaluation report data through multi-scheme comparison and evaluation; using the optimization scheme evaluation report data to perform AR / VR immersive feedback processing on the industrial design drawings to generate interactive design verification data, and making the final review decision to obtain the final design review result data.

7. The industrial design review method based on artificial intelligence according to claim 6, characterized in that, The final design review result data for final review decision-making includes: In the immersive feedback scenario data, designers operate through interactive terminals, and the system dynamically verifies the edited scheme by calling the compliance check model in real time, generating interactive design verification data; Based on the interactive design verification data, the multi-role collaborative review module is launched, and designers access the immersive scenario through hierarchical access permissions, submit their respective review opinions, and generate multi-role review opinion summary data; The opinion fusion algorithm is used to resolve conflicts and extract consensus from the multi-role review opinion summary data, and combined with the compliance check results in the interactive design verification data, the final review decision is made to obtain the final design review result data.

8. The industrial design review method based on artificial intelligence according to claim 1, characterized in that, Generate comprehensive industrial design review data: Based on the final design review results, the design quality score is processed to generate design quality score data. Based on the design quality score data, risk warnings and improvement suggestions are generated to produce comprehensive industrial design review data and design quality score data.

9. The industrial design review method based on artificial intelligence according to claim 8, characterized in that, The risk warning and improvement suggestion generation process based on design quality scoring data includes: conducting risk source analysis on indicators below the threshold in the design quality scoring data, identifying potential risk points by combining the risk case library in the multidimensional constraint knowledge graph, and generating design risk warning data; generating targeted improvement suggestions based on the modification suggestions in the design risk warning data and the final design review results data using a large language model, and generating design improvement suggestion data; and integrating the design quality scoring data, design risk warning data, and design improvement suggestion data, performing structured organization and visualization, and generating comprehensive industrial design review data.

10. An artificial intelligence-based industrial design review system, including: One or more processors; A memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method of any one of claims 1 to 9.

Images