Three-dimensional design model conformity analysis method and system based on large language model

Abstract

The invention discloses a three-dimensional design model conformity analysis method and system based on a large language model, and the method comprises the steps: obtaining the related data and design specification constraint conditions of a three-dimensional design model, and building a data interaction channel through a collaborative review platform; calling a semantic geometry collaborative verification algorithm to process the data and extracting feature parameters; generating a rule adaptation weight matrix through a space rule learning adaptation model; constructing a mapping relation between the feature vector and the design specification through a CAD point cloud feature coding mapping model; carrying out multi-dimensional conformity verification on the basis of the result; and outputting the comprehensive analysis data set. The system realizes deep cooperative processing of geometric and semantic data of a three-dimensional design model through multi-unit cooperative linkage and integration of an algorithm and a platform function, improves the accuracy of rule adaptation and feature mapping, comprehensively covers a conformity verification dimension, and meets a high-precision analysis requirement in a complex scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of product three-dimensional design analysis, and in particular to a three-dimensional design model compliance analysis method and system based on a large language model. BACKGROUND

[0002] Three-dimensional design models are increasingly widely used in complex industrial fields such as aerospace and high-end equipment manufacturing, and their design specification and compliance directly affect product research and development quality and production efficiency. With the increasing complexity of product structures and the increasing refinement of design specification systems, traditional analysis methods that rely on manual verification or single-dimensional algorithms have been unable to meet the needs of multi-source data collaborative verification and complex rule precise adaptation. Large language models, with their strong semantic understanding and multi-source data association capabilities, provide a new technical path for three-dimensional design model compliance analysis. The integration and application of core technologies such as product three-dimensional semantic geometry collaborative verification algorithms and spatial rule learning adaptation models can achieve deep association of geometric features, semantic information, and design specifications, and there is an urgent need to build a systematic analysis solution that integrates multiple algorithms and collaborative platforms to solve the efficiency and accuracy problems of three-dimensional design model multi-dimensional compliance verification.

[0003] The existing technology has two significant shortcomings: on the one hand, the geometric topology data and semantic description information of three-dimensional design models lack efficient collaborative processing mechanisms, and traditional analysis methods mostly verify geometric features or semantic information separately, failing to achieve deep association and collaborative verification of the two, resulting in deviations in the matching degree analysis of geometric constraints and semantic requirements in design specifications, making it difficult to fully cover the compliance verification dimensions of three-dimensional design models; on the other hand, there is a lack of dynamic learning and coding mapping capabilities to adapt to complex design rules, and existing models mostly use fixed rule libraries or simple coding methods, which cannot adaptively adjust according to the dynamic changes of design specifications, and the coding conversion of point cloud features of three-dimensional design models is not precise enough, leading to insufficient pertinence of rule adaptation and feature mapping, and thus affecting the reliability of compliance analysis results, which cannot meet the high-precision compliance verification needs of complex industrial product three-dimensional design models. SUMMARY

[0004] In order to overcome the shortcomings and deficiencies of the existing technology, the present application provides a three-dimensional design model compliance analysis method and system based on a large language model.

[0005] The technical solution adopted by the present application is: The three-dimensional design model compliance analysis method based on a large language model comprises the following steps: S1, obtain the geometric topology data, semantic description information, and design specification constraint conditions of the three-dimensional design model, build a multi-source data interaction channel through a product three-dimensional scene collaborative review platform, and establish an associated mapping relationship for data transmission and storage; S2, calling a product three-dimensional semantic geometry collaborative verification algorithm to cooperatively associate the obtained geometric topological data and semantic description information, extract a geometric feature point set, a semantic association factor, and a standard matching dimension parameter; S3, inputting the processed data into a spatial rule learning adaptation model to perform feature adaptation and rule mapping of design specification constraint conditions, and generating a rule adaptation weight matrix; S4, encoding and converting the geometric feature point set of the three-dimensional design model through a CAD point cloud feature coding mapping model, and constructing a mapping relationship between the feature vector space and the design specification; S5, based on the rule adaptation weight matrix output by the spatial rule learning adaptation model and the feature vector mapping relationship generated by the CAD point cloud feature coding mapping model, performing three-dimensional design model compliance multidimensional verification through the product three-dimensional semantic geometry collaborative verification algorithm; S6, outputting a three-dimensional design model compliance analysis result through a product three-dimensional scene collaborative review platform, and forming a comprehensive analysis data set including feature matching degree, rule adaptation degree, and semantic consistency.

[0006] Further, the expression of the product three-dimensional semantic geometry collaborative verification algorithm is:

[0007] wherein, is a three-dimensional semantic geometry collaborative verification coefficient, is a weight factor of the i-th geometric feature, is an i-th geometric topological feature parameter, is an i-th semantic description association parameter, is a specification constraint adjustment coefficient, is a design specification constraint parameter, is a topological structure association matrix, is a j-th normalized dimension weight, is a j-th data normalization parameter, is a collaborative verification correction coefficient, is a special verification parameter, n is the total number of geometric features, and m is the total number of normalized dimensions.

[0008] Further, the expression of the spatial rule learning adaptation model is:

[0009] wherein, is a spatial planning learning adaptation degree, is a k-th rule weight parameter, is a k-th learning iteration parameter, is a k-th rule mapping coefficient, For the q-th specification, adapt the weights. For the q-th design specification parameter, To adapt to the correction factor, For adaptive adjustment parameters, p is the total number of rules, and r is the total number of norm parameters.

[0010] Furthermore, the expression for the CAD point cloud feature encoding mapping model is:

[0011] in, The mapping value is encoded as the point cloud feature. Let be the weight of the s-th point cloud feature. Let s be the feature parameters of the s-th point cloud. Let be the eigenvalues ​​of the s-th vector. For the s-th eigenvector parameter, For the s-th encoding transformation matrix, For the u-th encoding weight, For the u-th encoding parameter, Let u be the coefficient of the hash mapping. Let u be the hash feature parameter. This is the mapping alignment coefficient. For alignment adjustment parameters, t is the total number of point cloud features, and v is the total number of encoding dimensions.

[0012] Furthermore, the collaborative interaction model expression of the product 3D scene collaborative review platform is as follows:

[0013] in, The efficiency coefficient for collaborative interaction. Let x be the number of connections to the x-th node. For the communication parameters of the x-th node, Let x be the data transmission volume of the x-th node. For the coordination and synchronization coefficient, For the synchronization parameters of the x-th node, This is the link adjustment coefficient. Let y be the bandwidth parameter of the y-th link. Let y be the transmission efficiency of the y-th link. This is the delay correction factor. Let w be the delay parameter of the y-th link, w be the total number of nodes, and z be the total number of links.

[0014] Furthermore, the comprehensive evaluation model expression for the conformity analysis of the three-dimensional design model is as follows:

[0015] in, This is the composite value for the compliance analysis. Weighting coefficients for each dimension. For semantic-geometric co-verification coefficients, To adapt to spatial rules learning, The mapping value is encoded as the point cloud feature. The efficiency coefficient for collaborative interaction. To standardize the matching parameters, Let the weight of the o-th evaluation factor be... Let q be the parameter of the o-th evaluation factor, and q be the total number of evaluation factors.

[0016] Furthermore, step S3 includes the following sub-steps: S31, based on the geometric topology data and semantic description information of the 3D design model, extract the calibration rule features in the design specification constraints, establish the initial association mapping between the rule features and the model data, and complete the data import and format adaptation through the spatial rule learning adaptation model input interface; S32, the imported design specification constraints are analyzed in layers, dividing them into basic rule layer, related rule layer and priority rule layer, with each layer of rules corresponding to different conformity verification dimensions of the 3D design model; S33. Based on the hierarchical analysis results, a rule adaptation weight matrix is ​​constructed. The matrix elements are determined by the rule importance ranking and the correlation degree of model features, so as to achieve accurate matching between rules and model features. S34. The constructed rule adaptation weight matrix is ​​input into the space rule learning adaptation model for iterative optimization. The matrix parameters are adjusted through multiple rounds of rule adaptation simulation to improve the accuracy and relevance of rule adaptation.

[0017] Furthermore, step S4 includes the following sub-steps: S41: Collect point cloud data of the 3D design model, filter the effective feature point set, remove redundant and abnormal points, and complete the filtering and classification of the feature point set through the preprocessing module of the CAD point cloud feature encoding mapping model. S42 expands the dimension of the classified feature point set, extracts the geometric coordinates, normal vector direction and neighborhood association features of each feature point, and constructs a multi-dimensional feature vector to provide data support for encoding mapping. S43, based on the constructed multi-dimensional feature vector, uses an encoding mapping algorithm to perform feature transformation, converts spatial feature information into encoded values ​​in a high-dimensional vector space, and establishes a one-to-one correspondence between feature vectors and encoded values; S44 outputs the encoding result through the output module of the CAD point cloud feature encoding mapping model, and preliminarily associates it with the design specification constraints of the 3D design model to form an initial matching dataset of encoding mapping and specification constraints.

[0018] Furthermore, step S5 includes the following sub-steps: S51 calls the product's 3D semantic geometry collaborative verification algorithm, loads the rule adaptation weight matrix output by S3 and the feature vector mapping relationship generated by S4, and establishes a correlation channel for multi-dimensional verification data. S52, according to the priority of the design specification constraints, the geometric topological compliance, semantic description consistency and rule adaptation completeness of the three-dimensional design model are checked in turn, with each check dimension corresponding to independent check logic and judgment criteria. S53, cross-validate the results of each verification dimension, integrate multi-dimensional verification data through the algorithm's collaborative verification module, eliminate the bias of single-dimensional verification, and form preliminary compliance analysis results; S54. The preliminary compliance analysis results are fed back to the spatial rule learning and adaptation model for secondary optimization. The rule adaptation weight matrix parameters are adjusted according to the feedback results to complete the final compliance verification of the 3D design model.

[0019] This system, based on a large-scale language model, is a 3D design model compliance analysis system. It applies the aforementioned 3D design model compliance analysis method based on a large-scale language model and includes: a 3D design model multi-source data acquisition and preprocessing unit, which establishes a bidirectional data transmission channel with the product 3D scene collaborative review platform to acquire geometric topology data, semantic description information, and design specification constraints of the 3D design model, completing data format conversion and association mapping; a product 3D semantic geometry collaborative verification unit, which connects to the 3D design model multi-source data acquisition and preprocessing unit and the spatial rule learning and adaptation unit, respectively, and calls the product 3D semantic geometry collaborative verification algorithm to perform collaborative association processing on the multi-source data, extracting geometric feature point sets, semantic association factors, and specification matching dimension parameters; and a spatial rule learning and adaptation unit, which communicates with the product 3D semantic geometry collaborative verification unit and the CAD point cloud feature encoding mapping unit, receives the collaboratively verified dataset, and performs design specification adaptation through spatial rule learning. The system performs feature adaptation and rule mapping under constraints to generate a rule adaptation weight matrix. A CAD point cloud feature encoding and mapping unit, connected to the spatial rule learning and adaptation unit and the 3D design model compliance comprehensive verification unit, encodes and transforms the geometric feature point set of the 3D design model, constructing a mapping relationship between the feature vector space and design specifications. A 3D design model compliance comprehensive verification unit, connected to the spatial rule learning and adaptation unit, the CAD point cloud feature encoding and mapping unit, and the product 3D scene collaborative review unit, completes multi-dimensional compliance verification of the 3D design model based on the rule adaptation weight matrix and feature vector mapping relationship through a product 3D semantic geometry collaborative verification algorithm. A product 3D scene collaborative review unit establishes a closed-loop data interaction with the 3D design model compliance comprehensive verification unit, receives the comprehensive verification results, integrates and outputs the data, forming a comprehensive analysis dataset including feature matching degree, rule adaptation degree, and semantic consistency, enabling collaborative interaction and result display of multi-source data.

[0020] Beneficial effects This invention proposes a method and system for conformity analysis of 3D design models based on large-scale language models. Utilizing product 3D semantic-geometric collaborative verification technology, it establishes a deep correlation processing mechanism between geometric topology data and semantic description information, breaking the limitations of traditional standalone verification. This achieves collaborative correlation and multi-dimensional matching between the two, comprehensively covering the verification dimensions of geometric constraints and semantic requirements in design specifications, eliminating the bias of single-dimensional analysis. Through the dynamic iteration capability of spatial rule learning adaptation technology, it achieves precise adaptation and adaptive adjustment of design specification constraints. Combined with CAD point cloud feature encoding mapping technology, it efficiently converts and accurately maps 3D design model features, solving the problem of insufficient targeting of traditional fixed rule bases and simple encoding methods. Simultaneously, by leveraging a product 3D scene collaborative review platform to build a multi-source data interaction and result integration channel, coupled with a comprehensive 3D design model conformity evaluation mechanism, it forms a complete technical chain from data acquisition, collaborative processing, rule adaptation, feature encoding to multi-dimensional verification. This significantly improves the comprehensiveness, accuracy, and reliability of conformity analysis, meeting the high-precision conformity verification requirements of complex industrial product 3D design models. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method steps of the present invention; Figure 2 This is a diagram showing the system unit composition of the present invention. Detailed Implementation

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] like Figure 1 As shown, the conformity analysis method for 3D design models based on large-scale language models includes the following steps: S1: Obtain the geometric topology data, semantic description information, and design specification constraints of the 3D design model; establish a multi-source data interaction channel through the product 3D scene collaborative review platform; and establish the association mapping relationship between data transmission and storage. Specifically, step S1, as the initial data preparation stage of the entire analysis method, completes the acquisition of multi-source data, the establishment of interactive channels, and the creation of correlation mappings. In practice, firstly, the geometric topology data of the 3D design model is acquired through a data acquisition interface, including core geometric information such as the model's vertex coordinate set, edge-face connection relationships, and surface curvature distribution. Simultaneously, semantic description information corresponding to the model is acquired, including textual semantic data such as design function specifications, component association annotations, and material property descriptions, as well as design specification constraints, including quantitative constraint parameters such as industry standard requirements, structural strength thresholds, and assembly gap ranges. The data acquisition accuracy is controlled within a feature point coordinate error range of no more than 0.5%. Subsequently, the communication module of the product 3D scene collaborative review platform is invoked to establish a multi-source data interaction channel. A dedicated data transmission link under the TCP / IP protocol is configured, and the data transmission rate is set to no less than 100Mbps to ensure the synchronous transmission of geometric data, semantic information, and constraints. Meanwhile, a data storage association mapping relationship is established in the platform's database module. Geometric topology data is bound to keywords of semantic description information according to the hierarchical structure of vertices, edges, and faces. Design specification constraints are indexed according to the constraint type and the corresponding dimension of geometric data, enabling fast retrieval and association of the three types of data, and providing structured and associative basic data support for subsequent collaborative processing.

[0024] S2, invoke the product's 3D semantic geometry collaborative verification algorithm to perform collaborative association processing on the acquired geometric topology data and semantic description information, and extract the geometric feature point set, semantic association factor and standard matching dimension parameter; Specifically, step S2 performs collaborative association processing and feature parameter extraction of multi-source data, achieving the technical objective through the execution of the product's 3D semantic geometry collaborative verification algorithm. During implementation, the geometric topology data and semantic description information collected and stored in step S1 are first imported into the algorithm's processing engine. The algorithm first performs feature recognition on the geometric topology data, traversing the model vertex set using a neighborhood search algorithm to identify key geometric feature point sets, including extreme points, inflection points, and surface transition points. The number of feature points selected is set according to the model complexity, with the size of a single model feature point set controlled between 5000 and 20000. Simultaneously, the semantic description information undergoes word segmentation and keyword extraction, identifying core semantic elements such as functional related words, component name words, and attribute description words. A semantic association factor is established based on a semantic similarity calculation method to quantify the association strength between geometric features and semantic elements. The association factor value is set between 0 and 1, with associations having values ​​higher than 0.7 included in the effective association set. Furthermore, the algorithm, based on the classification of design specification constraints, extracts specification matching dimension parameters, including geometric dimension matching, structural form matching, and functional attribute matching. Each matching dimension corresponds to 3-8 specific parameter indicators. For example, the geometric dimension matching includes parameters such as length, angle, and curvature, while the structural form matching dimension includes parameters such as topological structure type and component connection method. Through the algorithm's parallel computing module, the extraction and quantification of geometric feature point sets, semantic association factors, and specification matching dimension parameters are completed simultaneously, forming a structured intermediate processing dataset. This provides targeted data input for subsequent rule adaptation and encoding mapping.

[0025] S3, input the processed data into the spatial rule learning and adaptation model, perform feature adaptation and rule mapping of design specification constraints, and generate a rule adaptation weight matrix; Specifically, step S3 uses a spatial rule learning adaptation model to achieve feature adaptation and rule mapping between design specification constraints and intermediate data, generating a rule adaptation weight matrix. In practice, the geometric feature point set, semantic association factors, and specification matching dimension parameters extracted in step S2 are first converted into model input data according to the model's preset format. The input dimension is set to a 128-dimensional vector to ensure that the data matches the dimension of the model's input layer. After the model starts, the feature adaptation process is executed first. Features are extracted from the input data through the convolutional layers of a deep learning network. The convolutional kernel size is set to 3×3, and the stride is 1. After three layers of convolution, the extracted deep features are compared with the feature vectors of the design specification constraints using a cosine similarity algorithm to obtain the degree of adaptation between each constraint and the data features. Subsequently, the model enters the rule mapping stage, constructing an initial rule mapping relationship based on the adaptation degree value. Each constraint clause in the design specification is bound to the corresponding dimension of the data features. During the binding process, an adaptation threshold of 0.6 is set; constraint clauses below this threshold are not included in the initial mapping relationship. Based on this, a rule adaptation weight matrix is ​​generated. The number of rows in the matrix corresponds to the total number of design specification constraints, and the number of columns corresponds to the number of dimensions of data features. The matrix element values ​​are calculated by multiplying the adaptation degree value by the priority weight of the constraint clause. The priority of the constraint clause is divided into 5 levels according to its importance in affecting product performance, with weight values ​​of 0.2, 0.4, 0.6, 0.8, and 1.0, respectively. Finally, a rule adaptation weight matrix with a dimension of M×N (M is the number of constraints, and N is the dimension of data features) is formed. The matrix element values ​​are controlled between 0 and 1, realizing the accurate weighted mapping of design specification constraints to data features.

[0026] S4 uses a CAD point cloud feature encoding mapping model to encode and convert the geometric feature point set of the 3D design model, and constructs a mapping relationship between the feature vector space and the design specifications. Specifically, step S4 utilizes the CAD point cloud feature encoding mapping model to complete the encoding transformation of the geometric feature point set and construct the mapping relationship between the feature vector space and the design specifications. During implementation, the geometric feature point set extracted in step S2 is first imported into the model's preprocessing module. Coordinate standardization is performed on the feature point set, transforming the coordinates of all feature points to a unified coordinate system, normalizing the coordinate range to the [-1,1] interval, and eliminating the influence of differences in different model coordinate systems. Subsequently, the model initiates the encoding transformation process, employing a deep neural network structure to encode the geometric feature point set. The network includes four fully connected layers and two skip connection layers. The number of neurons in the fully connected layers is set to 1024, 512, 256, and 128, respectively, and the ReLU function is used as the activation function. Through forward propagation calculation, the three-dimensional coordinates and neighborhood features of each geometric feature point are converted into a 128-dimensional high-dimensional feature vector. Simultaneously, the model incorporates the dimensional division of design specification constraints to construct a mapping relationship between the feature vector space and the design specifications. The encoded high-dimensional feature vectors are grouped according to the matching dimensions of the design specifications, with each matching dimension corresponding to a subset of feature vectors. For example, the geometric dimension matching dimension corresponds to the dimension-related components in the feature vectors, and the structural morphology matching dimension corresponds to the topology-related components. The mapping relationship is established by calculating the correlation coefficient between the feature vector components and the design specification parameters. Components with an absolute correlation coefficient higher than 0.5 are included in the mapping set of their corresponding matching dimensions. This ultimately forms an encoded dataset that includes the feature vector encoding results and multi-dimensional mapping relationships. This dataset retains the core information of the geometric features while achieving a precise association with the design specifications, providing encoding-level data support for subsequent multi-dimensional verification.

[0027] S5, based on the mapping relationship between the rule adaptation weight matrix output by the spatial rule learning adaptation model and the feature vector generated by the CAD point cloud feature encoding mapping model, performs multi-dimensional verification of the conformity of the 3D design model through the product 3D semantic geometry collaborative verification algorithm; Specifically, step S5 is the core step in the multi-dimensional verification of the 3D design model's conformity. It integrates the outputs of previous steps through a secondary call to the product's 3D semantic-geometric collaborative verification algorithm to achieve comprehensive verification. In practice, the algorithm's association module is first invoked to load the rule-adaptation weight matrix generated in step S3 and the feature vector mapping relationship constructed in step S4. This establishes an association channel for the multi-dimensional verification data, binding each element in the weight matrix to its corresponding component in the feature vector mapping relationship, forming verification data association pairs. The number of association pairs is consistent with the number of elements in the rule-adaptation weight matrix. Subsequently, the algorithm prioritizes the design specification constraints and initiates a multi-dimensional verification process. The first dimension is geometric topology compliance verification. Based on the geometric related weights in the rule adaptation weight matrix, the deviation values ​​between the geometric components in the feature vector and the geometric constraint parameters of the design specification are calculated. The deviation values ​​are calculated using the Euclidean distance formula, and the geometric topology compliance score is obtained by weighted summation. The second dimension is semantic consistency verification. Combining the semantic association factor and the semantic related weights in the rule adaptation weight matrix, the algorithm verifies the degree of fit between the semantic association components in the feature vector mapping relationship and the semantic requirements of the design specification. The semantic consistency score is obtained by weighted calculation using semantic similarity. The third dimension is rule adaptation completeness verification. The algorithm counts the number of design specification constraint clauses covered in the feature vector mapping relationship and calculates the rule adaptation completeness score by combining the proportion of non-zero elements in the rule adaptation weight matrix. The score for each dimension ranges from 0 to 100. The algorithm calculates the comprehensive verification score by weighting the scores of the three dimensions in a 4:3:3 weight ratio using a weighted summation module. At the same time, it identifies the deviation items in each dimension, records the deviation location, deviation value and corresponding design specification clauses, and forms a preliminary verification result including the comprehensive score and deviation details, providing comprehensive verification data for subsequent result output.

[0028] S6 outputs the conformity analysis results of the 3D design model through the product 3D scene collaborative review platform, forming a comprehensive analysis dataset including feature matching degree, rule adaptation degree and semantic consistency.

[0029] Specifically, step S6 integrates and structures the verification results through the product 3D scene collaborative review platform. In practice, it first receives the preliminary verification results generated in step S5, including the comprehensive verification score, scores for each dimension, and detailed deviation data. The platform's data analysis module further integrates and processes this data, calculating three core indicators: feature matching degree, rule adaptation degree, and semantic consistency. Feature matching degree is calculated based on the geometric topology compliance score and feature vector mapping coverage; rule adaptation degree is calculated based on the rule adaptation completeness score and the proportion of effective elements in the weight matrix; and semantic consistency is directly expressed using the semantic consistency verification score. All three indicators are converted to percentages and displayed to two decimal places. Subsequently, the platform's result generation module constructs a comprehensive analysis dataset according to a preset format. The dataset includes four parts: a data header, a core indicator area, a deviation detail area, and a related data area. The data header records basic information such as analysis time, model name, and specification version. The core indicator area displays the specific values ​​of the three core indicators. The deviation detail area displays the deviation location, deviation value, corresponding specification clause, and rectification suggestion direction, sorted by deviation severity. The related data area provides an index linking the deviation location with geometric feature point sets and semantic description information. Finally, the platform outputs the comprehensive analysis dataset in the form of charts and graphs through the visualization module, generating visualization results such as 3D model deviation annotation charts, core indicator radar charts, and deviation trend line charts. It also provides dataset export function, supporting the export of files in multiple formats such as Excel, PDF, and JSON, to meet the application needs of results in different scenarios and realize closed-loop management of 3D design model conformity analysis.

[0030] Preferably, the expression for the product 3D semantic geometry collaborative verification algorithm is:

[0031] in, For three-dimensional semantic-geometric co-verification coefficients, Let be the weight factor for the i-th geometric feature. Let i be the i-th geometric topological feature parameter. For the i-th semantic description associated parameter, To standardize the constraint adjustment coefficient, To design specification constraint parameters, The topological correlation matrix, For the j-th normalized dimension weight, For the j-th data normalization parameter, For collaborative verification correction coefficients, Here are the specific verification parameters, where n is the total number of geometric features and m is the total number of normalized dimensions.

[0032] Specifically, the product 3D semantic geometry co-verification algorithm is used to achieve deep co-verification of geometric topology data and semantic description information of the 3D design model. Its implementation requires combining multi-dimensional parameters and weight adjustment mechanisms to complete accurate calculations. In practice, firstly, the total number of geometric features and the total number of normalized dimensions involved in the calculation are determined. The total number of geometric features is set to 50-200 based on the complexity of the 3D design model, and the total number of normalized dimensions is fixed at 10-15 to ensure coverage of core geometric and semantic related dimensions. Then, a weight factor is assigned to each geometric feature. The weight factor is determined based on the feature's influence on design compliance, ranging from 0.1 to 0.9. For key geometric features such as connection points of core components and key transition areas of curved surfaces, the weight factor should not be less than 0.7. Simultaneously, a specification constraint adjustment coefficient and a co-verification correction coefficient are set. The former is set to 0.3-0.8 to balance the influence of design specification constraints, and the latter is set to 0.2-0.5 to correct system deviations during the co-verification process. In the calculation process, the correlation strength between geometric topological feature parameters and semantic description correlation parameters is first quantified through collaborative operation. Then, the guidance role of the specifications on collaborative verification is strengthened by combining design specification constraint parameters and topological structure correlation matrix. Finally, the difference in data magnitude is eliminated by the sum of squares of normalized dimension parameters. The final collaborative verification coefficient ranges from 0 to 1. The closer the coefficient is to 1, the higher the degree of collaborative matching between geometry and semantics. Through the dynamic adjustment and comprehensive operation of multiple parameters, this algorithm achieves accurate collaborative verification of geometric features and semantic information, providing core data support for subsequent compliance analysis.

[0033] The expression for the spatial rule learning and adaptation model is as follows:

[0034] in, To improve the learning adaptability of spatial planning, For the k-th rule weight parameter, For the k-th learning iteration parameter, For the k-th rule mapping coefficient, For the q-th specification, adapt the weights. For the q-th design specification parameter, To adapt to the correction factor, For adaptive adjustment parameters, p is the total number of rules, and r is the total number of norm parameters.

[0035] Specifically, the spatial rule learning and adaptation model achieves precise adaptation between design specification constraints and 3D design model data through dynamic learning and rule mapping. Its implementation relies on multi-round iterative learning and multi-parameter collaborative calculation. During implementation, the total number of rules and specification parameters are first defined. The total number of rules is set to 30-80 according to the complexity of the design specification system. The total number of specification parameters corresponds to the quantitative indicators of each rule, and each rule includes 2-5 specification parameters to ensure comprehensive rule coverage. Rule weight parameters and rule mapping coefficients are assigned to each rule. The rule weight parameters are divided into five levels according to the importance of the rule, with values ​​of 0.2, 0.4, 0.6, 0.8, and 1.0 respectively. The weight parameter for core safety rules is set to 1.0, for general performance rules to 0.6-0.8, and for minor appearance rules to 0.2-0.4. The rule mapping coefficient is set to 0.1-0.7 to adjust the mapping strength between the rule and the model data. Simultaneously, learning iteration parameters are set, with the number of iterations controlled between 50 and 200 rounds. After each iteration, the parameter values ​​are dynamically adjusted based on the adaptation results. The adaptation correction coefficient is fixed at 0.3-0.6 to optimize mapping bias. During calculation, the learning adaptation effect of each rule is first quantified through the collaborative operation of rule weight parameters, learning iteration parameters, and rule mapping coefficients. Then, the influence of all specification parameters is integrated through the summation operation of specification adaptation weights and design specification parameters. Combined with adaptive adjustment parameters, the overall adaptation strength is balanced. The final spatial rule learning adaptation degree ranges from 0 to 1. A higher adaptation degree indicates a better fit between the model data and the design rules. This model achieves adaptive adaptation of design specifications through dynamic learning and precise calculation of multiple rules and parameters.

[0036] Preferably, the expression for the CAD point cloud feature encoding mapping model is:

[0037] in, The mapping value is encoded as the point cloud feature. Let be the weight of the s-th point cloud feature. Let s be the feature parameters of the s-th point cloud. Let be the eigenvalues ​​of the s-th vector. For the s-th eigenvector parameter, For the s-th encoding transformation matrix, For the u-th encoding weight, For the u-th encoding parameter, Let u be the coefficient of the hash mapping. Let u be the hash feature parameter. This is the mapping alignment coefficient. For alignment adjustment parameters, t is the total number of point cloud features, and v is the total number of encoding dimensions.

[0038] Specifically, the CAD point cloud feature encoding mapping model converts the point cloud features of a 3D design model into high-dimensional encoded values ​​and establishes a precise mapping with design specifications. The implementation process requires setting parameters for multiple stages of point cloud feature processing and encoding conversion. In practice, the total number of point cloud features and the total number of encoding dimensions are first determined. The total number of point cloud features is set to 5000-20000 based on the model's accuracy requirements, and the total number of encoding dimensions is fixed at 128 dimensions to ensure that the encoded feature vectors completely retain the core information of the point cloud. Point cloud feature weights and vector feature coefficients are assigned to each point cloud feature. The point cloud feature weights are set to 0.01-0.1 based on the feature's importance, with key feature points such as vertices and inflection points having weights set to 0.08-0.1, and ordinary feature points having weights set to 0.01-0.05. The vector feature coefficients are set to 0.2-0.8 to enhance the discriminative power of the feature vectors. The encoding transformation matrix is ​​constructed with a 3×3 dimension, and the matrix elements are random numbers between -1.0 and 1.0. The optimal matrix parameters are determined through multiple rounds of training and optimization. The encoding weights and hash mapping coefficients are set to 0.3-0.7 and 0.1-0.4, respectively, to adjust the influence intensity of each dimension during the encoding process. The mapping alignment coefficient is set to 0.2-0.5 to correct the alignment deviation after encoding mapping. In the calculation process, the initial encoding of point cloud features is completed first through the operation of point cloud feature parameters, vector feature parameters, and the encoding transformation matrix. Then, the encoding effect is optimized through the collaborative operation of encoding parameters and hash feature parameters. Finally, the mapping alignment coefficient is combined to achieve accurate alignment between the encoded value and the design specification. The obtained point cloud feature encoding mapping value ranges from 0 to 2. The closer the mapping value is to 1, the higher the matching degree between the encoding and the specification, providing accurate encoded data for subsequent verification.

[0039] The collaborative interaction model expression for the product 3D scene collaborative review platform is as follows:

[0040] in, The efficiency coefficient for collaborative interaction. Let x be the number of connections to the x-th node. For the communication parameters of the x-th node, Let x be the data transmission volume of the x-th node. For the coordination and synchronization coefficient, For the synchronization parameters of the x-th node, This is the link adjustment coefficient. Let y be the bandwidth parameter of the y-th link. Let y be the transmission efficiency of the y-th link. This is the delay correction factor. Let w be the delay parameter of the y-th link, w be the total number of nodes, and z be the total number of links.

[0041] Specifically, the collaborative interaction model of the product 3D scene collaborative review platform is used to quantify the efficiency of multi-source data collaborative interaction on the platform. The implementation process requires the configuration and calculation of multi-dimensional parameters for nodes and links. During implementation, the total number of nodes and links is first determined. The total number of nodes is set to 5-20 based on the number of collaborative participants, with each node corresponding to a data processing or interaction unit. The total number of links is set to 10-40 based on node connection relationships to ensure smooth data transmission channels between nodes. Each node is then assigned a number of node connections, node communication parameters, and node data transmission volume. The number of node connections is set to 2-8 based on node function, with 6-8 connections for core data processing nodes and 2-4 connections for ordinary interaction nodes. Node communication parameters are set to 0.5-1.0, representing the node's communication capability, with core nodes set to 0.9-1.0. Node data transmission volume is set to 100-1000MB based on data type, 800-1000MB for geometric data, and 100-300MB for semantic data. The collaborative synchronization coefficient ranges from 0.4 to 0.8, used to adjust the synchronization efficiency between nodes; the link adjustment coefficient is set to 0.3 to 0.6, the link transmission efficiency ranges from 0.6 to 1.0, the delay correction coefficient ranges from 0.2 to 0.5, and the link delay parameter is set to 0.1 to 1.0 according to the actual transmission situation. During calculation, the interaction capability of a single node is first quantified through collaborative calculations of node-related parameters, and then the link transmission performance is evaluated through calculations of link-related parameters. The final collaborative interaction efficiency coefficient ranges from 0 to 1; a higher coefficient indicates a better collaborative interaction effect of the platform. This model provides a quantitative basis for the collaborative optimization of the platform through comprehensive calculations of node and link parameters.

[0042] The comprehensive evaluation model expression for the conformity analysis of the 3D design model is as follows:

[0043] in, This is the composite value for the compliance analysis. Weighting coefficients for each dimension. For semantic-geometric co-verification coefficients, To adapt to spatial rules learning, The mapping value is encoded as the point cloud feature. The efficiency coefficient for collaborative interaction. To standardize the matching parameters, Let the weight of the o-th evaluation factor be... Let q be the parameter of the o-th evaluation factor, and q be the total number of evaluation factors.

[0044] Specifically, the comprehensive evaluation model for 3D design model compliance analysis integrates the output results of previous core algorithms and platforms to achieve a comprehensive quantitative evaluation of compliance. The implementation process requires combining multi-dimensional weight allocation and evaluation factor settings. In practice, firstly, evaluation weight coefficients are set for each dimension, including five dimensions: semantic-geometric collaborative verification, spatial rule learning and adaptation, point cloud feature encoding mapping, collaborative interaction efficiency, and specification matching. The weight coefficients are all between 0.1 and 0.4, with a total of 1.0. Specifically, the weight coefficients for semantic-geometric collaborative verification and spatial rule learning and adaptation are each set to 0.3, the weight coefficients for point cloud feature encoding mapping and specification matching are each set to 0.2, and the weight coefficient for collaborative interaction efficiency is set to 0.1, highlighting the impact of the core verification process. The total number of evaluation factors is set to 8-15, based on the subdivision indicators of the evaluation dimensions. Each evaluation factor corresponds to a specific evaluation item, such as geometric dimension matching degree and semantic description consistency. The parameter value of each evaluation factor ranges from 0 to 100 points. The weight of the evaluation factors is set from 0.05 to 0.2 according to the importance of the evaluation item. The weight of core evaluation items, such as safety specification matching degree, is set from 0.15 to 0.2, and the weight of secondary evaluation items, such as appearance specification matching degree, is set from 0.05 to 0.1. In the calculation process, the differences are first amplified by squaring the output results of each core algorithm and the platform. Then, a weighted sum is performed according to the weight coefficients. The influence of each dimension is balanced by the cube root operation. Finally, the evaluation accuracy is optimized by combining the weighted sum of the evaluation factors. The final comprehensive value of the conformity analysis ranges from 0 to 100 points. The higher the score, the better the conformity of the 3D design model. This model achieves a comprehensive and accurate evaluation of conformity through multi-dimensional and multi-factor comprehensive calculation.

[0045] Step S3 includes the following sub-steps: S31, based on the geometric topology data and semantic description information of the 3D design model, extract the calibration rule features in the design specification constraints, establish the initial association mapping between the rule features and the model data, and complete the data import and format adaptation through the spatial rule learning adaptation model input interface; S32, the imported design specification constraints are analyzed in layers, dividing them into basic rule layer, related rule layer and priority rule layer, with each layer of rules corresponding to different conformity verification dimensions of the 3D design model; S33. Based on the hierarchical analysis results, a rule adaptation weight matrix is ​​constructed. The matrix elements are determined by the rule importance ranking and the correlation degree of model features, so as to achieve accurate matching between rules and model features. S34. The constructed rule adaptation weight matrix is ​​input into the space rule learning adaptation model for iterative optimization. The matrix parameters are adjusted through multiple rounds of rule adaptation simulation to improve the accuracy and relevance of rule adaptation.

[0046] Specifically, step S3 achieves accurate rule adaptation and weight matrix construction for the spatial rule learning adaptation model through four sub-steps. In S31, key rule features are first extracted from the geometric topology data and semantic description information of the 3D design model. These features include core content such as quantitative indicators and correlation requirements in the design specification constraints. Then, an initial correlation mapping is established between the rule features and the model data. The mapping relationship is preliminarily defined according to the correspondence between data types and rule types. The correlated dataset is then imported through the input interface of the spatial rule learning adaptation model. The interface data transmission rate is controlled at 80-120Mbps, and automatic data format adaptation is completed to ensure that the input data is consistent with the structural requirements of the model input layer. In S32, the imported design specification constraints are analyzed hierarchically, divided into a basic rule layer, a correlation rule layer, and a priority rule layer according to the scope and importance of the constraints. The basic rule layer includes core clauses such as general geometric dimension constraints and basic assembly requirements. The correlation rule layer includes content such as inter-component coordination constraints and functional correlation requirements. The priority rule layer focuses on key clauses such as safety performance constraints and core function guarantees. Each rule layer corresponds to 3-5 compliance verification dimensions of the 3D design model, ensuring comprehensive verification coverage. S33 constructs a rule-adaptation weight matrix based on the hierarchical analysis results. The number of rows in the matrix is ​​equal to the total number of constraints in the design specifications, and the number of columns corresponds to the number of feature dimensions in the model data. Matrix elements are determined by multiplying the rule importance ranking result with the calculated model feature correlation value. Rule importance ranking is quantified and scored from 1 to 10, and model feature correlation is calculated using a cosine similarity algorithm, with a value range between 0 and 1. The final matrix element values ​​are controlled between 0 and 1 to achieve accurate matching between rules and model features. S34 inputs the constructed rule-adaptation weight matrix into the spatial rule learning adaptation model for iterative optimization. The number of iterations is set to 80-150 rounds. After each iteration, the matrix parameters are adjusted based on the adaptation error value, which is controlled between 0.01 and 0.05. Through multiple rounds of rule adaptation simulation, the distribution of matrix elements is continuously optimized to improve the accuracy and relevance of rule adaptation, providing reliable weight support for subsequent compliance verification.

[0047] Step S4 includes the following sub-steps: S41: Collect point cloud data of the 3D design model, filter the effective feature point set, remove redundant and abnormal points, and complete the filtering and classification of the feature point set through the preprocessing module of the CAD point cloud feature encoding mapping model. S42 expands the dimension of the classified feature point set, extracts the geometric coordinates, normal vector direction and neighborhood association features of each feature point, and constructs a multi-dimensional feature vector to provide data support for encoding mapping. S43, based on the constructed multi-dimensional feature vector, uses an encoding mapping algorithm to perform feature transformation, converts spatial feature information into encoded values ​​in a high-dimensional vector space, and establishes a one-to-one correspondence between feature vectors and encoded values; S44 outputs the encoding result through the output module of the CAD point cloud feature encoding mapping model, and preliminarily associates it with the design specification constraints of the 3D design model to form an initial matching dataset of encoding mapping and specification constraints.

[0048] Specifically, step S4 completes the feature processing and encoding mapping of the CAD point cloud feature encoding mapping model through four sub-steps. S41 first collects point cloud data from the 3D design model, with the collection density set to 3-8 points per cubic millimeter according to the model's accuracy requirements. Then, it filters the effective feature point set, using a neighborhood density filtering algorithm to remove redundant and outlier points. The filtering threshold is set to 0.02-0.05 to ensure that the retained feature point set accurately reflects the model's geometric features. Next, the preprocessing module of the CAD point cloud feature encoding mapping model classifies the filtered feature point set according to geometric feature type, dividing it into vertex sets, edge feature sets, and face feature sets, etc. The classification accuracy is controlled above 98%, laying the foundation for subsequent processing. S42 expands the dimensions of the classified feature point set, extracting multi-dimensional information such as the three-dimensional geometric coordinates, normal vector direction, curvature value, and neighborhood correlation features of each feature point. The expanded dimension of each feature point is set to 12-18 dimensions, with geometric coordinates occupying 3 dimensions, normal vector direction occupying 3 dimensions, curvature value occupying 1 dimension, and neighborhood correlation features occupying 5-11 dimensions. By integrating multi-dimensional information, a comprehensive feature vector is constructed, providing rich data support for encoding mapping. S43, based on the constructed multi-dimensional feature vector, uses a deep neural network encoding mapping algorithm for feature transformation. The network includes 3 convolutional layers and 2 fully connected layers. The convolutional kernel size is set to 3×3, and the number of neurons in the fully connected layers is 512 and 256, respectively. The activation function is the ReLU function. Through forward propagation of the network, the spatial feature information is converted into encoded values ​​in a high-dimensional vector space. The dimension of the encoded values ​​is fixed at 128 dimensions. At the same time, a one-to-one correspondence between feature vectors and encoded values ​​is established to ensure the uniqueness and reversibility of the encoding transformation. S44 outputs the encoding results through the output module of the CAD point cloud feature encoding mapping model. The output data format adopts a standardized vector format. Then, the encoding results are initially associated with the design specification constraints of the 3D design model. The correlation relationship is established by calculating the correlation coefficient between the encoding value and the specification parameters. The correlation coefficient threshold is set to 0.6-0.8. The encoding values ​​that meet the threshold requirements are bound to the specification clauses to form an initial matching dataset of encoding mapping and specification constraints, providing encoding-level correlation data for subsequent multi-dimensional verification.

[0049] Step S5 includes the following sub-steps: S51 calls the product's 3D semantic geometry collaborative verification algorithm, loads the rule adaptation weight matrix output by S3 and the feature vector mapping relationship generated by S4, and establishes a correlation channel for multi-dimensional verification data. S52, according to the priority of the design specification constraints, the geometric topological compliance, semantic description consistency and rule adaptation completeness of the three-dimensional design model are checked in turn, with each check dimension corresponding to independent check logic and judgment criteria. S53, cross-validate the results of each verification dimension, integrate multi-dimensional verification data through the algorithm's collaborative verification module, eliminate the bias of single-dimensional verification, and form preliminary compliance analysis results; S54. The preliminary compliance analysis results are fed back to the spatial rule learning and adaptation model for secondary optimization. The rule adaptation weight matrix parameters are adjusted according to the feedback results to complete the final compliance verification of the 3D design model.

[0050] Specifically, step S5 implements multi-dimensional compliance verification of the product's 3D semantic geometry collaborative verification algorithm through four sub-steps. S51 first calls the product's 3D semantic geometry collaborative verification algorithm, loading the rule adaptation weight matrix output in step S3 and the feature vector mapping relationship generated in step S4, establishing a connection channel for multi-dimensional verification data. The channel data transmission latency is controlled within 10-30 milliseconds. Simultaneously, a one-to-one correspondence binding relationship is established between weight matrix elements and feature vector components, ensuring 100% binding accuracy and achieving efficient association and retrieval of verification data. S52 sorts the design specification constraints according to their priority, with priorities divided into five levels based on their impact on product performance, corresponding to weight coefficients of 1.0, 0.8, 0.6, 0.4, and 0.2 from high to low. The geometric topology compliance, semantic description consistency, and rule adaptation completeness of the 3D design model are verified sequentially. Geometric topology compliance verification focuses on indicators such as dimensional accuracy and structural form; semantic description consistency verification focuses on functional descriptions and component associations; and rule adaptation completeness verification covers the matching of all specification clauses. Each verification dimension has independent verification logic and judgment criteria to ensure the targeted nature of the verification. S53 performs cross-validation on the results of each verification dimension using a weighted summation cross-validation algorithm. The verification weights for each dimension are allocated as follows: 40% for geometric topological compliance, 30% for semantic description consistency, and 30% for rule adaptation completeness. The algorithm's collaborative verification module integrates multi-dimensional verification data and calculates the deviation value of each dimension's verification results. The deviation value is controlled between 0.03 and 0.08 to eliminate biases from single-dimensional verification, forming preliminary compliance analysis results. The results include scores for each dimension and detailed deviations. S54 feeds the preliminary compliance analysis results back to the spatial rule learning adaptation model for secondary optimization. The feedback data transmission rate is set to 50-100 Mbps. Based on the feedback deviation details, the rule adaptation weight matrix parameters are adjusted, with the adjustment range set at 0.05-0.2 proportional to the deviation value. This parameter adjustment further optimizes the rule adaptation effect. The multi-dimensional verification process is then restarted to complete the final 3D design model compliance verification, ensuring the accuracy and reliability of the verification results.

[0051] The 3D semantic-geometric collaborative verification algorithm of this invention is an algorithm for multi-dimensional collaborative verification of 3D design models. It is a technical solution that integrates geometric topological data and semantic description information, and completes compliance verification through multi-parameter weighted calculation and collaborative operation. The implementation of this algorithm requires a multi-stage process: First, it receives the geometric topological feature parameters (such as vertex coordinates, edge-face connection relationships, etc.) and semantic description association parameters (such as functional descriptions, component annotations, etc.) of the 3D design model. Based on the model complexity, it sets the total number of geometric features to 50-200 and the total number of normalized dimensions to 10-15. Each feature is assigned a weight factor of 0.1-0.9, and a specification constraint adjustment coefficient of 0.3-0.8 and a collaborative verification correction coefficient of 0.2-0.5 are configured. Then, the correlation strength is quantified through collaborative operation of geometric and semantic parameters. The specification guidance is strengthened by combining design specification constraint parameters and the topological structure association matrix. After normalization processing to eliminate data magnitude differences, a collaborative verification coefficient with a value range of 0-1 is finally output. This algorithm breaks the fragmented processing model of geometric and semantic data, achieving deep correlation and synchronous verification between the two. It covers multi-dimensional matching requirements such as geometric dimensions, structural morphology, and functional attributes, solving the matching deviation problem caused by traditional separate verification. Through dynamic adjustment of multiple parameters and parallel computing, it improves the accuracy of verification, provides core data support for the conformity analysis of 3D design models, and ensures the comprehensiveness and reliability of subsequent comprehensive evaluation.

[0052] The spatial rule learning and adaptation model is a dynamic learning model that achieves accurate adaptation between design specification constraints and 3D design model data. It establishes a weighted mapping relationship between rules and data features through multiple rounds of iterative learning. The model implementation process is as follows: First, define the total number of rules (30-80) and the 2-5 specification parameters corresponding to each rule. Assign each rule a weight parameter of 0.2-1.0 and a mapping coefficient of 0.1-0.7, and set 50-200 iterations and an adaptation correction coefficient of 0.3-0.6. Then, convert the geometric feature point set, semantic association factors, and other data into a 128-dimensional input vector. Extract deep features through convolutional layers and calculate the cosine similarity with the specification feature vector. Construct an initial rule mapping relationship based on an adaptation threshold of 0.6. Generate an M×N dimension weight matrix based on rule importance ranking and feature association. Then, through multiple rounds of iterative optimization, control the adaptation error between 0.01 and 0.05. This model achieves adaptive adaptation to design specifications, dynamically adjusts rule weights to match model data features, and generates an accurate rule adaptation weight matrix. It breaks through the rigidity of traditional fixed rule bases and can flexibly adjust the adaptation strategy according to the dynamic changes in design specifications and the differences in model features, improving the pertinence and timeliness of rule matching. It provides a scientific weight basis for multi-dimensional compliance verification and ensures that the verification results are highly consistent with the design specifications.

[0053] The CAD point cloud feature encoding and mapping model is a technical model that converts the point cloud features of a 3D design model into high-dimensional codes and establishes a standardized mapping. Its core function is to achieve accurate encoding of geometric features and their association with standards. The specific implementation steps are as follows: Point cloud data is collected at a density of 3-8 points per cubic millimeter; effective feature point sets are filtered and classified using a neighborhood density filtering algorithm (threshold 0.02-0.05); each feature point is expanded into a 12-18 dimensional vector (including coordinates, normal vectors, curvature, etc.); a deep neural network with 3 convolutional layers and 2 fully connected layers is constructed, using a 3×3 convolutional kernel and the ReLU activation function to convert spatial features into 128-dimensional high-dimensional encoded values, establishing a one-to-one correspondence between feature vectors and encoded values; by calculating the correlation coefficient between encoded values ​​and standard parameters (threshold 0.6-0.8), encoded values ​​that meet the requirements are bound to standard clauses, forming an initial matching dataset. This model transforms three-dimensional spatial features into quantifiable and associative coded data, building a mapping bridge between geometric features and design specifications. It solves the problem of insufficient accuracy in traditional coding methods. Through multi-dimensional feature expansion and deep neural network coding, it fully preserves the core information of geometric features, achieves accurate association with design specifications, and provides standardized and highly recognizable coded data for subsequent collaborative verification, thereby improving the efficiency and accuracy of the verification process.

[0054] The product 3D scene collaborative review platform is a core support platform that integrates multi-source data interaction, processing result integration, and visualization output. Essentially, it provides a technical carrier for end-to-end data management and interaction for 3D design model conformity analysis. The platform's implementation relies on multi-module collaboration: establishing dedicated data transmission links under the TCP / IP protocol, setting a transmission rate of no less than 100Mbps and a hardware configuration of 5-20 nodes and 10-40 links; establishing association indexes and storage mappings for geometric, semantic, and specification data; configuring data analysis, result generation, and visualization modules; receiving multi-dimensional verification results; calculating core indicators such as feature matching degree and rule adaptation degree; constructing a comprehensive dataset including basic information, indicator data, and deviation details according to a preset format; providing export functions in multiple formats such as Excel and PDF; and generating visualization results such as 3D model deviation annotation charts and radar charts. This platform streamlines the entire process of data collection, processing, verification, and output, enabling efficient interaction of multi-source data, systematic integration of results, and intuitive display. It solves the problems of poor data transmission and scattered results from multiple sources, improves data processing efficiency through a unified interaction and management platform, provides users with comprehensive and intuitive analysis results, achieves closed-loop management of compliance analysis, and meets the collaborative review needs in complex industrial scenarios.

[0055] like Figure 2As shown, a 3D design model compliance analysis system based on a large language model is implemented. This system, applied to a 3D design model compliance analysis method based on a large language model, includes: a 3D design model multi-source data acquisition and preprocessing unit, which establishes a bidirectional data transmission channel with a product 3D scene collaborative review platform to acquire geometric topological data, semantic description information, and design specification constraints of the 3D design model, completing data format conversion and association mapping; a product 3D semantic-geometric collaborative verification unit, which connects to the 3D design model multi-source data acquisition and preprocessing unit and the spatial rule learning and adaptation unit, respectively, and calls the product 3D semantic-geometric collaborative verification algorithm to perform collaborative association processing on the multi-source data, extracting geometric feature point sets, semantic association factors, and specification matching dimension parameters; and a spatial rule learning and adaptation unit, which communicates with the product 3D semantic-geometric collaborative verification unit and the CAD point cloud feature encoding mapping unit, receiving the collaboratively verified dataset and performing design through spatial rule learning and model adaptation. The system includes: a feature adaptation and rule mapping unit for standard constraints, generating a rule adaptation weight matrix; a CAD point cloud feature encoding and mapping unit, connected to the spatial rule learning and adaptation unit and the 3D design model compliance comprehensive verification unit, which encodes and transforms the geometric feature point set of the 3D design model to construct a mapping relationship between the feature vector space and the design specifications; a 3D design model compliance comprehensive verification unit, connected to the spatial rule learning and adaptation unit, the CAD point cloud feature encoding and mapping unit, and the product 3D scene collaborative review unit, which performs multi-dimensional verification of the 3D design model compliance based on the rule adaptation weight matrix and feature vector mapping relationship, using a product 3D semantic geometry collaborative verification algorithm; and a product 3D scene collaborative review unit, which establishes a closed-loop data interaction with the 3D design model compliance comprehensive verification unit, receiving the comprehensive verification results and integrating and outputting the data to form a comprehensive analysis dataset including feature matching degree, rule adaptation degree, and semantic consistency, enabling collaborative interaction and result display of multi-source data.

[0056] This paper presents a method and system for conformity analysis of 3D design models based on large-scale language models. It breaks down data barriers using product 3D semantic-geometric collaborative verification technology, enabling synchronous processing and deep association of geometric topology data and semantic description information of 3D design models. Simultaneously, it leverages a product 3D scene collaborative review platform to establish an efficient data interaction channel, ensuring smooth data transmission and integration from multiple sources. This solves the problems of data fragmentation and low processing efficiency in traditional analysis. Spatial rule learning and adaptation technology possesses dynamic iterative optimization capabilities, flexibly adjusting adaptation strategies according to design specification characteristics. Combined with CAD point cloud feature encoding mapping technology for accurate conversion and mapping of geometric features, it significantly improves the targeting of feature extraction and rule matching. Furthermore, through a multi-dimensional conformity verification and comprehensive evaluation mechanism, it constructs a complete technical chain from data acquisition to result output, comprehensively covering all constraints and requirements of design specifications.

[0057] This invention establishes a correlation processing mechanism between geometric features and semantic information through product 3D semantic geometry collaborative verification technology, achieving synchronous verification of geometric features and semantic information, eliminating the bias of single-dimensional analysis, and comprehensively covering the matching scenarios of geometric constraints and semantic requirements in design specifications. Addressing the shortcomings of traditional model rule adaptation rigidity and insufficient feature mapping accuracy, it achieves adaptive adaptation to design specifications through the dynamic learning capability of spatial rule learning adaptation technology. Combined with CAD point cloud feature encoding mapping technology for refined transformation of 3D design model features, it improves the accuracy of rule adaptation and feature mapping. Furthermore, through multi-dimensional cross-validation and result feedback optimization mechanisms, it further corrects analytical biases, significantly improving the comprehensiveness and reliability of 3D design model conformity analysis, and meeting the high-precision verification requirements of complex industrial scenarios.

[0058] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for three-dimensional design model conformance analysis based on large language models, characterized in that, The method comprises the following steps: S1, obtaining geometric topology data, semantic description information and design specification constraint conditions of a three-dimensional design model, establishing a multi-source data interaction channel through a product three-dimensional scene collaborative review platform, and establishing an associated mapping relationship of data transmission and storage; S2, calling a product three-dimensional semantic geometry collaborative verification algorithm to perform collaborative association processing on the obtained geometric topology data and semantic description information, extracting geometric feature point sets, semantic association factors and specification matching dimension parameters; S3, inputting the processed data into a spatial rule learning adaptation model to perform feature adaptation and rule mapping of the design specification constraint conditions, and generating a rule adaptation weight matrix; S4, performing coding conversion on the geometric feature point sets of the three-dimensional design model through a CAD point cloud feature coding mapping model, and constructing a mapping relationship between the feature vector space and the design specification; S5, based on the rule adaptation weight matrix output by the spatial rule learning adaptation model and the feature vector mapping relationship generated by the CAD point cloud feature coding mapping model, performing three-dimensional design model compliance multidimensional verification through a product three-dimensional semantic geometry collaborative verification algorithm; S6, outputting three-dimensional design model compliance analysis results through a product three-dimensional scene collaborative review platform, and forming a comprehensive analysis data set including feature matching degree, rule adaptation degree and semantic consistency.

2. The large language model based three-dimensional design model compliance analysis method according to claim 1, wherein, The expression of the product three-dimensional semantic geometry collaborative verification algorithm is: wherein, is a three-dimensional semantic geometry collaborative verification coefficient, is a weight factor of the i-th geometric feature, is the i-th geometric topological feature parameter, is the i-th semantic description correlation parameter, is a specification constraint adjustment coefficient, is a design specification constraint parameter, is a topological structure correlation matrix, is the j-th normalized dimension weight, is the j-th data normalization parameter, is a collaborative verification correction coefficient, is a special verification parameter, n is the total number of geometric features, and m is the total number of normalized dimensions.

3. The large language model based three-dimensional design model compliance analysis method of claim 1, wherein, The expression of the spatial rule learning adaptation model is: wherein, is a spatial planning learning fitness, is a kth rule weight parameter, is a kth learning iteration parameter, is a kth rule mapping coefficient, is a qth norm fitness weight, is a qth design norm parameter, is a fitness correction coefficient, is an adaptive adjustment parameter, p is the total number of rules, and r is the total number of norm parameters.

4. The large language model based three-dimensional design model compliance analysis method of claim 1, wherein, An expression of the CAD point cloud feature encoding mapping model is: wherein, is a point cloud feature encoding mapping value, is a s-th point cloud feature weight, is a s-th point cloud feature parameter, is a s-th vector feature coefficient, is a s-th feature vector parameter, is a s-th encoding conversion matrix, is a u-th encoding weight, is a u-th encoding parameter, is a u-th hash mapping coefficient, is a u-th hash feature parameter, is a mapping alignment coefficient, is an alignment adjustment parameter, t is the total number of point cloud features, and v is the total number of encoding dimensions.

5. The large language model based three-dimensional design model compliance analysis method according to claim 1, wherein, The collaborative interaction model expression of the product three-dimensional scene collaborative review platform is: wherein, is a cooperative interaction efficiency coefficient, is the xth node connection number, is the xth node communication parameter, is the xth node data transmission amount, is a cooperative synchronization coefficient, is the xth node synchronization parameter, is a link adjustment coefficient, is the yth link bandwidth parameter, is the yth link transmission efficiency, is a delay correction coefficient, is the yth link delay parameter, w is the total number of nodes, and z is the total number of links.

6. The large language model based three-dimensional design model compliance analysis method according to claim 1, wherein, The comprehensive evaluation model expression of the three-dimensional design model conformity analysis is: wherein, is a conformity analysis integrated value, is a dimension evaluation weight coefficient, is a semantic geometric collaborative verification coefficient, is a spatial rule learning adaptation degree, is a point cloud feature encoding mapping value, is a collaborative interaction efficiency coefficient, is a specification matching parameter, is an oth evaluation factor weight, is an oth evaluation factor parameter, q is the total number of evaluation factors.

7. The large language model based three-dimensional design model compliance analysis method according to claim 1, wherein, The S3 comprises the following steps: S31, based on the geometric topology data and semantic description information of the three-dimensional design model, extracting the calibration rule features in the design specification constraint conditions, establishing an initial associated mapping of the rule features and the model data, and completing data import and format adaptation through an input interface of the spatial rule learning adaptation model; S32, performing hierarchical analysis on the imported design specification constraint conditions, dividing basic rule layer, associated rule layer and priority rule layer, and each layer of rules corresponding to different compliance verification dimensions of the three-dimensional design model; S33, according to the hierarchical analysis result, constructing a rule adaptation weight matrix, and determining the matrix elements through rule importance sorting and model feature correlation degree calculation to perform accurate matching of rules and model features; S34, inputting the constructed rule adaptation weight matrix into the spatial rule learning adaptation model for iterative optimization, adjusting the matrix parameters through multiple rule adaptation simulation, and improving the accuracy and pertinence of rule adaptation.

8. The large language model based three-dimensional design model compliance analysis method according to claim 1, wherein, The S4 comprises the following steps: S41, collecting point cloud data of the three-dimensional design model, screening effective feature point sets, eliminating redundant points and abnormal points, and completing feature point set screening and classification through a preprocessing module of the CAD point cloud feature coding mapping model; S42, performing dimension expansion on the classified feature point sets, extracting geometric coordinates, normal vector directions and neighborhood associated features of each feature point, and constructing multi-dimensional feature vectors to provide data support for coding mapping; S43, based on the constructed multi-dimensional feature vectors, using a coding mapping algorithm to perform feature conversion, converting spatial feature information into coding values in a high-dimensional vector space, and establishing a one-to-one correspondence between the feature vectors and the coding values; S44, the output module of the CAD point cloud feature coding mapping model outputs the coding result, preliminarily associates with the design specification constraint condition of the three-dimensional design model, and forms an initial matching data set of coding mapping and specification constraint.

9. The large language model based three-dimensional design model compliance analysis method according to claim 1, wherein, The S5 includes the following steps: S51, a product three-dimensional semantic geometry collaborative verification algorithm is called, the rule adaptation weight matrix output in S3 and the feature vector mapping relationship generated in S4 are loaded, and an associated channel of multi-dimensional verification data is established; S52, according to the priority order of the design specification constraint condition, the geometric topology compliance, semantic description consistency and rule adaptation integrity of the three-dimensional design model are verified in turn, and each verification dimension corresponds to independent verification logic and judgment standard; S53, the results of each verification dimension are cross-verified, multi-dimensional verification data are integrated through the collaborative verification module of the algorithm, the deviation of single-dimensional verification is eliminated, and a preliminary compliance analysis result is formed; S54, the preliminary compliance analysis result is fed back to the spatial rule learning adaptation model for secondary optimization, the rule adaptation weight matrix parameters are adjusted according to the feedback result, and the final three-dimensional design model compliance verification is completed.

10. A three-dimensional design model conformance analysis system based on a large language model, characterized by, The system is applied to the three-dimensional design model compliance analysis method based on the large language model in claim 1, comprising: A three-dimensional design model multi-source data acquisition and preprocessing unit, which establishes a bidirectional data transmission channel with a product three-dimensional scene collaborative review platform, is used for acquiring geometric topology data, semantic description information and design specification constraint conditions of the three-dimensional design model, and completing data format conversion and associated mapping; A product three-dimensional semantic geometry collaborative verification unit, which is connected with the three-dimensional design model multi-source data acquisition and preprocessing unit and the spatial rule learning adaptation unit respectively, calls a product three-dimensional semantic geometry collaborative verification algorithm to cooperatively associate and process multi-source data, extracts geometric feature point sets, semantic association factors and specification matching dimension parameters; A spatial rule learning adaptation unit, which is communicatively connected with the product three-dimensional semantic geometry collaborative verification unit and the CAD point cloud feature coding mapping unit, receives the data set after collaborative verification, performs feature adaptation and rule mapping of the design specification constraint condition through a spatial rule learning adaptation model, and generates a rule adaptation weight matrix; A CAD point cloud feature coding mapping unit, which is connected with the spatial rule learning adaptation unit and the three-dimensional design model compliance comprehensive verification unit, performs coding conversion on the geometric feature point set of the three-dimensional design model, and constructs a mapping relationship between the feature vector space and the design specification; A three-dimensional design model compliance comprehensive verification unit, which is connected with the spatial rule learning adaptation unit, the CAD point cloud feature coding mapping unit and the product three-dimensional scene collaborative review unit respectively, completes three-dimensional design model compliance multi-dimensional verification through a product three-dimensional semantic geometry collaborative verification algorithm based on the rule adaptation weight matrix and the feature vector mapping relationship; The product three-dimensional scene collaborative review unit is closed-loop data interaction with the three-dimensional design model conformity comprehensive checking unit, receives the comprehensive checking result and performs data integration and output, forms a comprehensive analysis data set including feature matching degree, rule adaptation degree and semantic consistency, and performs collaborative interaction and result display of multi-source data.

Images