Intelligent retrieval method and system for cad design features based on knowledge graph

Abstract

The application provides a CAD design feature intelligent retrieval method and system based on a knowledge graph, and relates to the technical field of data processing, wherein the method comprises the following steps: generating a first retrieval intention factor, a second retrieval intention factor and a third retrieval intention factor by analyzing retrieval request information input by a user into a CAD design space; performing multi-modal analysis graph structure mapping based on the CAD design space; performing retrieval fitness detection on the CAD design analysis graph space according to the first, second and third retrieval intention factors to obtain a first, second and third fitness optimization retrieval domain; and performing multi-domain retrieval collaborative analysis on the first, second and third fitness optimization retrieval domains according to a retrieval attention correlation mechanism to obtain a fourth fitness optimization retrieval domain. The technical problems of low CAD design feature retrieval efficiency and insufficient matching accuracy in the prior art are solved. The technical effects of improving design feature retrieval efficiency and accuracy are achieved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for intelligent retrieval of CAD design features based on knowledge graphs. Background Technology

[0002] With the rapid development of modern manufacturing and industrial design, Computer-Aided Design (CAD) has become an important tool for product development, encompassing multiple stages such as part modeling, assembly design, functional analysis, and process planning. In actual design processes, designers often need to quickly retrieve specific features or similar solutions from large amounts of historical design data to improve design efficiency and ensure design consistency. However, traditional CAD retrieval methods based on keywords or simple attribute matching suffer from low retrieval efficiency, insufficient matching accuracy, and difficulty in integrating multi-dimensional features.

[0003] Existing technologies suffer from low efficiency and insufficient matching accuracy in CAD design feature retrieval. Summary of the Invention

[0004] The purpose of this application is to provide a knowledge graph-based intelligent retrieval method and system for CAD design features, in order to solve the technical problems of low efficiency and insufficient matching accuracy in existing CAD design feature retrieval technologies.

[0005] In view of the above problems, this application provides a method and system for intelligent retrieval of CAD design features based on knowledge graphs.

[0006] The first aspect of this application provides an intelligent retrieval method for CAD design features based on a knowledge graph. The method includes: parsing a user-inputted retrieval request to a CAD design space to generate a first retrieval intent factor, a second retrieval intent factor, and a third retrieval intent factor; performing multimodal parsing graph structure mapping based on the CAD design space to establish a CAD design parsing graph space; performing retrieval fitness detection on the CAD design parsing graph space according to the first retrieval intent factor to obtain a first fitness-optimized retrieval domain; performing retrieval fitness detection on the CAD design parsing graph space according to the second retrieval intent factor to obtain a second fitness-optimized retrieval domain; performing retrieval fitness detection on the CAD design parsing graph space according to the third retrieval intent factor to obtain a third fitness-optimized retrieval domain; and performing multi-domain retrieval collaborative analysis on the first fitness-optimized retrieval domain, the second fitness-optimized retrieval domain, and the third fitness-optimized retrieval domain according to a retrieval attention association mechanism to obtain a fourth fitness-optimized retrieval domain.

[0007] Optionally, the request request information is categorized based on the historical event set to obtain a request geometric feature parsing event set, a request functional feature parsing event set, and a request process feature parsing event set; the retrieval request information is then parsed against distillation intent based on the request geometric feature parsing event set to obtain a first retrieval intent factor; the retrieval request information is then parsed against distillation intent based on the request functional feature parsing event set to obtain a second retrieval intent factor; and the retrieval request information is then parsed against distillation intent based on the request process feature parsing event set to obtain a third retrieval intent factor.

[0008] Optionally, the request geometric feature parsing event set is subjected to distribution equalization perturbation by an adversarial example generator to obtain an optimized request geometric feature parsing set; a CNN network is trained under supervision based on the request geometric feature parsing event set to construct a first geometric intent parsing network; an RNN network is trained under supervision based on the optimized request geometric feature parsing set to construct a second geometric intent parsing network; the first and second geometric intent parsing networks are fused through distillation learning to obtain a third geometric intent parsing network; the retrieval request information is input into the third geometric intent parsing network, and the first retrieval intent factor is output.

[0009] Optionally, based on the CAD design space, the m-th design drawing is read, where m is a positive integer, 1≤m≤M; geometric feature analysis is performed on the m-th design drawing to obtain multiple design geometric feature vectors; functional feature analysis is performed on the m-th design drawing to obtain multiple design functional feature vectors; process feature analysis is performed on the m-th design drawing to obtain multiple design process feature vectors; graph structure mapping is performed on the multiple design geometric feature vectors, the multiple design functional feature vectors, and the multiple design process feature vectors according to the knowledge graph to obtain the m-th design analysis graph, and the m-th design analysis graph is added to the CAD design analysis graph space.

[0010] Optionally, the matching degree of each design geometric feature vector in the m-th design analysis map is evaluated according to the first search intent factor to obtain multiple geometric matching evaluation coefficients; the stability of the multiple geometric matching evaluation coefficients is evaluated to obtain multiple geometric matching stability coefficients; the fitness of the multiple geometric matching evaluation coefficients and the multiple geometric matching stability coefficients is analyzed according to the geometric search fitness analysis conditions to obtain multiple geometric search fitness; the multiple geometric search fitnesss are optimized and identified according to the predetermined geometric search fitness to obtain the m-th optimal geometric fitness distribution; the geometric feature optimization is performed on the m-th design analysis map according to the m-th optimal geometric fitness distribution to obtain the m-th optimal geometric feature set; the m-th optimal geometric fitness distribution and the m-th optimal geometric feature set are added to the first fitness optimization retrieval domain, and the retrieval fitness detection of the CAD design analysis map space is continued based on the first search intent factor to update the first fitness optimization retrieval domain.

[0011] Optionally, the first search intent factor is perturbed to obtain multiple perturbed geometric intent factors; the matching degree of each design geometric feature vector in the m-th design analytical map is evaluated according to the multiple perturbed geometric intent factors to obtain multiple geometric matching evaluation groups; curve fitting is performed according to the multiple geometric matching evaluation groups to obtain each geometric matching evaluation curve corresponding to each design geometric feature vector; central tendency analysis is performed according to the multiple geometric matching evaluation curves to determine multiple perturbed matching evaluation set values; the stability of the multiple geometric matching evaluation coefficients is evaluated according to the multiple perturbed matching evaluation set values ​​to generate the multiple geometric matching stability coefficients.

[0012] Optionally, the retrieval attention association mechanism is activated, which includes geometric retrieval fitness weight, functional retrieval fitness weight, and process retrieval fitness weight; multi-domain retrieval is performed based on the first fitness optimization retrieval domain, the second fitness optimization retrieval domain, and the third fitness optimization retrieval domain to obtain multiple retrieval matching results, each corresponding to a retrieval fitness evaluation matrix; retrieval collaborative fitness is analyzed on the multiple retrieval matching results based on the retrieval attention association mechanism to obtain multiple retrieval collaborative fitness; and optimization identification is performed on the multiple retrieval matching results based on the multiple retrieval collaborative fitness to generate the fourth fitness optimization retrieval domain.

[0013] Optionally, the CAD design space includes M CAD design drawings, where M is a positive integer greater than 1.

[0014] Optionally, the geometric retrieval fitness parsing conditions include geometric matching evaluation weights and geometric matching stability weights.

[0015] A second aspect of this application provides a knowledge graph-based intelligent retrieval system for CAD design features. The system includes: a request parsing module, used to parse retrieval request information input by a user into a CAD design space, generating a first retrieval intent factor, a second retrieval intent factor, and a third retrieval intent factor; a space establishment module, used to perform multimodal parsing graph structure mapping based on the CAD design space, establishing a CAD design parsing graph space; a first fitness detection module, used to perform retrieval fitness detection on the CAD design parsing graph space according to the first retrieval intent factor, obtaining a first fitness optimization retrieval domain; a second fitness detection module, used to perform retrieval fitness detection on the CAD design parsing graph space according to the second retrieval intent factor, obtaining a second fitness optimization retrieval domain; a third fitness detection module, used to perform retrieval fitness detection on the CAD design parsing graph space according to the third retrieval intent factor, obtaining a third fitness optimization retrieval domain; and a multi-domain retrieval collaborative analysis module, used to perform multi-domain retrieval collaborative analysis on the first fitness optimization retrieval domain, the second fitness optimization retrieval domain, and the third fitness optimization retrieval domain according to a retrieval attention association mechanism, obtaining a fourth fitness optimization retrieval domain.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0017] The method provided in this application generates a first search intent factor, a second search intent factor, and a third search intent factor by parsing the search request information input by the user into the CAD design space; it then performs multimodal analytical graph structure mapping based on the CAD design space to establish a CAD design analytical graph space; it performs search fitness detection on the CAD design analytical graph space according to the first search intent factor to obtain a first fitness-optimized search domain; it performs search fitness detection on the CAD design analytical graph space according to the second search intent factor to obtain a second fitness-optimized search domain; it performs search fitness detection on the CAD design analytical graph space according to the third search intent factor to obtain a third fitness-optimized search domain; and it performs multi-domain search collaborative analysis on the first fitness-optimized search domain, the second fitness-optimized search domain, and the third fitness-optimized search domain according to a search attention association mechanism to obtain a fourth fitness-optimized search domain. This achieves the technical effect of improving the efficiency and accuracy of design feature retrieval.

[0018] The above description is merely an overview of the technical solution of this application. To enable a clearer understanding of the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 A flowchart illustrating the knowledge graph-based intelligent retrieval method for CAD design features provided in this application.

[0021] Figure 2 A schematic diagram of the structure of the CAD design feature intelligent retrieval system based on knowledge graph provided in this application.

[0022] Explanation of reference numerals in the attached diagram: Request parsing module 11, space establishment module 12, first fitness detection module 13, second fitness detection module 14, third fitness detection module 15, multi-domain retrieval collaborative analysis module. Detailed Implementation

[0023] This application provides a knowledge graph-based intelligent CAD design feature retrieval method and system to address the technical problems of low efficiency and insufficient matching accuracy in existing CAD design feature retrieval technologies. It achieves the technical effect of improving the efficiency and accuracy of design feature retrieval.

[0024] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0025] Example 1, as Figure 1As shown, this application provides a knowledge graph-based intelligent retrieval method for CAD design features, which includes:

[0026] The system parses the search request information input by the user into the CAD design space and generates a first search intent factor, a second search intent factor, and a third search intent factor.

[0027] Furthermore, the CAD design space includes M CAD design drawings, where M is a positive integer greater than 1.

[0028] Furthermore, parsing the user's search request information input into the CAD design space to generate a first search intent factor, a second search intent factor, and a third search intent factor includes: classifying the historical event set of request parsing to obtain a request geometric feature parsing event set, a request functional feature parsing event set, and a request process feature parsing event set; performing adversarial distillation intent parsing on the search request information based on the request geometric feature parsing event set to obtain the first search intent factor; performing adversarial distillation intent parsing on the search request information based on the request functional feature parsing event set to obtain the second search intent factor; and performing adversarial distillation intent parsing on the search request information based on the request process feature parsing event set to obtain the third search intent factor.

[0029] Specifically, the system receives retrieval request information input by the user into the CAD design space. This retrieval request information refers to the user-inputted requirements, including design geometric requirements, functional requirements, or technological constraints, used to locate the required design features within the CAD design space. The CAD design space is a collection of CAD design data that can be retrieved and analyzed, comprising M CAD design drawings, where M is a positive integer greater than 1. Each CAD design drawing contains three main categories of information: geometric features, functional features, and technological features. Geometric features refer to the spatial attributes of a part or assembly, such as its dimensions, shape, topological relationships, surfaces, holes, and slots. Functional features refer to the part's purpose, assembly relationships, stress characteristics, and functional constraints. Technological features refer to manufacturing process information, such as materials, processing methods, surface treatments, and assembly processes. The CAD design space can be obtained through the CAD system database or drawing parsing tools.

[0030] The received user search request information is parsed. Specifically, historical search records are first collected from the database. Key fields are extracted from each record, including request information, matching feature type, matching result, and user selection behavior. All records are summarized to form a request parsing history event set, which contains various situations of past user search requests. The historical request information in the request parsing history event set is analyzed to obtain request set feature parsing event sets, request function feature parsing event sets, and request process feature parsing event sets. For example, the request information, matching feature type, and matching result in the request parsing history event set are analyzed. Based on the matching results, multiple historical requests are clustered or classified to form three types of event sets. Among them, the request set feature parsing event set contains request records related to geometry, such as size, shape, hole position, surface, and topology. The request function feature parsing event set contains request records related to function, such as part purpose, stress conditions, and functional constraints. The request process feature parsing event set contains request records related to manufacturing process, such as processing method, material, and surface treatment.

[0031] For each type of feature parsing event set in the request geometric feature parsing event set, the request functional feature parsing event set, and the request process feature parsing event set, an adversarial distillation intent parsing method is employed. Adversarial distillation intent parsing is a technique combining adversarial example generation and knowledge distillation: it generates adversarial examples by introducing slight perturbations to maintain accurate feature understanding under different input biases; and it enhances the accuracy and reliability of the analysis by fusing the parsing capabilities of multiple networks through distillation learning. Based on the request geometric feature parsing event set, adversarial distillation intent parsing is performed on the retrieval request information to output a first retrieval intent factor, which accurately reflects the user's retrieval intent regarding geometric features. Similarly, based on the request functional feature parsing event set, adversarial distillation intent parsing is performed on the retrieval request information to obtain a second retrieval intent factor, which accurately reflects the user's retrieval intent regarding functional features. Based on the request process feature parsing event set, adversarial distillation intent parsing is performed on the retrieval request information to obtain a third retrieval intent factor, which accurately reflects the user's retrieval intent regarding process features.

[0032] By accurately analyzing users' search intent from different dimensions and generating corresponding search intent factors, precise input is provided for the fitness retrieval and multi-domain collaboration of CAD design analysis maps, thereby improving the intelligence, accuracy and targeting of the search.

[0033] Further, adversarial distillation of intent parsing is performed on the retrieval request information based on the requested geometric feature parsing event set to obtain the first retrieval intent factor, including: performing distribution equalization perturbation on the requested geometric feature parsing event set using an adversarial example generator to obtain an optimized requested geometric feature parsing set; supervising training of a CNN network based on the requested geometric feature parsing event set to construct a first geometric intent parsing network; supervising training of an RNN network based on the optimized requested geometric feature parsing set to construct a second geometric intent parsing network; performing distillation learning fusion on the first and second geometric intent parsing networks to obtain a third geometric intent parsing network; inputting the retrieval request information into the third geometric intent parsing network, and outputting the first retrieval intent factor.

[0034] Specifically, an adversarial distillation intent parsing technique is employed, where an adversarial sample generator applies a distributional equalization perturbation to the request geometric feature parsing event set. The adversarial sample generator is an algorithm for generating slightly perturbed samples, applying small-amplitude random perturbations to the feature vectors in the original request geometric feature parsing event set. The adversarial sample generator can generate an optimized sample set based on random noise, minute scaling, displacement, or geometric parameter transformations. The distributional equalization perturbation refers to simulating diverse user inputs by adding small random perturbations to the original geometric features. Specifically, the request geometric feature parsing event set is represented as a vector, with each vector containing geometric attribute features such as size, hole location, and topological relationships. The adversarial sample generator applies small random perturbations to these vectors, for example, a ±5% perturbation to the size attribute and a slight offset within ±0.5cm to the hole location coordinates. Through adjustment, the generated perturbed samples are integrated with the original combined feature parsing event set to form an optimized request geometric feature parsing set. This optimized set maintains the effectiveness of the original geometric features while expanding the diversity of the training samples.

[0035] Based on the requested geometric feature parsing optimization set, both CNN and RNN neural networks are trained for feature learning. The CNN network is trained using the requested feature parsing event set, extracting local geometric patterns such as combinations of holes and slots, surface shapes, or local topological structures through convolutional and pooling layers. Supervised training is used during training, with each request feature vector and its corresponding historical matching results as input and output, and cross-entropy or mean squared error loss functions used to optimize network weights. After training, a first geometric intent parsing network is constructed. This first network learns the representations of various geometric features in the requested geometric feature parsing optimization set, enabling it to analyze local geometric feature patterns and interpret the geometric intent in the retrieval request information from a local perspective. Simultaneously, the RNN network is trained under supervised conditions based on the requested geometric feature parsing optimization set. RNN networks excel at handling sequential or topology-related features, such as the arrangement sequence of part geometry on a drawing and topological relationships. During training, gradient descent is used to optimize network parameters, enabling the RNN to learn the relationships and change patterns of geometric features in the global sequence. After training, a second geometric intent parsing network is obtained, which can parse the geometric feature intent in the retrieval request information from a global perspective.

[0036] This paper utilizes a distillation-learning fusion method to integrate two geometric intent parsing networks: the first network and the second network. The first network acts as a teacher network, providing high-confidence predictions of local feature patterns, while the second network acts as another teacher network, providing high-confidence predictions of global sequence patterns. The outputs of both networks are used as guidance to train a student network, integrating the feature understanding capabilities of the two types of networks. By calculating the mean squared error, the difference between the student network output and the teacher network output is minimized to generate a third geometric intent parsing network. The retrieval request information is then input into this third network for analysis, outputting a first retrieval intent factor.

[0037] Similarly, by applying distributional equalization perturbations to the request function feature parsing event set and the request process feature parsing event set respectively using adversarial examples, optimized sets for request function feature parsing and request process feature parsing are obtained. Then, a CNN network is trained under supervision using these optimized sets to construct a first function intent parsing network and a first process intent parsing network. Simultaneously, an RNN network is trained under supervision using these optimized sets to construct a second function intent parsing network and a second process intent parsing network. Using a distillation learning fusion algorithm, the first and second function intent parsing networks are fused to obtain a third function intent parsing network. The first and second process intent parsing networks are then fused to obtain a third process intent parsing network. The retrieval request information is then input into the third function intent parsing network and the third process intent parsing network respectively to obtain a second retrieval intent factor and a third retrieval intent factor.

[0038] By using adversarial distillation intent parsing, the retrieval request information is comprehensively analyzed and processed, improving the accuracy and reliability of the first, second, and third retrieval intent factors, thereby enhancing the accuracy and relevance of the retrieval and enabling intelligent retrieval of CAD design features.

[0039] Based on the CAD design space, a multimodal analytical graph structure mapping is performed to establish a CAD design analytical graph atlas space.

[0040] Furthermore, based on the CAD design space, a multimodal analytical graph structure mapping is performed to establish a CAD design analytical graph space, including: reading the m-th design drawing (where m is a positive integer, 1 ≤ m ≤ M) according to the CAD design space; performing geometric feature analysis on the m-th design drawing to obtain multiple design geometric feature vectors; performing functional feature analysis on the m-th design drawing to obtain multiple design functional feature vectors; performing process feature analysis on the m-th design drawing to obtain multiple design process feature vectors; performing graph structure mapping on the multiple design geometric feature vectors, the multiple design functional feature vectors, and the multiple design process feature vectors according to a knowledge graph to obtain the m-th design analytical graph, and adding the m-th design analytical graph to the CAD design analytical graph space.

[0041] Specifically, based on the CAD design space, the m-th design drawing is read, where m is a positive integer, 1 ≤ m ≤ M, and the m-th design drawing contains all the design information of the part or component. M is the total number of design drawings in the CAD design space. For the m-th design drawing, three types of feature analysis are performed: Geometric feature analysis is performed on the m-th design drawing to extract information such as the shape, size, hole positions, surfaces, and topological relationships of the part or component, and these are vectorized into design geometric feature vectors. These design geometric feature vectors are used to describe the spatial structural features of the part or component. Functional feature analysis is performed on the m-th design drawing to analyze the functional attributes of the part or component within the m-th design drawing, such as the part's purpose, stress conditions, and functional constraints, and to generate corresponding design functional feature vectors. These design functional feature vectors reflect the function undertaken by the part or component in the overall design. Simultaneously, process feature analysis is performed on the m-th design drawing to extract manufacturing process-related information, such as processing methods, material selection, and surface treatment, and convert this information into design process feature vectors for process constraint matching and optimization.

[0042] After obtaining the three types of feature vectors, the Neo4j graph database is used to perform graph structure mapping on multiple design geometric feature vectors, multiple design functional feature vectors, and multiple design process feature vectors. A knowledge graph is a structure that organizes knowledge in the form of a graph, consisting of nodes and edges. Through graph structure mapping, different types of feature vectors are used as nodes, and edges are constructed based on the relationships between different categories of features, such as the association between geometric features and functional features, and the association between functional features and process features, forming the m-th design analytical graph. In the m-th design analytical graph, nodes represent entities, such as geometric features, functional features, and process features, and edges represent the relationships between entities. The m-th design analytical graph is added to the CAD design analytical graph space. By traversing all M design drawings and performing the above processing, a multimodal graph space encompassing the entire CAD design space is established.

[0043] The CAD design analysis diagram space integrates the geometric, functional, and technological features of all design drawings and presents them in a diagrammatic form, providing comprehensive and accurate data support for CAD design feature retrieval and improving the efficiency and accuracy of retrieval.

[0044] Based on the first retrieval intent factor, the CAD design analysis map space is subjected to retrieval fitness detection to obtain the first fitness optimization retrieval domain.

[0045] Further, the retrieval fitness detection of the CAD design analysis map space based on the first retrieval intent factor is performed to obtain a first fitness optimization retrieval domain, including: evaluating the matching degree of each design geometric feature vector in the m-th design analysis map based on the first retrieval intent factor to obtain multiple geometric matching evaluation coefficients; evaluating the stability of the multiple geometric matching evaluation coefficients to obtain multiple geometric matching stability coefficients; performing fitness analysis on the multiple geometric matching evaluation coefficients and the multiple geometric matching stability coefficients based on geometric retrieval fitness analysis conditions to obtain multiple geometric retrieval fitnesss; performing optimization identification on the multiple geometric retrieval fitnesss based on predetermined geometric retrieval fitnesss to obtain the m-th optimal geometric fitness distribution; performing geometric feature optimization identification on the m-th design analysis map based on the m-th optimal geometric fitness distribution to obtain the m-th optimal geometric feature set; adding the m-th optimal geometric fitness distribution and the m-th optimal geometric feature set to the first fitness optimization retrieval domain, and continuing to perform retrieval fitness detection on the CAD design analysis map space based on the first retrieval intent factor to update the first fitness optimization retrieval domain.

[0046] Furthermore, the geometric retrieval fitness parsing conditions include geometric matching evaluation weights and geometric matching stability weights.

[0047] Specifically, based on the first retrieval intent factor, each m-th design analysis map in the CAD design analysis map is traversed. The first retrieval intent factor is then compared with each design geometric feature vector within the m-th design analysis map using a similarity calculation algorithm, such as cosine similarity or Euclidean distance, to calculate the matching degree and obtain multiple geometric matching evaluation coefficients. These geometric matching evaluation coefficients reflect the degree of matching between each geometric feature and the user's intent. The stability of these multiple geometric matching coefficients is evaluated by subjecting the first retrieval intent factor to neighboring semantic perturbations. Without altering the semantic core, the first retrieval intent factor is slightly modified, and the matching degree is recalculated multiple times and analyzed to obtain a geometric matching stability coefficient for each feature. This coefficient measures the stability of the matching results of the geometric feature vector under different input conditions.

[0048] Based on the geometric retrieval fitness analysis conditions, the geometric retrieval fitness of each feature is calculated by combining multiple geometric matching evaluation coefficients and multiple geometric matching stability coefficients. The geometric retrieval fitness analysis conditions include geometric matching evaluation weights and geometric matching stability weights. The formula for calculating geometric retrieval fitness is: F(i) = w1 × S m (i)+w2×S s (i), where F(i) is the retrieval fitness of the i-th geometric feature, S m (i) is the matching evaluation coefficient, S s(i) represents the matching stability coefficient, where w1 and w2 are the geometric matching evaluation weight and geometric matching stability weight, respectively, satisfying w1+w2=1. This can be dynamically set according to actual needs. The geometric retrieval fitness calculation formula achieves accurate analysis of geometric feature retrieval fitness by weightedly fusing the geometric matching evaluation coefficient and the geometric matching stability coefficient.

[0049] By analyzing historical data and using the mean and standard deviation of features, a predetermined geometric retrieval fitness is dynamically set. This predetermined geometric retrieval fitness serves as a threshold condition. Multiple geometric retrieval fitness values ​​are optimized and identified based on this predetermined fitness value. All geometric retrieval fitness values ​​greater than or equal to the predetermined fitness value are selected and integrated to form the m-th optimal geometric fitness distribution. Based on this distribution, the m-th design analytical map undergoes geometric feature optimization and identification: First, the m-th optimal geometric fitness distribution is clustered using K-means or DBSCAN clustering algorithms to group geometrically similar features. This clustering effectively aggregates similar geometric features, such as hole-like, groove-like, and chamfer-like structures. After clustering, a weighted center vector is calculated for each cluster, representing the average geometric feature weighted by the feature fitness value. A higher weight indicates that the geometric feature is closer to the user's search intent. The clusters are sorted according to their average fitness values, with the highest value ranked first. Based on a set output ratio or quantity threshold, several top-ranked clusters are selected to form the m-th optimal geometric feature set, which is the set of geometric candidate features that best matches the user's search intent. The m-th optimal geometric fitness distribution and the m-th optimal geometric feature set are added to the first fitness optimization retrieval domain. Based on the first search intent factor, iterative retrieval and fitness updates are performed on the CAD design analysis map space, gradually optimizing and updating the first fitness optimization retrieval domain. The first fitness optimization retrieval domain refers to the set of high-fitness feature clusters formed based on geometric search intent factors in the CAD design analysis map space. It contains all feature results that best match the user's intent geometrically after fitness detection and optimization screening.

[0050] By performing retrieval fitness detection based on retrieval intent factors, dynamic matching and optimal filtering from semantic intent to geometric features are achieved, enabling rapid and accurate identification of design features that meet user geometric requirements within the CAD design atlas space. Furthermore, by introducing stability and weighted fitness, the reliability and accuracy of retrieval results are effectively improved.

[0051] Furthermore, the stability evaluation of the plurality of geometric matching evaluation coefficients is performed to obtain a plurality of geometric matching stability coefficients, including: perturbing the first retrieval intent factor to obtain a plurality of perturbed geometric intent factors; evaluating the matching degree of each design geometric feature vector in the m-th design analytical map according to the plurality of perturbed geometric intent factors to obtain a plurality of geometric matching evaluation groups; performing curve fitting according to the plurality of geometric matching evaluation groups to obtain each geometric matching evaluation curve corresponding to each design geometric feature vector; performing central tendency analysis according to the plurality of geometric matching evaluation curves to determine a plurality of perturbed matching evaluation set values; and evaluating the stability of the plurality of geometric matching evaluation coefficients according to the plurality of perturbed matching evaluation set values ​​to generate the plurality of geometric matching stability coefficients.

[0052] Specifically, the first retrieval intent factor is subjected to neighboring semantic perturbation. Neighboring semantic perturbation refers to making minor perturbations to it while maintaining the core semantic meaning of the retrieval intent, simulating semantic variations that might occur when a user expresses the same design requirement. For example, for the intent of a circular hole structure, neighboring semantic perturbation generates adjacent semantic representations such as through-hole structure and cylindrical hole structure in the vector space. Multiple perturbed geometric intent factors are generated by applying a word vector replacement mechanism to the first retrieval intent factor. Based on these multiple perturbed geometric intent factors, a similarity calculation algorithm is used to calculate the matching degree of each design geometric feature vector within the m-th design analytical graph, forming multiple geometric matching evaluation groups. Each evaluation group corresponds to the full graph matching result under one perturbation. Curve fitting is performed on the multiple geometric matching evaluation groups, with the number of matches on the horizontal axis and the matching degree on the vertical axis, to generate a geometric matching evaluation curve corresponding to each geometric feature vector. The geometric matching evaluation curve reflects the sensitivity of the design geometric feature to semantic perturbation. If the curve fluctuation amplitude is small, it indicates that the design geometric feature can still maintain a high matching degree under different intent perturbations. After obtaining multiple geometric matching evaluation curves, statistical analysis methods, such as calculating the mean, median, or standard deviation, are used to perform central tendency analysis on each geometric matching evaluation curve, resulting in multiple perturbation matching evaluation set values. These set values ​​reflect the core stable region of the design geometric feature matching degree under multiple perturbations. The stability of multiple geometric matching evaluation coefficients is evaluated based on these set values, i.e., the deviation between the set values ​​and the coefficients is calculated. A small difference and a small fluctuation range indicate high stability of the matching result. Normalized difference is used to characterize the stability of feature matching under perturbation conditions, mapping the deviation value to the 0-1 interval. The closer the value is to 1, the more stable the matching evaluation under perturbation. The stability value corresponding to each geometric feature is used as the geometric matching stability coefficient.

[0053] By introducing stability evaluation indicators, a more stable and reliable basis is provided for CAD feature retrieval, effectively avoiding the deviation of retrieval results caused by unstable coefficients, and improving the accuracy and effectiveness of CAD feature retrieval.

[0054] Based on the second retrieval intent factor, the CAD design analysis map space is subjected to retrieval fitness detection to obtain the second fitness optimization retrieval domain.

[0055] Based on the third retrieval intent factor, the retrieval fitness of the CAD design analysis map space is tested to obtain the third fitness optimization retrieval domain.

[0056] Specifically, the overall logic and implementation of performing retrieval fitness testing on the CAD design analysis map space based on the second retrieval intent factor to obtain the second fitness optimization retrieval domain, and performing retrieval fitness testing on the CAD design analysis map space based on the third retrieval intent factor to obtain the third fitness optimization retrieval domain, are similar to those of performing retrieval fitness testing on the CAD design analysis map space based on the first retrieval intent factor to obtain the first fitness optimization retrieval domain. Specifically, based on the second retrieval intent factor, functional feature matching is performed on each m-th design analysis map in the CAD design analysis map space, calculating multiple functional matching evaluation coefficients. Multiple perturbation functional intent vectors are generated through neighboring perturbations, and stability analysis is performed to obtain multiple functional matching stability coefficients. Based on the functional retrieval fitness analysis conditions, including functional matching evaluation weights and functional matching stability weights, multiple functional matching evaluation coefficients and multiple functional matching stability coefficients are weighted and fused to calculate the functional retrieval fitness of each functional feature. The m-th optimal function fitness distribution is obtained by filtering according to the predetermined functional fitness threshold. Then, the functional feature is identified by clustering and feature geometry to form the m-th optimal function feature set. The m-th optimal function fitness distribution and the m-th optimal function feature set are added to the second fitness retrieval domain. The retrieval fitness of the CAD design analysis map space is further detected according to the second retrieval intent factor, and the second fitness retrieval domain is updated.

[0057] Similarly, for the third search intent factor, in the CAD design analysis map space, for each m-th design analysis map, a search fitness detection and matching at the process feature level is performed to obtain multiple process matching evaluation coefficients and multiple process matching stability coefficients. Based on the process search fitness analysis conditions, including process matching evaluation weights and process matching stability weights, multiple process search fitnesss are obtained. By setting the process search fitness, optimization identification is performed to obtain the m-th optimal process fitness distribution. Based on the m-th optimal process fitness distribution, process feature optimization identification is performed on the m-th design analysis map to obtain the m-th optimal process feature set. The m-th optimal process fitness distribution and the m-th optimal process feature set are added to the third fitness optimization search domain, and search fitness detection continues based on the third search intent factor to update the third fitness optimization search domain. Among them, the second fitness optimization search domain is the feature matching and optimization result at the functional level, and the third fitness optimization search domain is the feature matching and optimization result at the process level.

[0058] By obtaining the second fitness optimization retrieval domain and the third fitness optimization retrieval domain, the accuracy and reliability of CAD design feature retrieval are further improved, as are the intelligence and precision of CAD design feature retrieval.

[0059] Based on the retrieval attention association mechanism, a multi-domain retrieval collaborative analysis is performed on the first fitness optimization retrieval domain, the second fitness optimization retrieval domain, and the third fitness optimization retrieval domain to obtain a fourth fitness optimization retrieval domain.

[0060] Furthermore, based on the retrieval attention association mechanism, multi-domain retrieval collaborative analysis is performed on the first fitness optimization retrieval domain, the second fitness optimization retrieval domain, and the third fitness optimization retrieval domain to obtain a fourth fitness optimization retrieval domain. This includes: activating the retrieval attention association mechanism, which includes geometric retrieval fitness weight, functional retrieval fitness weight, and process retrieval fitness weight; performing multi-domain retrieval sorting based on the first fitness optimization retrieval domain, the second fitness optimization retrieval domain, and the third fitness optimization retrieval domain to obtain multiple retrieval matching results, each retrieval matching result corresponding to a retrieval fitness evaluation matrix; performing retrieval collaborative fitness analysis on the multiple retrieval matching results based on the retrieval attention association mechanism to obtain multiple retrieval collaborative fitness; and performing optimization identification on the multiple retrieval matching results based on the multiple retrieval collaborative fitness to generate the fourth fitness optimization retrieval domain.

[0061] Specifically, a retrieval attention association mechanism is activated. This mechanism includes three core weight parameters: geometric retrieval fitness weight, functional retrieval fitness weight, and process retrieval fitness weight. These weights measure the importance of different feature domains in the overall retrieval and can be dynamically adjusted based on user input intent or historical matching experience, thereby allocating attention to the retrieval process. Multi-domain retrieval is performed on the first, second, and third fitness optimization retrieval domains. The design graph features of each domain are matched and combined to form multiple retrieval matching results. Each result corresponds to a retrieval fitness evaluation matrix, which includes geometric retrieval fitness, functional retrieval fitness, and process retrieval fitness. The retrieval attention association mechanism is used to analyze the collaborative fitness of the multiple matching results. The fitness values ​​from the three fitness optimization retrieval domains are weighted and merged into a single global collaborative fitness value, resulting in multiple collaborative fitness values. The collaborative fitness is calculated as follows: F... c (i)=w g ⋅F g (i)+w f ⋅F f (i)+w p ⋅F p (i), F c (i) represents the retrieval co-fitness of the matching result of the i-th design. The higher the value, the more the design meets the user's needs in the three dimensions of geometry, function, and process. g (i), F f (i), F p (i) represent the fitness of geometric retrieval, functional retrieval, and process retrieval, respectively. g w f w p These are the fitness weights for geometric retrieval, functional retrieval, and process retrieval, respectively. g w f w pThe sum is 1. A search collaboration fitness threshold is set. Multiple search collaboration fitness values ​​are used to optimize and identify the multiple search matching results. All search collaboration fitness values ​​greater than or equal to the threshold are selected as final candidates, generating a fourth fitness optimization search domain. This fourth fitness optimization search domain integrates multi-domain information, considering not only the individual matching effect of each feature domain but also achieving dynamic adjustment between domains through attention weights. It realizes cross-domain collaborative optimization of geometric, functional, and technological features, thereby improving the accuracy and reliability of search results. By constructing the fourth fitness optimization search domain, a final globally optimal design feature set is provided for CAD intelligent retrieval, improving the efficiency, effectiveness, and high accuracy of CAD design feature intelligent retrieval.

[0062] Example 2, based on the same inventive concept as the knowledge graph-based intelligent CAD design feature retrieval method in the preceding examples, such as... Figure 2 As shown, this application provides a knowledge graph-based intelligent CAD design feature retrieval system, wherein the knowledge graph-based intelligent CAD design feature retrieval system includes:

[0063] The request parsing module 11 is used to parse the retrieval request information input by the user into the CAD design space, and generate a first retrieval intent factor, a second retrieval intent factor, and a third retrieval intent factor; the space establishment module 12 is used to perform multimodal analytical graph structure mapping based on the CAD design space to establish a CAD design analytical graph space; the first fitness detection module 13 is used to perform retrieval fitness detection on the CAD design analytical graph space according to the first retrieval intent factor to obtain a first fitness optimization retrieval domain; the second fitness detection module 14 is used to perform retrieval fitness detection on the CAD design analytical graph space according to the second retrieval intent factor to obtain a second fitness optimization retrieval domain; the third fitness detection module 15 is used to perform retrieval fitness detection on the CAD design analytical graph space according to the third retrieval intent factor to obtain a third fitness optimization retrieval domain; the multi-domain retrieval collaborative analysis module 16 is used to perform multi-domain retrieval collaborative analysis on the first fitness optimization retrieval domain, the second fitness optimization retrieval domain, and the third fitness optimization retrieval domain according to the retrieval attention association mechanism to obtain a fourth fitness optimization retrieval domain.

[0064] Furthermore, the request parsing module 11 is also configured to: classify the request parsing historical event set to obtain a request geometric feature parsing event set, a request functional feature parsing event set, and a request process feature parsing event set; perform adversarial distillation intent parsing on the retrieval request information based on the request geometric feature parsing event set to obtain a first retrieval intent factor; perform adversarial distillation intent parsing on the retrieval request information based on the request functional feature parsing event set to obtain a second retrieval intent factor; and perform adversarial distillation intent parsing on the retrieval request information based on the request process feature parsing event set to obtain a third retrieval intent factor.

[0065] Furthermore, the request parsing module 11 is also configured to: perform distribution equalization perturbation on the request geometric feature parsing event set according to the adversarial example generator to obtain a request geometric feature parsing optimization set; perform supervised training on a CNN network according to the request geometric feature parsing event set to construct a first geometric intent parsing network; perform supervised training on an RNN network according to the request geometric feature parsing optimization set to construct a second geometric intent parsing network; perform distillation learning fusion on the first and second geometric intent parsing networks to obtain a third geometric intent parsing network; input the retrieval request information into the third geometric intent parsing network, and output the first retrieval intent factor.

[0066] Furthermore, the space establishment module 12 is also used to: read the m-th design drawing according to the CAD design space, where m is a positive integer, 1≤m≤M; perform geometric feature analysis on the m-th design drawing to obtain multiple design geometric feature vectors; perform functional feature analysis on the m-th design drawing to obtain multiple design functional feature vectors; perform process feature analysis on the m-th design drawing to obtain multiple design process feature vectors; perform graph structure mapping on the multiple design geometric feature vectors, the multiple design functional feature vectors, and the multiple design process feature vectors according to the knowledge graph to obtain the m-th design analysis graph, and add the m-th design analysis graph to the CAD design analysis graph space.

[0067] Furthermore, the first fitness detection module 13 is also used to: evaluate the matching degree of each design geometric feature vector in the m-th design analysis map according to the first retrieval intent factor, and obtain multiple geometric matching evaluation coefficients; evaluate the stability of the multiple geometric matching evaluation coefficients, and obtain multiple geometric matching stability coefficients; perform fitness analysis on the multiple geometric matching evaluation coefficients and the multiple geometric matching stability coefficients according to the geometric retrieval fitness analysis conditions, and obtain multiple geometric retrieval fitness; perform optimization identification on the multiple geometric retrieval fitness according to the predetermined geometric retrieval fitness, and obtain the m-th optimal geometric fitness distribution; perform geometric feature optimization identification on the m-th design analysis map according to the m-th optimal geometric fitness distribution, and obtain the m-th optimal geometric feature set; add the m-th optimal geometric fitness distribution and the m-th optimal geometric feature set to the first fitness optimization retrieval domain, and continue to perform retrieval fitness detection on the CAD design analysis map space based on the first retrieval intent factor, and update the first fitness optimization retrieval domain.

[0068] Furthermore, the first fitness detection module 13 is also configured to: perturb the first retrieval intent factor to obtain multiple perturbation geometric intent factors; evaluate the matching degree of each design geometric feature vector in the m-th design analytical map according to the multiple perturbation geometric intent factors to obtain multiple geometric matching evaluation groups; perform curve fitting according to the multiple geometric matching evaluation groups to obtain each geometric matching evaluation curve corresponding to each design geometric feature vector; perform central tendency analysis according to the multiple geometric matching evaluation curves to determine multiple perturbation matching evaluation set values; and evaluate the stability of the multiple geometric matching evaluation coefficients according to the multiple perturbation matching evaluation set values ​​to generate the multiple geometric matching stability coefficients.

[0069] Furthermore, the multi-domain retrieval collaborative analysis module 16 is also used to: activate the retrieval attention association mechanism, which includes geometric retrieval fitness weight, functional retrieval fitness weight, and process retrieval fitness weight; perform multi-domain retrieval sorting based on the first fitness optimization retrieval domain, the second fitness optimization retrieval domain, and the third fitness optimization retrieval domain to obtain multiple retrieval matching results, each retrieval matching result corresponding to a retrieval fitness evaluation matrix; perform retrieval collaborative fitness analysis on the multiple retrieval matching results based on the retrieval attention association mechanism to obtain multiple retrieval collaborative fitness; and perform optimization identification on the multiple retrieval matching results based on the multiple retrieval collaborative fitness to generate the fourth fitness optimization retrieval domain.

[0070] Furthermore, the request parsing module 11 is also used to: the CAD design space includes M CAD design drawings, where M is a positive integer greater than 1.

[0071] Furthermore, the first fitness detection module 13 is also used to: the geometric retrieval fitness parsing conditions include geometric matching evaluation weights and geometric matching stability weights.

[0072] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The knowledge graph-based intelligent retrieval method and specific examples in the aforementioned Embodiment 1 are also applicable to the knowledge graph-based intelligent retrieval system for CAD design features in this embodiment. Through the foregoing detailed description of the knowledge graph-based intelligent retrieval method for CAD design features, those skilled in the art can clearly understand the knowledge graph-based intelligent retrieval system for CAD design features in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0073] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0074] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A knowledge graph-based intelligent CAD design feature retrieval method, characterized in that, include: Parse the search request information input by the user into the CAD design space to generate the first search intent factor, the second search intent factor, and the third search intent factor; Based on the CAD design space, a multimodal analytical graph structure mapping is performed to establish a CAD design analytical graph atlas space; Based on the first retrieval intent factor, the CAD design analysis map space is subjected to retrieval fitness detection to obtain a first fitness optimization retrieval domain. The first fitness optimization retrieval domain refers to the set of high fitness feature clusters formed based on geometric retrieval intent factors in the CAD design analysis map space, which includes all feature results that best meet the user's intent in the geometric dimension after fitness detection and optimization screening. Based on the second search intent factor, the CAD design analysis map space is subjected to search fitness detection to obtain the second fitness optimization search domain. Based on the third search intent factor, the CAD design analysis map space is searched for fitness to obtain the third fitness optimization search domain. Based on the retrieval attention association mechanism, a multi-domain retrieval collaborative analysis is performed on the first fitness optimization retrieval domain, the second fitness optimization retrieval domain, and the third fitness optimization retrieval domain to obtain a fourth fitness optimization retrieval domain; Based on the first retrieval intent factor, a retrieval fitness test is performed on the CAD design analysis map space to obtain a first fitness-optimized retrieval domain, including: Based on the first retrieval intent factor, the matching degree of each geometric feature vector of the m-th design analytical map is evaluated to obtain multiple geometric matching evaluation coefficients. The stability of the plurality of geometric matching evaluation coefficients is evaluated to obtain a plurality of geometric matching stability coefficients, wherein the geometric matching stability coefficients are used to measure whether the matching results of the geometric feature vectors are stable under different input conditions; Based on the geometric retrieval fitness analysis conditions, fitness analysis is performed on the multiple geometric matching evaluation coefficients and the multiple geometric matching stability coefficients to obtain multiple geometric retrieval fitnesss; Based on the predetermined geometric retrieval fitness, the multiple geometric retrieval fitness values ​​are optimized and identified to obtain the distribution of the m-th optimized geometric fitness value; Based on the m-th optimization geometric fitness distribution, the m-th design analytical map is subjected to geometric feature optimization identification to obtain the m-th optimization geometric feature set; The m-th optimal geometric fitness distribution and the m-th optimal geometric feature set are added to the first fitness optimization retrieval domain, and the retrieval fitness detection is continued on the CAD design analysis map space based on the first retrieval intent factor to update the first fitness optimization retrieval domain.

2. The intelligent CAD design feature retrieval method based on knowledge graph as described in claim 1, characterized in that, Parse the user's search request information input into the CAD design space to generate a first search intent factor, a second search intent factor, and a third search intent factor, including: Based on the request parsing history event set, we can classify them to obtain the request geometric feature parsing event set, the request functional feature parsing event set, and the request process feature parsing event set; Based on the requested geometric feature parsing event set, the retrieval request information is adversarially distilled to obtain the first retrieval intent factor; Based on the request function feature parsing event set, the retrieval request information is subjected to adversarial distillation intent parsing to obtain the second retrieval intent factor; Based on the requested process feature parsing event set, the retrieval request information is subjected to adversarial distillation intent parsing to obtain the third retrieval intent factor.

3. The intelligent CAD design feature retrieval method based on knowledge graph as described in claim 2, characterized in that, Based on the requested geometric feature parsing event set, adversarial distillation intent parsing is performed on the retrieval request information to obtain the first retrieval intent factor, including: The request geometric feature parsing optimization set is obtained by performing a distribution equalization perturbation on the request geometric feature parsing event set using an adversarial sample generator. The CNN network is trained under supervision based on the requested geometric feature parsing event set to construct a first geometric intent parsing network; The RNN network is trained under supervision based on the requested geometric feature parsing optimization set to construct a second geometric intent parsing network. Distillation learning and fusion are performed on the first geometric intent parsing network and the second geometric intent parsing network to obtain the third geometric intent parsing network; The retrieval request information is input into the third geometric intent parsing network, and the first retrieval intent factor is output.

4. The intelligent CAD design feature retrieval method based on knowledge graph as described in claim 1, characterized in that, Based on the CAD design space, a multimodal analytical graph structure mapping is performed to establish a CAD design analytical graph atlas space, including: Based on the CAD design space, read the m-th design drawing, where m is a positive integer, 1≤m≤M; Based on the m-th design drawing, geometric feature analysis is performed to obtain multiple design geometric feature vectors; Based on the m-th design drawing, functional feature analysis is performed to obtain multiple design functional feature vectors; Based on the m-th design drawing, process feature analysis is performed to obtain multiple design process feature vectors; Based on the knowledge graph, the multiple design geometric feature vectors, the multiple design functional feature vectors, and the multiple design process feature vectors are mapped to a graph structure to obtain the m-th design analytical graph, and the m-th design analytical graph is added to the CAD design analytical graph space.

5. The intelligent CAD design feature retrieval method based on knowledge graph as described in claim 1, characterized in that, The stability of the plurality of geometric matching evaluation coefficients is evaluated to obtain a plurality of geometric matching stability coefficients, including: The first search intent factor is perturbed by neighboring senses to obtain multiple perturbation geometric intent factors; Based on the multiple perturbation geometric intention factors, the matching degree of each design geometric feature vector in the m-th design analytical map is evaluated to obtain multiple geometric matching evaluation groups; Curve fitting is performed based on the multiple geometric matching evaluation groups to obtain the geometric matching evaluation curves corresponding to each design geometric feature vector; Based on the geometric matching evaluation curves, a central tendency analysis is performed to determine multiple perturbation matching evaluation central values. The stability of the plurality of geometric matching evaluation coefficients is evaluated based on the plurality of perturbation matching evaluation set values, thereby generating the plurality of geometric matching stability coefficients.

6. The intelligent CAD design feature retrieval method based on knowledge graph as described in claim 1, characterized in that, Based on the retrieval attention association mechanism, a multi-domain retrieval collaborative analysis is performed on the first fitness optimization retrieval domain, the second fitness optimization retrieval domain, and the third fitness optimization retrieval domain to obtain a fourth fitness optimization retrieval domain, including: Activate the retrieval attention association mechanism, which includes geometric retrieval fitness weight, functional retrieval fitness weight, and process retrieval fitness weight; Multi-domain retrieval is performed based on the first fitness optimization retrieval domain, the second fitness optimization retrieval domain, and the third fitness optimization retrieval domain to obtain multiple retrieval matching results. Each retrieval matching result corresponds to a retrieval fitness evaluation matrix. Based on the retrieval attention association mechanism, the retrieval collaboration fitness of the multiple retrieval matching results is analyzed to obtain multiple retrieval collaboration fitness; The multiple search matching results are optimized and identified based on the multiple search collaborative fitness values ​​to generate the fourth fitness optimization search domain.

7. The intelligent CAD design feature retrieval method based on knowledge graph as described in claim 1, characterized in that, The CAD design space includes M CAD design drawings, where M is a positive integer greater than 1.

8. The intelligent CAD design feature retrieval method based on knowledge graph as described in claim 1, characterized in that, The geometric retrieval fitness parsing conditions include geometric matching evaluation weights and geometric matching stability weights.

9. A CAD design feature intelligent retrieval system based on knowledge graphs, characterized in that, The steps for implementing the knowledge graph-based intelligent CAD design feature retrieval method according to any one of claims 1 to 8 include: The request parsing module is used to parse the retrieval request information input by the user into the CAD design space and generate the first retrieval intent factor, the second retrieval intent factor and the third retrieval intent factor; The space establishment module is used to perform multimodal analytical graph structure mapping based on the CAD design space and establish the CAD design analytical graph atlas space; The first fitness detection module is used to perform retrieval fitness detection on the CAD design analysis map space according to the first retrieval intent factor, and obtain the first fitness optimization retrieval domain. The second fitness detection module is used to perform retrieval fitness detection on the CAD design analysis map space according to the second retrieval intent factor, and obtain the second fitness optimization retrieval domain. The third fitness detection module is used to perform retrieval fitness detection on the CAD design analysis map space according to the third retrieval intent factor to obtain the third fitness optimization retrieval domain; the multi-domain retrieval collaborative analysis module is used to perform multi-domain retrieval collaborative analysis on the first fitness optimization retrieval domain, the second fitness optimization retrieval domain and the third fitness optimization retrieval domain according to the retrieval attention association mechanism to obtain the fourth fitness optimization retrieval domain.

Images