An industrial design intelligent drawing data management system
By constructing an industrial design AI generation model based on neural symbol fusion and implementing multi-dimensional verification and compensation, the problem that existing AI generation models cannot generate data that satisfies topological consistency and ISO geometric constraints has been solved, and high-precision drawing data management has been achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHANGSHA YUNZHAN NETWORK TECHNOLOGY CO LTD
- Filing Date
- 2026-04-28
- Publication Date
- 2026-07-21
AI Technical Summary
Existing AI-generated models are unable to generate accurate boundary representations that satisfy topological consistency and ISO geometric constraints in industrial design, resulting in geometric and topological inaccuracies and geometric discontinuities.
We construct an industrial design AI generation model based on neural symbol fusion. By injecting supervision signals of ISO geometric constraints and dimensional constraints during the model training process, we generate initial design drawing data that satisfies topological consistency and constraint legality, and implement multi-dimensional verification and error compensation.
It effectively solves the geometric discontinuity problem of AI-generated models, ensures that the generated data meets ISO geometric tolerance standards, improves drawing accuracy and reliability, and avoids production rework.
Smart Images
Figure CN122433145A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of industrial design management technology, specifically to an intelligent drawing data management system for industrial design. Background Technology
[0002] Industrial design intelligent drawing data management is an upgrade of traditional computer-aided design (CAD) drawing tools into a system that integrates intelligent design, data-driven approach, and full lifecycle management. By integrating intelligent rules, a unified data platform, and collaborative technologies, it solves core pain points in traditional design such as version chaos, data silos, and inefficient collaboration.
[0003] In fields such as computer vision or web design, minute rendering deviations, such as a few pixels of misalignment, are usually tolerable. However, in industrial design, an error of ±0.1mm can cause an entire component to be unassembled, or even lead to serious engineering accidents. Current mainstream AI generation models, such as diffusion models, are essentially based on rasterization or simple vector paths for output. They are good at generating visually “correct” images, but they cannot understand or generate accurate boundary representation (B-Rep) topological relationships and NURBS surfaces. This results in serious geometric discontinuities in AI-generated drawings, with problems of geometric and topological inaccuracies. Summary of the Invention
[0004] The purpose of this invention is to provide an intelligent drawing data management system for industrial design, which solves the technical problems mentioned in the background art.
[0005] The objective of this invention can be achieved through the following technical solutions:
[0006] An intelligent drawing data management system for industrial design includes:
[0007] Initial design drawing data constraint generation module: Constructs an industrial design AI generation model based on neural symbol fusion. By injecting supervision signals of ISO geometric constraints and dimensional constraints during the model training process, the model is forced to learn and comply with engineering rules, generating initial design drawing data that satisfies topological consistency and constraint legality.
[0008] Multi-dimensional verification and compensation processing fusion module: Based on the generated initial design drawing data, it performs multi-dimensional verification of constraint compliance and multi-dimensional error compensation, verifies whether the compensated constraints meet the standards, and outputs manufacturable drawings that meet the assembly accuracy requirements.
[0009] Furthermore, the constructed industrial design AI generation model has an overall architecture that includes a generator, a symbolic logic module, and a constraint supervisor.
[0010] The generator is a U-Net structure, which specifically includes an input layer, an encoder, a decoder, and an output layer.
[0011] The symbolic logic module includes the formal expression of ISO constraints and the construction of a constraint rule base.
[0012] Furthermore, the constraint supervisor includes a matching degree calculation module and a supervision signal generation module;
[0013] The matching degree calculation module is used to extract the attribute data of vertices, edges and faces from the B-Rep topology information tensor output by the generator, convert it into a first-order logical expression, match the logical expression of the generated result with the predicates in the constraint rule base, and calculate the matching degree.
[0014] The supervision signal generation module is used to analyze the calculated matching degree, dynamically generate gradient penalty signals based on the analysis results, and adjust the convolutional layer weights of the generator through backpropagation.
[0015] Furthermore, when constructing the constraint-annotated industrial design training dataset, high-precision B-Rep models and NURBS surface data were selected from publicly available industrial design databases and enterprise internal design libraries;
[0016] Ensure the data contains complete topological information and geometric dimension annotations;
[0017] In addition, ISO geometric constraints and dimensional constraints are manually labeled for each model.
[0018] Furthermore, when generating constraint supervision signals, the B-Rep topological adjacency matrix output by the generator and the adjacency matrix of the labeled data are extracted;
[0019] Calculate the edit distance between two topologies and normalize the edit distance to the range [0,1] as the geometric constraint loss;
[0020] Additionally, the geometric dimension tensor output by the generator and the dimension values of the annotation data are extracted, and the mean square error of the generated dimension and the annotation dimension is calculated as the dimension constraint loss.
[0021] The calculated geometric constraint loss and dimensional constraint loss are fused together to obtain the final constraint loss term.
[0022] Furthermore, when fusing the constraint loss and the diffusion model loss, the original loss term of the diffusion model adopts the noise prediction loss, which is the mean square error between the noise predicted by the model and the actual noise.
[0023] Set constraint loss weights to ensure that the model learns engineering rules first, while preserving the diversity of the generated data, and construct the total loss function.
[0024] Furthermore, constraint validity and topology consistency verification are performed. During geometric constraint verification, it is checked whether the generated initial B-Rep topology matches the constraint logic predicate.
[0025] During topology consistency verification, the adjacency matrix of the B-Rep topology is calculated to check for topology errors; if the verification fails, the constraint supervisor generates a constraint correction signal.
[0026] Furthermore, the constraint compliance is verified in multiple dimensions, including automated verification of geometric constraints, comparative verification of dimensional constraints tolerances, and detection of topological consistency defects, and outputs geometric constraint verification reports and dimensional constraint out-of-tolerance reports.
[0027] Furthermore, based on the output geometric constraint verification report and dimensional constraint out-of-tolerance report, the number and value of deviation items are counted according to the constraint type, and differentiated compensation operations are performed for different geometric constraint types, including parallel constraint compensation, coaxial constraint compensation, and tangential constraint compensation.
[0028] Furthermore, when implementing parallel constraint compensation, the angular deviation between the current side direction vector and the target direction vector is calculated, and the angle compensation amount is calculated.
[0029] Using the vertex of the edge as the rotation center, rotate the direction vector of the edge around an axis perpendicular to both vectors by an angle compensation amount to obtain the corrected direction vector.
[0030] Furthermore, when implementing coaxial constraint compensation, the actual distance between the axes of the two cylindrical surfaces is calculated, and the amount of axis translation is also calculated.
[0031] Additionally, calculate the deviation distance between the axes of the two cylindrical surfaces and the ideal datum, and translate the cylindrical surface with the larger deviation distance along the direction of the axis connection to ensure that the axes coincide.
[0032] Furthermore, when implementing tangent constraint compensation, the curvature continuity of the surface and the edge is calculated, and the adjustment amount of the surface control points is calculated; the coordinates of the surface control points are adjusted so that the curvature continuity meets the curvature continuity requirement.
[0033] Furthermore, when implementing dimensional constraint tolerance compensation, the topological entity corresponding to the out-of-tolerance dimension is located based on the preset dimension annotation mapping table;
[0034] And, calculate the vertex coordinate adjustment amount;
[0035] The target vertex is translated using the vertex coordinate adjustment to achieve size compensation.
[0036] Compared to existing solutions, the beneficial effects achieved by this invention are:
[0037] This invention uses symbolic logic constraints and supervisory signal injection to ensure that the initial design data generated by the model can effectively meet ISO geometric constraints, avoiding the geometric discontinuity problem of traditional AI generation models. The introduced constraint loss term forces the model to learn engineering rules, which can effectively improve the constraint compliance rate of the generated data compared with traditional models.
[0038] This invention implements multi-dimensional constraint compliance verification and multi-dimensional error compensation, automatically verifying and compensating for constraints, ensuring that drawings conform to ISO geometric tolerance standards, avoiding production rework due to design violations, eliminating the need for secondary conversion, effectively improving drawing accuracy and reliability, and effectively solving the geometric and topological inaccuracies of existing technical solutions. Attached Figure Description
[0039] The invention will now be further described with reference to the accompanying drawings.
[0040] Figure 1 This is a flowchart illustrating the operation of an intelligent drawing data management system for industrial design according to the present invention. Detailed Implementation
[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0042] like Figure 1 As shown, this invention is an intelligent drawing data management system for industrial design, including an initial design drawing data constraint generation module and a multi-dimensional verification and compensation processing fusion module; specifically:
[0043] Initial Design Drawing Data Constraint Generation Module: This module constructs a neural symbolic fusion-based industrial design AI generation model. By injecting supervisory signals of ISO geometric and dimensional constraints during model training, it forces the model to learn and adhere to engineering rules, generating initial design drawing data that satisfies topological consistency and constraint validity. Specific steps include:
[0044] When constructing the neural symbolic fusion industrial design AI generation model architecture, a diffusion model is used as the basic generation network, and a symbolic logic module is introduced to realize the formal expression and supervision of engineering constraints. The overall model architecture includes a generator, a symbolic logic module, and a constraint supervisor.
[0045] The generator uses a U-Net architecture, specifically comprising an input layer, encoder, decoder, and output layer; specifically:
[0046] The input layer receives a 128×128×3 random noise tensor and a 768-dimensional design task description embedding vector, which are then fused into a 128×128×771 input tensor through a splicing layer.
[0047] The encoder contains four convolutional blocks, each consisting of two 3×3 convolutional layers, a ReLU activation function, and a 2×2 max pooling layer, which extract features and reduce the tensor size to 8×8×512.
[0048] The decoder contains four deconvolution blocks, each consisting of two 3×3 deconvolution layers, a ReLU activation function, and skip connections, restoring the tensor size to 128×128×3.
[0049] The output layer uses a 1×1 convolutional layer to convert the tensor into a 128×128×256 B-Rep topological information tensor and a 128×128×64 geometric dimension tensor, which represent the connection relationships of vertices, edges, and faces and the key dimension values, respectively.
[0050] Initialize the U-Net convolutional layers with ImageNet pre-trained weights to accelerate model convergence;
[0051] Furthermore, the generator was pre-trained on an unconstrained industrial design dataset to enable it to generate basic design graphics. The pre-training methods used were all existing conventional techniques, and the specific implementation steps will not be elaborated here.
[0052] The symbolic logic module includes the formal expression of ISO constraints and the construction of a constraint rule base; specifically:
[0053] When implementing the formal expression of ISO constraints, first-order logical predicates are defined for 32 types of ISO geometric constraints, including parallel constraints, tangent constraints, and coaxial constraints.
[0054] Parallel constraint, including parallel (edge 1, edge 2), indicates that the angle between the direction vectors of edge 1 and edge 2 is 0°;
[0055] Tangency constraint, including tangency (face 1, face 2), indicates that the normal vectors of face 1 and face 2 are collinear at the intersection point;
[0056] Coaxial constraint, including coaxial (cylinder 1, cylinder 2), indicates that the central axes of cylinder 1 and cylinder 2 coincide;
[0057] In addition, for 17 types of ISO dimensional constraints, numerical constraint predicates are defined, including linear distance constraints and angular tolerance constraints;
[0058] Linear distance constraint, including distance (point1, point2, 10±0.1), indicating that the distance between point1 and point2 is 10mm, with a tolerance of ±0.1mm;
[0059] Angle tolerance constraint, including angle (side 1, side 2, 90±0.5), means that the angle between side 1 and side 2 is 90°, with a tolerance of ±0.5°;
[0060] When constructing the constraint rule base, all logical predicates of ISO constraints are stored in the Pyke knowledge base to build a rule matching engine; and a constraint parsing interface is developed to support the automatic extraction of constraint information from the design task description and its conversion into logical predicates.
[0061] The constraint supervisor includes a matching degree calculation module and a supervision signal generation module;
[0062] The matching degree calculation module is used to extract the attribute data of vertices, edges and faces from the B-Rep topological information tensor output by the generator and convert it into a first-order logical expression. This is a conventional technical solution, and the specific implementation steps will not be elaborated here.
[0063] The Prover9 theorem prover is used to match the generated logical expression with the predicates in the constraint rule base, and the matching degree is calculated. : ;in, The number of constraints to be satisfied; This represents the total number of constraints.
[0064] The supervision signal generation module is used to analyze the matching degree of the calculation. If the matching degree is lower than 95%, a gradient penalty signal is generated and the weights of the convolutional layer of the generator are adjusted through backpropagation. This is a conventional technical solution, and the specific implementation steps are not described here.
[0065] Furthermore, for specific violations of constraints, constraint correction suggestions are generated, such as "edge 1 and edge 2 are not parallel, it is recommended to adjust the direction vector of edge 2", and the suggestions are fed back to the generator for local adjustment.
[0066] When constructing the constraint-annotated industrial design training dataset, 1000 sets of high-precision B-Rep models and NURBS surface data were selected from publicly available industrial design databases and enterprise internal design libraries, covering typical parts in fields such as machinery, aerospace, and automobiles; industrial design databases, such as GrabCAD and Thingiverse; typical parts, such as gear shafts, aircraft airfoils, and engine blocks;
[0067] Ensure that the data contains complete topological information and geometric dimension annotations, with an accuracy of no less than ±0.01mm; the topological information includes the connection relationships of vertices, edges, and faces;
[0068] Additionally, using the constraint annotation tool developed by OpenCASCADE, 32 types of ISO geometric constraints and 17 types of dimensional constraints were manually annotated for each model; ISO geometric constraints include parallelism, tangency, and coaxiality; dimensional constraints include linear distance and angular tolerance.
[0069] The B-Rep model is converted into a trainable tensor format, the vertex coordinates are normalized to the range of [0,1], the topology is converted into an adjacency matrix, and the constraint annotation information is converted into a first-order logic expression that can be recognized by the symbolic logic module and stored as a rule file in Prover9 format.
[0070] The datasets are divided into training, validation, and test sets in an 8:1:1 ratio to ensure that each dataset covers the same type of constraints and parts.
[0071] When generating constraint supervision signals, extract the B-Rep topological adjacency matrix output by the generator and the adjacency matrix of the labeled data;
[0072] The NetworkX library is used to calculate the edit distance between two topologies. The edit distance is the minimum number of operations required to add, delete, or modify vertices or edges. The edit distance is normalized to the range [0,1] and used as the geometric constraint loss. The corresponding expression is:
[0073] ;in, For geometric constraint loss; To edit distance; To maximize the possible edit distance;
[0074] For example, if the topology of the generated model needs to be modified by 3 edges compared to the annotation, and the maximum possible edit distance is 10, then the geometric constraint loss... ;
[0075] Furthermore, the geometric dimension tensor output by the generator and the dimension values of the annotation data are extracted. The mean square error of the generated dimensions and annotation dimensions is calculated using PyTorch's MSELoss function, which serves as the dimension constraint loss. The corresponding expression is:
[0076] ;in, This is due to dimensional constraint losses; Index for dimensional constraints; The number of dimensional constraints; The actual dimensions are those indicated. Dimension values generated for the model;
[0077] The calculated geometric constraint loss and dimensional constraint loss are then fused to obtain the final constraint loss term: ;in, All are loss weights. The default values are 0.6 and 0.4 respectively;
[0078] When merging constraint loss and diffusion model loss, the original loss term of the diffusion model... The noise prediction loss is used, which is the mean square error between the noise predicted by the model and the actual noise.
[0079] Set constraint loss weights To ensure the model prioritizes learning engineering rules while preserving the diversity of generated data, the total loss function is constructed, and its corresponding expression is:
[0080] ;
[0081] When setting training parameters, the optimizer is AdamW, and the learning rate is set to... The weight decay coefficient is 0.01;
[0082] The batch size is set to 16, the training rounds are 100, and the validation set is evaluated every 5 rounds.
[0083] The total loss, constraint loss, and diffusion loss of the training and validation sets are recorded in real time. After each round of validation, the constraint compliance rate, i.e., the proportion of samples that meet the ISO constraints, is calculated. Training is terminated early when the constraint compliance rate of the validation set exceeds 95% for 5 consecutive rounds. The model training uses existing conventional technical solutions, and the specific implementation steps are not described here.
[0084] After training, input the task encoding vector corresponding to the design task description, for example, to generate a gear shaft model with parallel constraints. The model outputs initial design drawing data containing B-Rep topology, NURBS surface parameters and dimension information.
[0085] The B-Rep topology includes the connection relationships between vertices, edges, and faces, ensuring that ISO geometric constraints are met.
[0086] NURBS surface parameters include control points, weight factors, and node vectors to ensure the G2 continuity and smoothness of the surface;
[0087] The dimensioning information includes the values and tolerance ranges of key dimensions to ensure compliance with ISO dimensional constraints;
[0088] It should be noted that during generator inference, the task encoding vector and random noise are input into the U-Net generator. Through the learned constraint rules, the direction vector of the keyway side edge that satisfies the parallel constraint is automatically constructed, and the initial B-Rep topology tensor and initial geometric dimension tensor are output.
[0089] Meanwhile, the symbolic logic module loads the constraint logic predicates into the rule base, and the constraint supervisor synchronously verifies the constraint matching degree of the generated result. If the matching degree is less than 95%, a gradient signal is generated to trigger a local adjustment of the generator.
[0090] Perform constraint legality and topology consistency verification. When verifying geometric constraints, use the Prover9 theorem prover to check whether the generated initial B-Rep topology matches the constraint logic predicate. For example, verify whether the angle between the direction vectors of the side edges of the keyway is 0°, which corresponds to the parallel constraint.
[0091] During topology consistency verification, the adjacency matrix of the B-Rep topology is calculated using NetworkX to check for topology errors such as isolated vertices, edges, and unclosed surfaces, ensuring that the model has no geometric breaks.
[0092] If the verification fails, the constraint supervisor generates a constraint correction signal, such as "the angle between the direction vectors of side 1 and side 2 of the keyway is 2°, and the direction vector of side 2 needs to be adjusted", and feeds it back to the generator for local fine-tuning until the constraint matching degree is ≥95%;
[0093] Output verified B-Rep topology data, including the final parameters of vertices, edges, and faces;
[0094] When optimizing NURBS surface parameters, OpenCASCADE is used to convert the B-Rep surface into a NURBS surface, automatically generating initial control points, weight factors, and node vectors, and adjusting the key parameters of the NURBS surface, including control points, node vectors, and weight factors, and outputting the optimized NURBS surface parameters, specifically including control point coordinates, weight factors, and node vectors.
[0095] Among them, control points: the coordinates of control points on the surface boundary are adjusted by the least squares method to ensure the curvature continuity of adjacent surfaces;
[0096] Node vectors: In areas where the curvature of the surface changes significantly, such as the root of a gear tooth, increase the node density and set the node vectors to non-uniform triple nodes to ensure a smooth transition of the surface;
[0097] Weighting factor: The default setting is 1.0. If local deformation occurs on the surface, the weighting factor of the corresponding control point will be adjusted. For example, adjusting the weighting factor to 1.2 will bring the surface closer to the control point.
[0098] When generating ISO-compliant dimensioning based on B-Rep topology and NURBS surface parameters, a rule engine is used to identify key dimensional features from the validated B-Rep topology, such as the shaft diameter, length, and keyway width of the gear shaft.
[0099] Tolerances are assigned to critical dimensions according to ISO 2768 standard. The default tolerance for linear dimensions is ±0.1mm, and the default tolerance for angular dimensions is ±0.5°.
[0100] When standardizing parameters, the dimensioning information is converted to the ISO standard format, for example, the linear distance is marked as 100±0.1mm, and the geometric tolerance is marked as ○0.02, which is the roundness tolerance;
[0101] It outputs standardized dimensioning information, including dimension values, tolerance ranges, and datum features, and converts B-Rep topology, NURBS surface parameters, and dimensioning information into industry standard formats such as STEP and IGES to ensure that they can be directly read by CAD software.
[0102] In this embodiment of the invention, by injecting symbolic logic constraints and supervisory signals, the initial design data generated by the model can effectively meet ISO geometric constraints, avoiding the geometric discontinuity problem of traditional AI generation models; the introduced constraint loss term forces the model to learn engineering rules, which can effectively improve the constraint compliance rate of the generated data compared with traditional models.
[0103] Multi-dimensional verification and compensation processing fusion module: Based on the generated initial design drawing data, it performs multi-dimensional verification of constraint compliance and multi-dimensional error compensation, verifies whether the compensated constraints meet the standards, and outputs manufacturable drawings that meet assembly accuracy requirements. Specific steps include:
[0104] The constraint compliance verification is carried out in multiple dimensions, including automated verification of geometric constraints, comparative verification of dimensional constraint tolerances, and detection of topological consistency defects, and these are performed in parallel.
[0105] When performing automated geometric constraint verification, the 32 types of geometric constraints defined in ISO 10303-21 are loaded into the rule base of the constraint verification engine.
[0106] The actual values of parallel constraints, coaxial constraints, and tangent constraints are measured by traversing the topological entities of the model using the BRepCheck_Analyzer module of OpenCASCADE.
[0107] The actual value of the parallel constraint is calculated by measuring the angle between the direction vectors of the two sides.
[0108] The actual value of the coaxial constraint is calculated by measuring the distance between the axes of the two cylindrical surfaces;
[0109] The actual value of the tangency constraint is calculated by determining the curvature continuity of the surface and the edge;
[0110] The measured actual value is compared with its corresponding ISO standard target value. If the measured actual value is greater than the corresponding ISO standard target value, it is marked as non-compliant.
[0111] Conversely, it is marked as meeting the standard;
[0112] When outputting the geometric constraint verification report, include the actual value, target value, and whether the standard is met for each constraint;
[0113] When performing dimensional constraint tolerance comparison and verification, based on the accuracy requirements of the design task, the ISO 2768-m medium precision tolerance standard is matched, and the dimension annotation information is read using a Pandas script. The actual values of each linear dimension and angular dimension are compared with the ISO allowable tolerance range, and the type of out-of-tolerance dimension is marked, such as critical dimension or non-critical dimension. The dimensional constraint out-of-tolerance report is output, which includes out-of-tolerance dimension, actual value, allowable range, and deviation value.
[0114] For linear dimensions, such as a shaft diameter of 20mm, the allowable range is 19.9~20.1mm. If the actual value is 20.12mm, it is marked as "out of tolerance".
[0115] Angular dimensions, such as a keyway angle of 90°, have an allowable range of 89.5° to 90.5°;
[0116] When performing topology consistency defect detection, the TopologyChecker module of OpenCASCADE is used to traverse the vertices, edges, and faces of the model and construct the topology adjacency matrix.
[0117] Adjacency matrix analysis is used to identify topological defects, specifically including:
[0118] Surface overlap: The spatial overlap of two surfaces is ≥90%;
[0119] Edge break: An edge is incorrectly split into two discontinuous topological entities;
[0120] Unclosed surface: The boundary edges of a surface do not form a closed loop;
[0121] Isolated vertex: A vertex that is not connected to any edge;
[0122] Record the topology ID and spatial coordinates of each defect, generate a visual marker map, and output a topology defect report, including defect type, location, quantity, and visual marker map;
[0123] Based on the output geometric constraint verification report and dimensional constraint out-of-tolerance report, the number and value of deviation items are counted according to the constraint type. Differentiated compensation operations are performed for different geometric constraint types, including parallel constraint compensation, coaxial constraint compensation, and tangent constraint compensation. The specific constraint types include parallel, coaxial, and dimensional.
[0124] When implementing parallel constraint compensation, calculate the angular deviation between the current side direction vector and the target direction vector. And calculate the angle compensation amount. ;in, The unit direction vector of the edge to be corrected is the actual geometric feature parameter extracted from the B-Rep topology. The ideal direction vector is required by the ISO 10303-21 standard, and it must be completely consistent with the direction vector of the reference edge in the parallel constraint; This is the parallel constraint compensation coefficient, with a value range of 1 to 1.5. The critical constraint is set to 1.5, and the non-critical constraint is set to 1. The critical and non-critical constraints can be customized according to the actual application scenario and application requirements; 0 corresponds to a parallel constraint target angle of 0°.
[0125] Rotate the direction vector of the edge around an axis perpendicular to both vectors, using the vertex of the edge as the center of rotation. This yields the corrected direction vector;
[0126] For example, the original direction vector (0.996, 0.087, 0) and the target vector (1.0, 0, 0) form an angle of approximately 5°, and the compensation amount... =1.2×5°=6°, and the direction vector is corrected to (1.0,0,0) after rotation;
[0127] When implementing coaxial constraint compensation, calculate the actual distance d between the axes of the two cylindrical surfaces;
[0128] Calculate the translation of the axes based on the actual distance between the axes of the two cylindrical surfaces. ; This is a compensation coefficient specifically for coaxial constraints, with a value range of 1.2 to 1.8 and a default value of 1.5.
[0129] Additionally, calculate the deviation distance between the axes of the two cylindrical surfaces and the ideal datum, and quantify the degree of deviation; if the design specification clearly specifies the reference feature, then the axis of the reference feature is used as the ideal datum axis; if it is a symmetrical coaxial constraint, such as two symmetrical cylindrical surfaces being coaxial, then calculate the geometric mean axis of the two axes as the ideal datum.
[0130] The cylindrical surface with the large deviation is translated along the direction of the axis. Ensure the axes coincide;
[0131] When implementing tangent constraint compensation, calculate the curvature continuity of the surface and edges. ;in, The curvature of a surface is the curvature value at the point where the surface is tangent to an edge, reflecting the degree of curvature of the surface at that point. Curvature is the curvature of the edge, which is the curvature value of the edge at the point where it is tangent to the surface, reflecting the degree of bending of the edge; the curvature of the surface and the curvature of the edge can be obtained through the GeomAPI_Curvature module and the BRepAdaptor_Curve module of OpenCASCADE, respectively;
[0132] Calculate the adjustment amount of the surface control points based on the curvature continuity of the surface and its edges. ;in, This is a compensation coefficient specific to tangent constraints, with a value range of 1 to 1.2 and a default value of 1.1.
[0133] Adjust the coordinates of the control points on the surface to ensure that the curvature continuity meets the requirements, for example... ;
[0134] Output the B-Rep topology after geometric constraint compensation and the geometric constraint compensation log;
[0135] When implementing dimensional constraint tolerance compensation, the topological entity corresponding to the out-of-tolerance dimension is located based on the preset dimension annotation mapping table. For example, the shaft diameter corresponds to the generatrix vertex of the cylindrical surface. The dimension annotation mapping table can associate the dimension annotation information in the B-Rep topology structure and NURBS surface parameters converted to JSON format with the geometric features in the STEP file to ensure that each dimension corresponds to a unique topological entity. For example, the shaft diameter corresponds to the cylindrical surface.
[0136] And, calculate the vertex coordinate adjustment amount. ;in, These are the actual linear dimensions measured from the B-Rep model; The target size value;
[0137] For example, the actual shaft diameter is 20.12 mm, the target diameter is 20.0 mm, the upper tolerance limit is 20.1 mm, and the adjustment amount... =0.12mm, shift the vertex of the generatrix of the cylindrical surface 0.06mm towards the axis, adjust the radius by 0.06mm, adjust the diameter by 0.12mm, and the corrected shaft diameter is 20.0mm;
[0138] The target vertex is translated using the BRepBuilderAPI_Transform module of OpenCASCADE to complete the size compensation, and the B-Rep topology after size constraint compensation and size compensation log are output.
[0139] When verifying whether the compensated constraints meet the standards, the geometric constraint verification tool is called to remeasure the actual values of the constraints for the compensated deviation items; the geometric constraint verification tool includes OpenCASCADE BRepCheck_Analyzer and local verification scripts;
[0140] If the deviation still exceeds the threshold after compensation, readjust the compensation coefficient and compensate again, for example, increase the compensation coefficient by 10%; if new topological defects appear after compensation, such as face overlap, revert to the state before compensation and adjust the compensation amount.
[0141] The compensated geometric constraint deviation must meet the corresponding threshold requirements, and the dimensional constraints must fall completely within the ISO tolerance range.
[0142] In this embodiment of the invention, by implementing multi-dimensional verification and multi-dimensional error compensation for constraint compliance, automatic constraint verification and compensation ensure that the drawings conform to ISO geometric tolerance standards, avoid production rework caused by design violations, eliminate the need for secondary conversion, effectively improve drawing accuracy and reliability, and effectively solve the geometric and topological inaccuracies of existing technical solutions.
[0143] In the several embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for example, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.
[0144] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.
[0145] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module described above can be implemented in hardware or in the form of hardware plus software functional modules.
[0146] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the essential characteristics of the present invention.
[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims
1. An intelligent drawing data management system for industrial design, characterized in that, include: Initial design drawing data constraint generation module: Constructs an industrial design AI generation model based on neural symbol fusion. By injecting supervision signals of ISO geometric constraints and dimensional constraints during the model training process, the model is forced to learn and comply with engineering rules, generating initial design drawing data that satisfies topological consistency and constraint legality. Multi-dimensional verification and compensation processing fusion module: Based on the generated initial design drawing data, it performs multi-dimensional verification of constraint compliance and multi-dimensional error compensation, verifies whether the compensated constraints meet the standards, and outputs manufacturable drawings that meet the assembly accuracy requirements.
2. The intelligent drawing data management system for industrial design according to claim 1, characterized in that, The constructed industrial design AI generation model has an overall architecture that includes a generator, a symbolic logic module, and a constraint supervisor.
3. The intelligent drawing data management system for industrial design according to claim 2, characterized in that, The constraint supervisor includes a matching degree calculation module and a supervision signal generation module; The matching degree calculation module is used to extract the attribute data of vertices, edges and faces from the B-Rep topology information tensor output by the generator, convert it into a first-order logical expression, match the logical expression of the generated result with the predicates in the constraint rule base, and calculate the matching degree. The supervision signal generation module is used to analyze the calculated matching degree, dynamically generate gradient penalty signals based on the analysis results, and adjust the convolutional layer weights of the generator through backpropagation.
4. The intelligent drawing data management system for industrial design according to claim 3, characterized in that, When generating constraint supervision signals, extract the B-Rep topological adjacency matrix output by the generator and the adjacency matrix of the labeled data; Calculate the edit distance between two topologies and normalize the edit distance to the range [0,1] as the geometric constraint loss; Additionally, the geometric dimension tensor output by the generator and the dimension values of the annotation data are extracted, and the mean square error of the generated dimension and the annotation dimension is calculated as the dimension constraint loss. The calculated geometric constraint loss and dimensional constraint loss are fused together to obtain the final constraint loss term.
5. The intelligent drawing data management system for industrial design according to claim 4, characterized in that, Perform constraint validity and topology consistency verification. During geometric constraint verification, check whether the generated initial B-Rep topology matches the constraint logic predicate. During topology consistency verification, the adjacency matrix of the B-Rep topology is calculated to check for topology errors. If the verification fails, the constraint supervisor generates a constraint correction signal.
6. The intelligent drawing data management system for industrial design according to claim 1, characterized in that, The constraint compliance verification is performed in multiple dimensions, including automated geometric constraint verification, dimensional constraint tolerance comparison verification, and topological consistency defect detection, and outputs geometric constraint verification report and dimensional constraint out-of-tolerance report.
7. The intelligent drawing data management system for industrial design according to claim 6, characterized in that, Based on the output geometric constraint verification report and dimensional constraint out-of-tolerance report, the number and value of deviation items are counted according to the constraint type. Differentiated compensation operations are performed for different geometric constraint types, including parallel constraint compensation, coaxial constraint compensation, and tangential constraint compensation.
8. The intelligent drawing data management system for industrial design according to claim 7, characterized in that, When implementing parallel constraint compensation, calculate the angle deviation between the current side direction vector and the target direction vector, and calculate the angle compensation amount; Using the vertex of the edge as the rotation center, rotate the direction vector of the edge around an axis perpendicular to both vectors by an angle compensation amount to obtain the corrected direction vector.
9. The intelligent drawing data management system for industrial design according to claim 7, characterized in that, When implementing coaxial constraint compensation, calculate the actual distance between the axes of the two cylindrical surfaces and the amount of axis translation; Additionally, calculate the deviation distance between the axes of the two cylindrical surfaces and the ideal datum, and translate the cylindrical surface with the larger deviation distance along the direction of the axis connection to ensure that the axes coincide.
10. The intelligent drawing data management system for industrial design according to claim 7, characterized in that, When implementing tangent constraint compensation, calculate the curvature continuity of the surface and edges, and calculate the adjustment amount of the surface control points; adjust the coordinates of the surface control points so that the curvature continuity meets the curvature continuity requirement.