Parametric design generation and consistency checking method and device, and storage medium

Abstract

This invention discloses a method, apparatus, and storage medium for parametric design generation and consistency verification. By integrating engineering intent text and visual evidence, and utilizing unified representation, constraint learning and feasibility feedback, constrained reasoning and generation, and consistency verification and evidence-based derivation, it completes the engineering generation and management of parametric design results. Employing the technical solution of this invention, through element binding, learnable feasibility injection, and consistency-driven generation, improves the compilability, assembly feasibility, and editing stability of the design results.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to a parametric design generation and consistency verification method and apparatus, and a storage medium. Background Technology

[0002] In practical engineering design, simply having a shape that "looks correct" is far from enough. Any result lacking parameters, constraints, and topological semantics will fail at the very first step of implementation: it cannot be compiled into a boundary representation, sketch constraints are unsolvable, dimensions and tolerances cannot be transferred, assembly import will cause interference, and even minor parameter modifications will trigger a complete topology flip. Existing geometry generation methods, primarily based on vision or text, excel at providing visual shapes but are almost powerless when it comes to the fundamental engineering requirements of "editability," "assemblyability," and "traceability." A seemingly successful model will immediately reveal serious flaws such as missing parameters, lack of constraints, and topological instability once it enters the design change or part derivation stage, becoming a non-reusable "one-off result."

[0003] Multimodal input nominally alleviates this dilemma, but traditional workflows generally treat text, images, and point clouds separately, inevitably leading to inconsistent units and coordinates. Key elements of the engineering intent cannot be stably bound to the sketch dimension and topology. Any subtle noise on the input side will be amplified in subsequent generation: the diameter tolerance of the same hole system, expressed in millimeters in text, may be misread as pixels in image annotation; the same datum lacks consistent registration relationships across different views, causing constraint solutions to be based on the wrong coordinate system from the outset. As a result, the parametric program appears complete on the surface but harbors contradictions; once secondary editing is performed, the constraint chain breaks, and the model instantly collapses.

[0004] More importantly, existing methods often place feasibility and consistency checks after generation, which is equivalent to reaching the end of a wrong path before turning back. Constraints that determine the usability of an engineering project, such as minimum wall thickness, assembly gaps, reachability, and manifold, do not provide feedforward guidance during the generation process; there is no minimum repair mechanism when candidate solutions approach the feasible boundary, the search process involves repeated trial and error, and computational power is consumed by meaningless branches. Even if a compilable model is occasionally obtained, unacceptable interference often occurs during assembly verification, or irreversible topological mutations occur during parameter fine-tuning. The so-called "success" is just a one-time, unmaintainable accident.

[0005] Finally, almost all existing processes lack evidentiary and auditable capabilities. There's no stable index between the generated results and their sources, making it impossible to answer questions like "Why is this parameter the current value?", "Where was this rule triggered?", and "Which modification led to subsequent instability?" Without a chain of evidence, design teams are forced to rely on verbal explanations and manual review, causing rework costs to rise exponentially, and rendering version management and compliance documentation meaningless. Faced with stringent manufacturing and assembly standards, such processes are virtually unusable: they cannot guarantee delivery quality, support large-scale reuse, or survive the rapid iteration of complex products. Summary of the Invention

[0006] To address the problems existing in the prior art, the present invention provides a parametric design generation and consistency verification method and apparatus, and a storage medium.

[0007] To achieve the above objectives, the present invention provides the following solution: A parametric design generation and consistency verification method includes: Step S1: Preprocess the engineering intent text and visual evidence respectively to obtain the parameter items, sketch parameters, topology parameters, and manufacturing and assembly rules in the engineering intent; Step S2: Bind the parameter items in the engineering intent with the sketch parameters, topology parameters, and manufacturing and assembly rules to form an auditable unified representation; Step S3: Construct a soft constraint system for manufacturing and assembly. The soft constraint system includes: minimum wall thickness, interference volume, manifold and self-intersection risk, accessibility and assembly feasibility, and dimensional and tolerance deviation indices. Simultaneously, call the symbolic constraint solver in a low-frequency manner to obtain feasibility labels and minimum repair amounts. Use the judgment result of the symbolic constraint solver as a teacher signal to train the lightweight verification module. Perform temperature calibration on the feasibility score output by the lightweight verification module to obtain the calibrated feasibility probability. Step S4: Generate parameterized candidate solutions based on unified representation, and use a scoring function consisting of feasibility probability, representation consistency and complexity penalty to progressively screen the candidates; perform early stopping and pruning on branches that violate hard constraints or have scores below the threshold; trigger minimum repair and verification for critical candidates, generate a compilable parameterized design program consisting of sketch sequence, parameter list and constraint set, and generate boundary representation model; Step S5: Verify and threshold based on topological legality, assembly readiness, and editing stability to obtain the comprehensive confidence level; at the same time, establish an evidence index for the parametric design program and boundary representation model, associate rendering fragments, engineering icon annotations, rule hit entries, parameter modification trajectories, and minimum repair records, and output a structured result containing the comprehensive confidence level and evidence list.

[0008] The present invention also provides a parametric design generation and consistency verification device, comprising: The first processing module is used to preprocess the engineering intent text and visual evidence respectively to obtain the parameter items, sketch parameters, topology parameters and manufacturing and assembly rules in the engineering intent. The second processing module is used to bind the parameter items in the engineering intent with the sketch parameters, topology parameters, and manufacturing and assembly rules to form an auditable unified representation; The third processing module is used to construct a soft constraint system for manufacturing and assembly. The soft constraint system includes: minimum wall thickness, interference volume, manifold and self-intersection risk, accessibility and assembly feasibility, and dimensional and tolerance deviation indices. At the same time, it calls the symbolic constraint solver in a low-frequency manner to obtain feasibility labels and minimum repair amounts, and uses the judgment result of the symbolic constraint solver as a teacher signal to train the lightweight verification module. The feasibility score output by the lightweight verification module is calibrated by temperature to obtain the calibrated feasibility probability. The fourth processing module is used to generate parameterized candidate solutions based on a unified representation, and to progressively screen the candidates using a scoring function consisting of feasibility probability, representation consistency, and complexity penalty; to perform early stopping and pruning on branches that violate hard constraints or have scores below the threshold; to trigger minimum repair and verification on critical candidates; to generate a compilable parameterized design program consisting of a sketch sequence, parameter list, and constraint set; and to generate a boundary representation model. The fifth processing module is used to verify and determine the threshold based on topological legality, assembly readiness, and editing stability to obtain the comprehensive confidence level. At the same time, it establishes an evidence index for the parametric design program and boundary representation model, associates rendering fragments, engineering icon annotations, rule hit entries, parameter modification trajectories, and minimum repair records, and outputs a structured result containing the comprehensive confidence level and evidence list.

[0009] The present invention also provides a storage medium storing a computer program, which executes a parametric design generation and consistency verification method during runtime.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves an integrated closed loop from engineering intent and visual evidence to parameterized results without altering existing design processes and data formats, demonstrating significant engineering practical value and scalability. Firstly, at the input and representation level, by unifying units, coordinates, and benchmarks, and binding key parameters in the engineering intent with sketch dimensions, topological relationships, manufacturing, and assembly rules, a stable semantic-to-geometric correspondence is established, reducing ambiguity and mismatches and improving the determinism of downstream calculations. Secondly, at the generation process level, soft constraints such as minimum wall thickness, interference, reachability, and dimensional and tolerance deviations are directly embedded into the inference process, and low-frequency constraint solver feedback is introduced to form feedforward feasibility guidance. Combined with early stopping pruning and minimum repair mechanisms, invalid searches and boundary oscillations are effectively suppressed, improving compileability success rate and result convergence speed. Third, regarding result quality, a triple check is implemented, encompassing topological legality, assembly readiness, and editing stability, which are integrated with the generated score. This significantly reduces risks such as self-intersection, Boolean failures, assembly interference, and topological mutations caused by parameter fine-tuning, ensuring the results can be used for secondary editing and assembly verification. Fourth, regarding evidence and compliance, a queryable evidence index is established for parameterized programs, boundary representations, rule hits, parameter modification trajectories, and minimum repair records, forming an auditable closed-loop traceability capability, facilitating design review, version management, and process documentation. Fifth, at the system and ecosystem level, the method is highly modular, and the rule base and thresholds can be conditionally configured according to industry standards and enterprise specifications, possessing good portability and evolution potential, facilitating integration with existing computer-aided design platforms, assembly planning, and design retrieval systems. In summary, this invention can reduce manual review and rework costs, shorten design iteration cycles, and enhance structured reuse and scalable delivery capabilities while ensuring compilability, assembly success, and editing stability. Attached Figure Description

[0011] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of the parametric design generation and consistency verification method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the shaft and support assembly according to an embodiment of the present invention. Detailed Implementation

[0013] 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.

[0014] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0015] Example 1 like Figure 1 As shown, this invention provides a parametric design generation and consistency verification method. It combines image, text, and point cloud inputs, first completing feature anchoring and parameter extraction, then constructing parametric sketches and 3D features. Subsequently, consistency verification and minimum repair are performed based on geometry, structure, and rules, ultimately outputting reusable parametric design program structure blocks, assembly exploded thumbnails, verification result tables, and evidence index lists. Specifically, it includes: Step S100: Multimodal input processing; preprocess the engineering intent text and visual evidence (multi-view images, point clouds, engineering drawings) to obtain the parameter items and sketch parameters, topology parameters, and manufacturing and assembly rules in the engineering intent, including: Step S101: Text parsing; standardize the terminology, extract the hierarchy and unify the units of the engineering intent text to obtain the parameter item set P (target size, allowable tolerance, material / working condition); at the same time, based on the terminology-geometry dictionary, establish semantic alignment items with the sketch parameters S, candidate elements with the topology parameters T, and locate the corresponding manufacturing and assembly rule candidates R (identified by standard number / rule ID).

[0016] Step S102: Image processing; perform distortion correction, edge enhancement and feature extraction on the view / image to form two explicit sets of sketch parameters S (line segments / arcs / apertures / angles and their constraints) and topological parameters T (face-edge-volume, adjacent / contact pairs, normals); and trigger candidate terms of manufacturing and assembly rules R (minimum wall thickness, interference volume, assembly direction accessibility, etc.) based on the identified geometric features.

[0017] Step S103: Engineering drawing / point cloud analysis; outlier removal, scale correction, and wireframe vectorization are performed on the point cloud / engineering drawing; the connection relationship between the dimension-tolerance network and topology parameter T of the sketch parameter S is completed; the specific thresholds and applicable conditions of manufacturing and assembly rules R are extracted from the title block, technical specifications, and BOM. At this point, stage S1 produces four sets of elements {P, S, T, R}, along with their source indices and uncertainties.

[0018] Step S200: Unified Representation Construction and Feature Binding; Using the above four feature sets P (parameter items), S (sketch parameters), T (topology parameters), and R (manufacturing and assembly rules) as input, the binding, mapping, and uncertainty assessment are completed sequentially, and an auditable unified representation is output, along with an uncertainty assessment. This includes: Step S201: Unify element definition and identification; After standardizing units and terminology, establish standardized dictionaries for the parameter item set P (target size, tolerance, material / working condition, etc.), the sketch parameter set S (line segment / arc / aperture / angle and its sketch constraints), and the topology parameter set T (face / edge / volume, adjacent / contact pairs, normal), and assign a unique identifier (ID) and source index (text line number, image fragment ID, slice ID) to each item. Simultaneously, establish a ternary index table. Used to record parameter items With sketch parameters Topological elements The candidate correspondence. Output the normalized result. and index table .

[0019] Step S202: Binding rules; based on index table The parameter item—sketching dimension—topology element is bound and mapped in a one-to-many / many-to-one manner, and each mapping is associated with the corresponding entry in the manufacturing and assembly rule R (including threshold, tolerance direction, unit of measurement and rule ID).

[0020] Binding example: aperture "Bind to sketch circle diameter / radius" Topological surface of hole wall ; "Thickness" is bound to the minimum measure (t) between two adjacent surfaces / walls and associated with the "Minimum Wall Thickness" rule entry. ; The draft angle α is bound to the sketch / topological taper of the forming direction. And associate it with the "draft angle" rule entry. .

[0021] Output binding relationship table Reference index with rules.

[0022] Step S203: Uncertainty assessment; For terms in P, S, and T that have insufficient, conflicting, or missing sources, calculate the confidence interval / confidence score, and... and Marked with "Requires Review" or "Insufficient Evidence"; when the necessary conditions for triggering rule R are not met, a review prompt is output (e.g., "Aperture D is not explicitly given; the range is estimated based on the view and array detection"). Output with uncertainty and verification markers. , .

[0023] Step S204: Unified representation generation; based on maintaining consistency between geometric and semantic context. and Generate a unified representation It includes at least: traceable binding of parameters, sketches, topology, and rules; field units and coordinate systems; thresholds and tolerances; and evidence indexes (text line numbers, image fragment IDs, and slice IDs). It provides retrieval and auditing interfaces using structured data (such as JSON / CSV) for subsequent constraint learning and feasibility assessment. The output is presented in a unified format. .

[0024] Step S300: Constraint Learning and Feasibility Feedback; Construct a soft constraint system for manufacturing and assembly, which includes: minimum wall thickness, interference volume, manifold and self-intersection risk, accessibility and assembly feasibility, and dimensional and tolerance deviation indices; Simultaneously, call the symbolic constraint solver in a low-frequency manner to obtain feasibility labels and minimum repair amounts, and use the judgment result of the symbolic constraint solver as a teacher signal to train the lightweight verification module; Perform temperature calibration on the feasibility score output by the lightweight verification module to obtain the calibrated feasibility probability, which is used for subsequent scoring and threshold determination; For ease of subsequent reference, denote the soft constraint index vector as... , where represent minimum wall thickness, interference volume, manifold score, self-intersection risk, assembly direction accessibility, assembly feasibility, and dimensional / tolerance deviation, respectively; the threshold (or target interval) is configured by the rule base and denoted as . ,include: Step S301: Symbolic constraint solving and index calculation; the symbolic constraint solver is called infrequently to determine the feasibility of candidate geometries / parameters; and each parameter is calculated simultaneously. Six categories of indicators: 1. Minimum wall thickness Find the minimum measure between adjacent walls; and the threshold. Comparison requires the use of markers. .

[0025] 2. Interference volume : Boolean intersection of the assembly; pass condition is .

[0026] 3. Manifold score Risk of self-fertilization Detect and quantize open edges / non-manifolds / self-intersections; and use thresholds. , The comparison is generated by the marker.

[0027] 4. Assembly direction accessibility : Perform reachability determination on the assembly direction cone, and output 0 / 1.

[0028] 5. Assembly Feasibility Scoring or judging based on comprehensive rules such as fit / fastener accommodation, and threshold values. Compare.

[0029] 6. Dimensional / Tolerance Deviation : Calculate the normalized deviation between the target size and the measured / reconstructed size; subject to the following conditions. Based on this, a vector-based index-by-index structure is formed. Detect conflicts (including rule IDs and geometric element IDs) and output the "overall feasible / infeasible" teacher label; finally, output... Conflict location records.

[0030] Step S302: Calculate the minimum repair amount; solve the problem while keeping the design intent unchanged. And record the indicator vectors before / after the repair. With vector If multiple solutions exist, the correct solution should be chosen first. and The solution with the greatest improvement; when repair is not feasible, output the constraint alternative / mitigation solution and its impact. Repair discrepancies in the record.

[0031] Step S303: Lightweight verification module distillation and probability calibration; based on S301 and A lightweight verification module is trained for teacher signals, and the lightweight verification module simultaneously outputs the overall feasibility probability. and index-wise probability vector The probabilities are calibrated using temperature to obtain comparable values, which are then fed back to the generation and scoring modules for early stopping / pruning and boundary sample retention; by default, the probabilities are generated by the fusion function. Consistent overall probability.

[0032] Step S400: Constrained Reasoning and Generation; A constrained guided search is employed to perform early stopping and pruning on infeasible candidates, and trigger minimum repair and verification on critical candidates, generating a compilable parametric design program and boundary representation model. This includes: Step S401: Candidate generation; generate parameterized candidates based on unified representation and feature binding; sample uncertain parameters within the confidence interval.

[0033] Step S402: Scoring function; candidate scores consist of feasibility probability, uniform representation consistency, and complexity penalty. in, Let be the feasibility probability, representing the confidence level that the candidate solution satisfies the hard constraints; The consistency score for the unified representation is used to measure the geometric / semantic matching degree between candidate parameterization schemes and the unified representation; As a complexity penalty, the number of sketch commands, constraints, Boolean iterations, and reconstruction time are normalized and then linearly combined; the larger the value, the more complex the task. Non-negative weights and = 1, and its value is determined during the training or deployment configuration phase.

[0034] Step S403: Early stopping and pruning; Prune immediately when the candidate's comprehensive score is below the threshold or violates hard constraints; Stop obviously infeasible paths early to avoid error propagation and computational waste.

[0035] Step S404: Minimum Repair Trigger and Review; When the candidate approaches the feasible threshold, call the minimum repair amount in step S302 to fine-tune a few key parameters; Immediately after the repair, review the soft and hard constraints and update the score.

[0036] Step S405: Output of results; output a parametric design program consisting of a sketch sequence, parameter list, constraint set, and boundary representation model; compilability criteria include geometric closure, normal consistency, Boolean operation stability, and self-intersection exclusion.

[0037] Step S500: Consistency Verification and Evidence-Based Export; Verification and threshold determination are performed based on topological legality, assembly readiness, and editing stability. Results are output on high-confidence paths, and the evidence and results are indexed to achieve traceable export, including: Step S501: Topology validity check; check topological constraints such as loops, connectivity, holes and weak features; provide location and repair suggestions for non-manifolds.

[0038] Step S502: Assembly readiness check; mainly based on fit clearance, interference volume, assembly direction accessibility, and fastener accommodation space, the range of hole-shaft fit clearance g is determined as follows: in, This represents the decision function (0 / 1 quantity) for whether the match passes. ,but =1, otherwise =0 indicates the inner diameter of the hole. Indicates the outer diameter of the shaft. Indicates radial clearance, [ , ] represents the qualified range given by the rule base.

[0039] The interference check results are shown in Figure 2. Candidate generation and pruning are performed in the "Constrained Search" pipeline: when interval determination or interference check fails, a "Repair" step is triggered to minimize the adjustment amount Δp on the allowed parameter set; after repair, it re-enters "Review," only proceeding if the condition is met. ∈[ , ]and The process is successful. The "Parameter Disturbance Comparison" below shows the case where the diameter is adjusted from d to d′, indicating that the repair eliminates interference and maintains the wall thickness / gap as required without changing the assembly relationship. Finally, the "Verification Result Table" and "Evidence Index Table" are output at the bottom.

[0040] Step S503: Edit the stability check; compare small parameter perturbations to verify topological invariance and constraint solvability; if reconstruction failure or constraint collapse occurs, mark it as unstable.

[0041] Step S504: Confidence fusion and threshold determination; The scores from steps S501 to S503 are weighted or fused with the scores from the generation stage to obtain a comprehensive confidence score and the results are selected accordingly; Evidence is retained for boundary cases and a review is prompted.

[0042] Export the evidence index for the parametric design program and model, which includes at least rendering fragments, engineering diagram annotations, rule entries, parameter modification trajectories, and repair suggestion records, and provide it to external parties for auditing and traceability in a structured format.

[0043] Table 1

[0044] Table 2

[0045] The key point of this invention is: Multimodal input processing: On the input side, collaborative parsing of engineering intent text, multi-view images, point clouds, and engineering drawings is performed. On the text side, recognition, terminology standardization, and dimensional unification are completed, extracting target structures, key dimensions and tolerances, and process and assembly elements. On the image side, denoising, enhancement, and edge stabilization are performed, extracting contours and candidate sketch primitives. On the point cloud side, outlier removal, normal estimation, and multi-source registration are completed. Through the unification of units, coordinate systems, and reference surfaces, a stable reference is provided for feature binding and subsequent calculations.

[0046] Unified Representation and Feature Binding: A unified processing layer defines parameters such as holes, slots, ribs, chamfers, fillets, fits, dimensions, and tolerances. These parameters are then traceably mapped to sketch dimensions, constraints, and topology, while their origin and priority are recorded. Uncertainty assessments are performed for source conflicts and missing evidence, and confidence intervals are provided, ensuring consistent referencing of semantic and geometric features in subsequent processes.

[0047] Constraint Learning and Feasibility Feedback: A soft constraint system is constructed for manufacturing and assembly, including indicators such as minimum wall thickness, interference volume, manifold risk, accessibility and assembly feasibility, and dimensional and tolerance deviations. Tolerance bands and penalty terms are set. The constraint solver is invoked in a low-frequency manner to output feasible or infeasible labels, conflict location, and minimum repair suggestions. The feedback is distilled into a lightweight verification module and probabilistically calibrated to ensure that different rule scores have a unified dimension and comparability, providing feedforward guidance for the generation process.

[0048] Constrained Reasoning and Minimal Repair: Parameterized candidate solutions are generated based on a unified representation. The scoring function integrates feasibility probability, representation consistency, and complexity penalties. Early stopping and pruning are performed on branches that violate hard constraints or have scores below a threshold. Minimal repair is triggered and verified when a candidate approaches the feasible boundary. The final output is a parameterized design program consisting of a sequence of sketches, a parameter list, and a constraint set, and a boundary representation model is generated to ensure compilability for subsequent verification and downstream use.

[0049] Consistency verification and confidence fusion: A joint verification is conducted from three aspects: topological validity, assembly readiness, and editing stability, and a threshold determination is performed. Topological validity includes geometric closure, normal consistency, Boolean operation stability, and self-intersection exclusion; assembly readiness includes mating clearance, interference, assembly sequence reachability, and fastener accommodation space; editing stability examines topological invariance and constraint solvability under small parameter perturbations. The three types of scores are fused with the generated score to form a comprehensive confidence score, which is used for result screening and quality grading, suppressing false positives and structural instability.

[0050] Evidence Indexing and Export: Establishes an evidence chain for parametric design programs and boundary representation models, linking at least rendered fragments, engineering diagram annotations, rule hit entries, parameter modification trajectories, and minimum repair records to form a queryable evidence index. Provides structured export and retrieval interfaces, supporting direct invocation by systems such as design review, assembly planning, and design retrieval, and allows for conditional configuration and dynamic updates according to industry standards and enterprise specifications.

[0051] Example 2 The present invention also provides a parametric design generation and consistency verification device, comprising: The first processing module is used to preprocess the engineering intent text and visual evidence respectively to obtain the parameter items, sketch parameters, topology parameters and manufacturing and assembly rules in the engineering intent. The second processing module is used to bind the parameter items in the engineering intent with the sketch parameters, topology parameters, and manufacturing and assembly rules to form an auditable unified representation; The third processing module is used to construct a soft constraint system for manufacturing and assembly. The soft constraint system includes: minimum wall thickness, interference volume, manifold and self-intersection risk, accessibility and assembly feasibility, and dimensional and tolerance deviation indices. At the same time, it calls the symbolic constraint solver in a low-frequency manner to obtain feasibility labels and minimum repair amounts, and uses the judgment result of the symbolic constraint solver as a teacher signal to train the lightweight verification module. The feasibility score output by the lightweight verification module is calibrated by temperature to obtain the calibrated feasibility probability. The fourth processing module is used to generate parameterized candidate solutions based on a unified representation, and to progressively screen the candidates using a scoring function consisting of feasibility probability, representation consistency, and complexity penalty; to perform early stopping and pruning on branches that violate hard constraints or have scores below the threshold; to trigger minimum repair and verification on critical candidates; to generate a compilable parameterized design program consisting of a sketch sequence, parameter list, and constraint set; and to generate a boundary representation model. The fifth processing module is used to verify and determine the threshold based on topological legality, assembly readiness, and editing stability to obtain the comprehensive confidence level. At the same time, it establishes an evidence index for the parametric design program and boundary representation model, associates rendering fragments, engineering icon annotations, rule hit entries, parameter modification trajectories, and minimum repair records, and outputs a structured result containing the comprehensive confidence level and evidence list.

[0052] Example 3 The present invention also provides a storage medium storing a computer program, which executes a parametric design generation and consistency verification method during runtime.

[0053] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for parametric design generation and consistency verification, characterized in that, include: Step S1: Preprocess the engineering intent text and visual evidence respectively to obtain the parameter items, sketch parameters, topology parameters, and manufacturing and assembly rules in the engineering intent; Step S2: Bind the parameter items in the engineering intent with the sketch parameters, topology parameters, and manufacturing and assembly rules to form an auditable unified representation; Step S3: Construct a soft constraint system for manufacturing and assembly. The soft constraint system includes: minimum wall thickness, interference volume, manifold and self-intersection risk, accessibility and assembly feasibility, and dimensional and tolerance deviation indices. Simultaneously, call the symbolic constraint solver in a low-frequency manner to obtain feasibility labels and minimum repair amounts. Use the judgment result of the symbolic constraint solver as a teacher signal to train the lightweight verification module. Perform temperature calibration on the feasibility score output by the lightweight verification module to obtain the calibrated feasibility probability. Step S4: Generate parameterized candidate solutions based on unified representation, and use a scoring function consisting of feasibility probability, representation consistency and complexity penalty to progressively screen the candidates; perform early stopping and pruning on branches that violate hard constraints or have scores below the threshold; trigger minimum repair and verification for critical candidates, generate a compilable parameterized design program consisting of sketch sequence, parameter list and constraint set, and generate boundary representation model; Step S5: Verify and threshold based on topological legality, assembly readiness, and editing stability to obtain the comprehensive confidence level; at the same time, establish an evidence index for the parametric design program and boundary representation model, associate rendering fragments, engineering icon annotations, rule hit entries, parameter modification trajectories, and minimum repair records, and output a structured result containing the comprehensive confidence level and evidence list.

2. A parametric design generation and consistency verification device, characterized in that, include: The first processing module is used to preprocess the engineering intent text and visual evidence respectively to obtain the parameter items, sketch parameters, topology parameters and manufacturing and assembly rules in the engineering intent. The second processing module is used to bind the parameter items in the engineering intent with the sketch parameters, topology parameters, and manufacturing and assembly rules to form an auditable unified representation; The third processing module is used to construct a soft constraint system for manufacturing and assembly. The soft constraint system includes: minimum wall thickness, interference volume, manifold and self-intersection risk, accessibility and assembly feasibility, and dimensional and tolerance deviation indices. At the same time, it calls the symbolic constraint solver in a low-frequency manner to obtain feasibility labels and minimum repair amounts, and uses the judgment result of the symbolic constraint solver as a teacher signal to train the lightweight verification module. The feasibility score output by the lightweight verification module is calibrated by temperature to obtain the calibrated feasibility probability. The fourth processing module is used to generate parameterized candidate solutions based on a unified representation, and to progressively screen the candidates using a scoring function consisting of feasibility probability, representation consistency, and complexity penalty; to perform early stopping and pruning on branches that violate hard constraints or have scores below the threshold; to trigger minimum repair and verification on critical candidates; to generate a compilable parameterized design program consisting of a sketch sequence, parameter list, and constraint set; and to generate a boundary representation model. The fifth processing module is used to verify and determine the threshold based on topological legality, assembly readiness, and editing stability to obtain the comprehensive confidence level. At the same time, it establishes an evidence index for the parametric design program and boundary representation model, associates rendering fragments, engineering icon annotations, rule hit entries, parameter modification trajectories, and minimum repair records, and outputs a structured result containing the comprehensive confidence level and evidence list.

3. A storage medium, characterized in that, The storage medium stores a computer program, which executes the parameterized design generation and consistency verification method as described in claim 1 when it runs.

Images