Search

CN-121981230-A - Sheet metal part manufacturability reasoning method based on space-semantic pattern alignment

CN121981230ACN 121981230 ACN121981230 ACN 121981230ACN-121981230-A

Abstract

The invention discloses a sheet metal part manufacturability reasoning method based on space-semantic graph alignment, which relates to the field of design evaluation and industrial knowledge reasoning for manufacturing, and comprises the following steps of carrying out geometric analysis on a CAD geometric model of a sheet metal part to be evaluated, and abstracting geometric features and topological/metric space relations thereof into a computable space semantic graph; the method comprises the steps of carrying out semantic analysis on a process specification described by natural language, converting the process rule into formal logic verification expression by utilizing a large language model through contextual learning, generating an executable domain-specific language verification script, executing the script on a space semantic graph, realizing deterministic reasoning through graph matching and attribute verification, completing precise mapping of text rules and geometric features and detecting violations, and outputting an interpretable diagnosis result comprising locating violations features, triggering rules and numerical evidence. Aiming at the problems of the process rule of 'natural language-geometric model' semantic gap, poor adaptability of the traditional rule system and strong data dependence and unexplained end-to-end learning method, the invention can quickly adapt to new rules without a large amount of labeling data and retraining the model, can accurately identify the infraction of the micro-size and the space relation, and has reasoning rigor, interpretability and engineering usability.

Inventors

  • ZHANG YUHANG

Assignees

  • 西南科技大学

Dates

Publication Date
20260505
Application Date
20260326

Claims (8)

  1. 1. The sheet metal part manufacturability reasoning method based on space-semantic pattern alignment is characterized by comprising the following steps of: firstly, carrying out geometric analysis on a sheet metal part geometric model to be evaluated, and constructing a space semantic graph containing geometric feature nodes, space relationship edges and attribute information; performing semantic analysis on the process rules of the natural language description, converting the process rules into formalized logic verification expression, and further generating executable domain-specific language scripts; step three, executing the domain specific language script on the space semantic graph, and realizing deterministic reasoning through sub-graph matching and attribute verification to obtain a rule compliance judging result; And step four, outputting interpretable diagnostic information corresponding to the compliance determination result.
  2. 2. The method for reasoning manufacturability of sheet metal parts based on space-semantic graph alignment according to claim 1, wherein in the step of performing geometric analysis on the geometric model of the sheet metal parts to be evaluated to construct a space semantic graph: The continuous boundary representation model is discretized into a graph structure in which nodes are used to represent geometric features with engineering semantics, edges are used to explicitly encode spatial topological relationships between features, and attributes are used to store metrology data.
  3. 3. The method for reasoning manufacturability of sheet metal parts based on space-semantic graph alignment according to claim 1, wherein in the step of performing semantic parsing on the process rules of natural language description: The unstructured natural language rule set is converted into a formalized first-order predicate logic expression set by using a large language model to convert fuzzy constraints in the rule into deterministic logic predicates.
  4. 4. The method for sheet metal manufacturability inference based on space-semantic graph alignment of claim 1, wherein in the step of generating executable domain-specific language scripts from process rules: domain-specific languages based on Python, which define strict context-free grammar and contain predefined domain primitives to limit the generation space, are used as logical verification script carriers.
  5. 5. The method for reasoning about manufacturability of sheet metal parts based on alignment of spatial-semantic graphs according to claim 1, wherein in the step of performing script-implemented deterministic reasoning on the spatial semantic graph: and calculating constraint satisfaction states through sub-graph matching and attribute verification traversing candidate nodes or node pairs in the atlas, so that mapping and conflict detection of text rules and geometric features are realized.
  6. 6. The method for reasoning about manufacturability of sheet metal parts based on alignment of spatial-semantic graphs according to claim 1, wherein in the step of performing script-implemented deterministic reasoning on the spatial semantic graph: The reasoning process includes logical variable map binding, candidate pruning, and diagnostic report generation.
  7. 7. The method for inference of manufacturability of sheet metal parts based on space-semantic pattern alignment according to claim 6, wherein in the step of candidate pruning: aiming at the spatial locality of the industrial design rule, introducing a spatial index pruning strategy based on R-Tree, screening a candidate pair set meeting a topological adjacency or distance threshold, and then executing logic operation on the candidate set.
  8. 8. The method of sheet metal manufacturability inference based on spatio-semantic pattern alignment of claim 1, wherein in said outputting interpretable diagnostic information step: When the violation set is detected, node identification and geometric attribute values participating in operation are extracted, a rule threshold is combined to generate an evidence chain, and the diagnosis information follows the structural format of entity index-logical basis-numerical evidence.

Description

Sheet metal part manufacturability reasoning method based on space-semantic pattern alignment Technical Field The invention belongs to the field of computer aided design and manufacturing (CAD/CAM), and particularly relates to a method for abstracting a sheet metal part CAD model into a space semantic graph, converting natural language process rules into executable logic verification scripts, and carrying out deterministic reasoning on the space semantic graph to realize automatic evaluation of sheet metal part manufacturability. Background Manufacturability design evaluation (DfM) is an important link to shorten product development cycle and reduce production cost. However, there is a significant semantic gap in practical industrial scenarios, where the process specifications are mostly presented in unstructured natural language in the design manual, while the design objects are presented in highly structured geometric models, which are difficult to interact directly. The prior art mainly comprises two types of (1) a geometric verification method based on rules and expert systems, wherein the method has strong reasoning certainty and good interpretability, but the rule base construction depends on a large amount of manual hard codes, when the process specification is changed, the bottom layer codes are often required to be reconstructed, the maintenance cost is high, and meanwhile, fuzzy knowledge in a natural language form is difficult to process, so that a large amount of unstructured experience is difficult to effectively use. (2) The data driving method based on geometric deep learning generally needs large-scale high-quality marking data, industrial three-dimensional defect marking data are scarce and sensitive, and moreover, the end-to-end model has black box property, so that clear violation reasons and positioning information are difficult to give, and interpretable feedback required by engineering design modification is difficult to meet. Meanwhile, the large language model has the capability of reading the process text, but the inherent 'illusion' problem can not directly replace strict engineering calculation, so that a neural symbol cooperative paradigm needs to be explored, namely, the large language model is responsible for translating natural language rules into deterministic logic codes, and a graph algorithm carries out accurate geometric operation on a space semantic graph, so that the unification of flexibility and strictness is realized. Disclosure of Invention Aiming at the semantic isomerism problem of unstructured process text-structured geometric model in sheet metal part DfM evaluation, the invention provides a neural symbol reasoning method based on space-semantic graph alignment, which forms manufacturability analysis tasks into cross-modal constraint satisfaction problems, constructs global mapping from geometric model and rule input to structured diagnosis result output, and realizes automatic checking of design rules, illegal feature positioning and interpretable closed-loop feedback. In order to achieve the above purpose, the invention has the following design scheme: Step one, constructing an integral mapping and an input-output space. The system inputs a Cartesian product formed by a space geometric model and a natural language rule set, outputs a diagnosis triplet set for describing geometric entity compliance, and decouples a global mapping function into a compound operation of three subfunctions of geometric analysis, rule analysis and nerve symbol reasoning. And step two, geometrically analyzing and constructing a Space Semantic Graph (SSG). Defining a geometric analysis function, discretizing a continuous boundary representation (B-Rep) model into a machine-inferable space semantic graph structure, namely enabling node representations to have engineering semantic geometric features (such as holes, grooves, bending lines and the like), and enabling edges to explicitly encode spatial topology/geometric relationships (such as adjacency, inclusion, parallelism and the like) among the features, wherein an attribute set stores measurement data, so that abstraction from pure geometry to structured symbol knowledge is realized, and key topology invariants are reserved. And thirdly, formalized definition and attribute modeling are carried out on the space semantic graph. The method comprises the steps of defining a space semantic graph as a four-element group, wherein the four-element group comprises a geometric feature node set, a space relation edge set, a node attribute mapping space and a heterogeneous relation type set, representing node attributes by adopting mixed attribute vectors, at least comprising centroid coordinates, fixed-length geometric parameter vectors (aperture, depth, bending angle and the like, and insufficient dimension adopts zero filling), and One-hot semantic coding of feature types, modeling the graph as an attribute multiple graph, enabling a plurali