From CAD Data to Semantic Graphs: Integration and Derivation of Domain-Specific Knowledge within Classified Graphs

Abstract

This paper presents a hybrid framework that integrates machine learning and ontological reasoning in graphbased representations of CAD assemblies to enable functional analysis. Building on the Core Product Model (CPM), we define multiple levels of information richness within our hybrid-data CAD assembly graphs, which fuse geometric and parametric data from STEP and CATIA sources into a unified semantic representation. By embedding CAD domain expertise into ontological rule sets and coupling them with machine learning models refined through ontology-guided calibration, our method elevates low-information assembly graphs to higher levels within a defined CPM hierarchy of semantic significance. The enriched graphs support the inference of implicit degrees of freedom and the automated propagation of motion by operationalizing ontological rules. Demonstrated on gearbox assemblies, the approach reconstructs complete kinematic chains from input to output. By uniting rule-based reasoning with machine-learned prediction, this work establishes a foundation for semantic, interpretable, and knowledge-driven CAD graph analysis.