STEP-based Model Recommendation Method for the Exchange and Reuse of Digital Twins

Abstract

To support the design and optimization of human-centric manufacturing systems in the Industry 5.0 era, Model Based Definition (MBD) models with STEP knowledge graph (STEP KG) recommendation are crucial for exchanging and reusing digital twin models. Existing methods based on graph convolutional networks (GCN) focus on geometric semantics but overlook the needed correlation engineering semantics in the STEP KG. Our paper introduces a Quaternion Diffusion Graph Convolutional Network (QDGCN) recommendation framework, comprising quaternion semantic diffusion and quaternion parameter diffusion. The quaternion semantic diffusion method uses quaternion to combine multiple layers of semantic diffusion into a single set transformation operation and constructs the quaternion-based multi-layer semantic model on the STEP KG. The quaternion parameter diffusion method uses a quaternion parameter generation mechanism based on the diffusion model. It generates different weight coefficients for identifying the main node features in the STEP KG. The fusion of the two solves the inconsistency problem between geometric and engineering semantics. We compared QDGCN with state-of-the-art methods on real datasets, and the detailed analysis of experimental results demonstrates the effectiveness of QDGCN.