A novel graph neural networks approach for 3D product model retrieval

Abstract

ABSTRACT With the quick development of intelligent manufacturing technology, increasing of 3D product models data resource raises the need for product models retrieval method that can fully exploit model information, whereas the existing methods only concern geometric information with little or no manufacturing information. Model-Based Definition (MBD) 3D engineering has gained wide popularity in recent times which enables smart manufacturing by attaching product manufacturing information. Its tip orientation to the effective reuse of MBD for reducing development time and creating an efficient digital thread. In this paper, we proposed the MBD-based attribute adjacency graph neural networks (MAAGNN) with attention mechanism. Firstly, the MBD-based attribute adjacency graph (MAAG) is established to fully express the product semantic information, which includes geometric and non-geometric information of the MBD model. Then graph neural network (GNN) is employed to embed the MAAG into continuous vector spaces, and attention mechanism is introduced to adaptively recognize more important features information, which can solve the ambiguity problem caused by blend features such as chamfer. Finally, the model is trained to achieve fast and accurate retrieval of the MBD models. Experiments demonstrate its potential advantages to meet the requirements of engineering applications.