MP-GFormer: A 3D-Geometry-Aware Dynamic Graph Transformer Approach for Machining Process Planning

TL;DR

MP-GFormer addresses machining process planning by integrating evolving 3D geometry into a dynamic graph learning framework. It combines a Graph Attention Network encoder for geometry-aware feature extraction with a transformer decoder to capture spatio-temporal dependencies between sequential process graphs and the initial design, enabling accurate machining operation sequence predictions. Experiments on synthesized STL/BRep data show MP-GFormer outperforms DyGFormer and DiffPool, with notable gains in main and sub-operation accuracy, validating the importance of geometry-aware dynamics in MP. The work lays a foundation for geometry-driven planning and suggests future directions, including diffusion-based approaches to jointly predict 3D geometry and operation sequences.

Abstract

Machining process planning (MP) is inherently complex due to structural and geometrical dependencies among part features and machining operations. A key challenge lies in capturing dynamic interdependencies that evolve with distinct part geometries as operations are performed. Machine learning has been applied to address challenges in MP, such as operation selection and machining sequence prediction. Dynamic graph learning (DGL) has been widely used to model dynamic systems, thanks to its ability to integrate spatio-temporal relationships. However, in MP, while existing DGL approaches can capture these dependencies, they fail to incorporate three-dimensional (3D) geometric information of parts and thus lack domain awareness in predicting machining operation sequences. To address this limitation, we propose MP-GFormer, a 3D-geometry-aware dynamic graph transformer that integrates evolving 3D geometric representations into DGL through an attention mechanism to predict machining operation sequences. Our approach leverages StereoLithography surface meshes representing the 3D geometry of a part after each machining operation, with the boundary representation method used for the initial 3D designs. We evaluate MP-GFormer on a synthesized dataset and demonstrate that the method achieves improvements of 24\% and 36\% in accuracy for main and sub-operation predictions, respectively, compared to state-of-the-art approaches.

Figures (10)

• Figure 1: The proposed framework integrates a GAT, an attention mechanism, and a transformer to capture and learn the spatio-temporal and geometry-aware dependencies in MP.
• Figure 2: The MP-GFormer architecture. Sequential machining process graphs and the initial design graph are used as inputs. The architecture consists of three main stages: (a) an encoder, where node and edge are mapped into a latent space using a GAT encoder, and a cross-attention mechanism captures interdependencies between process and design graphs; (b) a Transformer decoder that learns dependencies across sequential graphs, machining operations, and their interactions through three attention mechanisms; and (c) a classifier that predicts machining operations.
• Figure 3: Samples of parts geometries used for synthesized data generation: (1) simple geometry (bottom row) and (2) complex geometry (top row).
• Figure 4: Machining operation prediction via MP-GFormer. The case study utilizes STL geometry data and the BRep model as inputs, which are converted into graph representations. These graph-based inputs are then processed by MP-GFormer to predict sequential machining operations.
• Figure 5: Loss curves.