Voxel-Based Generative Modeling for Dynamic Process Planning in Subtractive Manufacturing

Abstract Dynamic process planning in machining is essential for optimizing production efficiency, particularly in high-precision manufacturing. While traditional process planning methods rely on predefined machining sequences, they often lack the flexibility to adapt to geometric transformations and evolving process parameters. This paper presents a generative modeling framework that enables real-time, adaptive process planning at the micro-level. Our approach systematically captures the evolution of part geometry and surface quality (e.g., surface roughness) in response to machining parameters such as feed rate, depth of cut, and cutting speed. By discretizing geometric transformations into high-resolution voxel units, we create a multi-modal temporal dataset that maps the interactions between part geometry, surface quality, material properties, and machining parameters. To enhance machining adaptability, we integrate surface roughness predictions derived from deviations between actual and theoretically fitted smooth surfaces. These roughness estimations, alongside voxel-based spatial feature extraction, serve as the foundation for training deep generative models—such as Auto-Encoders—which dynamically predict successive machining steps and optimize toolpath adjustments in real-time.