Generative Modeling of the Shape Transformation Capability of Machining Processes

The shape, material property, and part quality transformation capabilities of a manufacturing process are essential process capability knowledge that are traditionally acquired by process planners through experience. While efforts have been made over the years to develop automated systems that utilize known process capabilities for process selection and manufacturability assessment of part designs, such systems are hampered by the lack of a systematic approach to capture and model the shape, material property, and part quality transformation capabilities from design and manufacturing data. In this paper, the shape transformation capabilities of representative machining operations are modeled using 3D Variational Autoencoders and Generative Adversarial Networks (3D-VAE-GANs.) The proposed approach models the shape transformation capability as a latent probability distribution from which visualizations of realistic machinable features can be sampled for shape decomposition and reconstruction, thereby assisting machining process selection by a process planner and manufacturability assessment of part shapes generated by a designer.