Imaginary Vessels

Clay is one of the foundational materials in art and architecture, traced in the development of mud walls and the adobe module, and showcased in utilitarian and ornamental pottery. Wheel throwing is the process of shaping clay mainly into symmetrical objects, a complex craft in which the master potter has the knowledge and skill to manipulate the clay into the final design of various physical objects. This project explores how Machine Learning can be used to translate the richness and complexity of wheel throwing for digital fabrication. In this paper we present a surrogate digital dataset for robotic fabrication and geometric prediction used to train neural networks and provide a bridge between digital fabrication and handcraft. We report on the parametric model which abstracts wheel throwing as the interaction between a rotating mass and a given set of forces, as well as on data wrangling methods, dataset composition considerations, and training methodology. We present two models, one in which geometry is predicted based on a given set of forces, and a second in which forces are predicted based on a given geometry. Lastly, we give a critical assessment of the predictions of both networks and discuss future steps.