Accelerated and inexpensive Machine Learning for manufacturing processes with incomplete mechanistic knowledge

Abstract

Listen

Translate article

Machine Learning (ML) enables deployable modeling of parametric effects in manufacturing processes. But this paradigm is largely limited to established processes, since the state-of-the-art ignores the significant cost of creating qualitatively accurate physics-based models for new processes. We propose a transfer learning based method that addresses this issue by pushing the boundaries of the qualitative accuracy demanded of the physics-based model. Our approach is evaluated for modeling the printed line width in Fused Filament Fabrication and shows reduction in the model development cost by multiple human-years, experimental cost by 56–76%, computational cost by orders of magnitude, and error by 16–24%.