Self-Supervised Learning of Temporally Varying Process Parameter Models for Direct Ink Writing

Abstract Direct ink writing (DIW) is an additive manufacturing (AM) process known for its flexibility in printing a variety of materials. However, ink drying is one of the common issues in DIW process. Time-dependent changes in ink properties require us to use a temporally varying process parameter models to efficiently utilize ink. To address this problem, we propose a learning approach for constructing reliable process parameter models that compensate for the temporal changes in ink properties. In the study, we use a mixture of silicones as the ink and a robotic arm with a fluid dispenser system as the experimental setup. We begin by printing test artifacts and collecting initial data with various combinations of process parameters. This data is then used to construct surrogate models for the DIW process and estimate the proper ranges of process parameters to use over time. With self-supervised learning approach and the process parameter adjustment, our method maximizes ink utilization. Additionally, we apply image processing techniques to analyze the printed artifacts. Our results demonstrate the accuracy of the models, compare the DIW process outcomes for actual artifacts with and without temporal adjustment, and estimate the feasible duration for successful printing.