Using Reinforcement Learning for Direct Ink Writing Control


Closed-loop printing enhanced by machine learning. © Michal Piovarci/ISTA

Using fluids for 3D printing may seem paradoxical at first glance, but not all fluids are aqueous. Many useful materials are more viscous, from inks to hydrogels, and therefore qualify for printing. Yet their potential has been relatively unexplored due to the limited control of their behavior. Today, researchers from the Bickel group at the Austrian Institute of Science and Technology (ISTA) are using machine learning in virtual environments to achieve better results in real experiments.

3D printing is on the rise. Many people are familiar with characteristic plastic structures. However, attention has also turned to different printing materials, such as inks, viscous pastes and hydrogels, which could potentially be used to 3D print biomaterials and even food. But printing such fluids is a challenge. Exact control over them requires careful trial and error experiments, as they generally tend to warp and spread after application.

A team of researchers, including Michal Piovarči and Bernd Bickel, are tackling these challenges. In their labs at the Austrian Institute of Science and Technology (ISTA), they use reinforcement learning – a type of machine learning – to improve the technique of printing viscous materials. The results were presented to the SIGGRAPH Conferencethe annual meeting of researchers in simulation and visual computing.

A critical part of manufacturing is identifying the parameters that consistently produce high quality structures. Admittedly, an assumption is implicit here: the relationship between the parameters and the result is predictable. However, actual processes always exhibit some variability due to the nature of the materials used. In printing with viscous materials, this notion is more prevalent, as they take a significant amount of time to settle after deposition. The question is: how can we understand and manage complex dynamics?

“Instead of printing thousands of samples, which is not only expensive, but rather time-consuming, we put our expertise in computer simulations into action,” replies Piovarči, lead author of the study. While computer graphics often trade physical precision for faster simulation, here the team offered a simulated environment that accurately reflects physical processes. “We modeled the current and future short-horizon states of the ink based on fluid physics. The efficiency of our model allowed us to simulate hundreds of impressions simultaneously, more often than we ever could in the experiment. We used the dataset for reinforcement learning and gained knowledge on how to control ink and other materials. »

Learn in virtual environments how to control ink. © Michal Piovarci/ISTA

The machine learning algorithm established various policies, including one to control the movement of the ink dispensing nozzle around a corner so that no unwanted drips occur. The printing device would no longer follow the baseline of the desired shape, but instead would follow a slightly modified path that would ultimately yield better results. To verify that these rulers can handle various materials, they trained three models using liquids of different viscosity. They tested their method with experiments using inks of different thicknesses.

The team opted for closed-loop shapes instead of simple lines or writing, because “closed loops are the standard case of 3D printing and that’s our target application,” says Piovarči. While single-layer printing in this project is sufficient for printed electronics use cases, he wants to add another dimension. “Naturally, three-dimensional objects are our goal, so that one day we will be able to print optical designs, foods or functional mechanisms. I find it fascinating that we, as a community of computer graphics, can be the primary driver of machine learning for 3D printing.

Read the full research

Closed-loop control of direct ink writing via reinforcement learning
Michal Piovarči, Michael Foshey, Jie Xu, Timothy Erps, Vahid Babaei, Piotr Didyk, Szymon Rusinkiewicz, Wojciech Matusik, Bernd Bickel

Austrian Institute of Science and Technology

AIhub is a non-profit organization dedicated to connecting the AI ​​community to the public by providing free, high-quality AI information.

AIhub is a non-profit organization dedicated to connecting the AI ​​community to the public by providing free, high-quality AI information.


Comments are closed.