Jason E. Johnson, a Ph.D. candidate in mechanical engineering at Purdue University, will give a seminar titled "Active Machine Learning for Optimization of 3D Printing at Nanoscale."

Abstract: Additive manufacturing, or 3D printing, has the potential to reshape the manufacturing industry by offering greater design flexibility, faster prototyping, reduced waste, and numerous other advantages over traditional methods. Multi-photon polymerization is a type of additive manufacturing for 3D printing at the micro/nanoscale, where a high peak intensity femtosecond laser is focused into a photosensitive resin to induce multi-photon absorption and initiate polymerization. The nonlinearity of multi-photon absorption allows for confinement of the reaction to sub-100 nm scales, well below the diffraction limit. This method is widely used in fields such as nanophotonics, metamaterials, and bioengineering. Yet, as with most additive manufacturing methods, determining the optimal process parameters for a 3D printed structure can be complex and often requires extensive experimentation. Moreover, as the field of nanoscale 3D printing continues to advance, parameter spaces are constantly evolving. Active machine learning for optimal experimental design can help guide exploration of these frontiers. This presentation will discuss the use of active machine learning in nanoscale 3D printing and present work demonstrating its effectiveness in optimizing the recently developed projection multi-photon 3D printing method. In this work, an active machine learning framework is introduced that leverages Bayesian optimization to guide optimal experimentation, enabling the adaptive collection of the most informative data for efficiently training a Gaussian process regression model. This model acts as a surrogate for the manufacturing process, predicting the ideal process parameters to achieve a target geometry, i.e. the 2D shape of each printed layer. With as few as 5 experiments, or 339 training data points, the framework can reduce geometric errors in projection multi-photon printing to sub-100 nm levels. The success of these case studies, combined with the versatility of the Gaussian-process-regression-based surrogate model, indicates that the framework could be broadly applied to other additive manufacturing processes, improving accuracy with reduced experimental effort.

Biography: Jason E. Johnson is a PhD candidate in the School of Mechanical Engineering at Purdue University, where he works under the advisement of Prof. Xianfan Xu. His research focuses on improving light-based advanced manufacturing techniques, with a particular emphasis on using active machine learning to advance micro/nanoscale 3D printing. Specifically, Jason’s work explores strategies to optimize direct-laser-writing-based and projection-based multi-photon polymerization. He earned his Bachelor of Science in Mechanical Engineering from Missouri S&T in 2020. Jason is a recipient of the NSF Graduate Research Fellowship and Purdue University’s Ingersoll-Rand Fellowship.