Using machine learning to perfect nanoscale 3D printing

“We had demonstrated a significant enhancement in speed with the 3D printing process called projection multi-photon lithography, which prints 2D slices of the object at once and stacks them continuously as they are printed to form a 3D object. This lends itself perfectly to analyzing the shapes produced in those 2D slices using machine learning.” said Xianfan Xu, James J. and Carol L. Shuttleworth Professor of Mechanical Engineering. “In macroscale manufacturing methods, there are well-established processes for characterization and quality control. Some have started to use machine learning to speed up the optimization process. Conventional machine learning process requires a large number of data points, which will be very time-consuming to produce and measure for micro- and nano- scale printing.”
In a new paper published in Light Science & Application, Xu's team describes an active machine learning-based framework to determine the optimal parameters for the rapid, projection micro- and nano- scale printing process. They developed an active learning framework using Bayesian optimization to inform optimal experimentation in order to adaptively collect the most informative data for effective training of a Gaussian-process-regression-based machine learning model. This model then serves as a surrogate for the manufacturing process: predicting optimal process parameters for achieving a target geometry.
“The active learning framework serves as a digital twin for the manufacturing process to predict optimal process parameters for achieving a target geometry, improve the geometric accuracy, with drastic reductions of the errors to within the measurement accuracy," said Guang Lin, Professor of Mathematics & Mechanical Engineering, and Associate Dean for Research and Innovation in the College of Science. “Bayesian optimization acts like a guide for our experiments. It helps us figure out where to collect more training data, avoiding unnecessary work and speeding up the process. In our case, it will help us print more accurate 3D parts, but it’s also designed to be broadly applicable. We hope the success of our framework in this work will encourage others to adopt this machine learning approach,” said Jason Johnson, the lead author of the study.
Based on a number of case studies of printing various geometries of different sizes, it is shown that this active learning framework leads to more accurate prints more quickly, using much less experimental data than typically used in a machine learning process. This active learning framework can potentially be applied broadly to other additive manufacturing processes to increase accuracy, with significantly reduced experimental data collection requirements for optimization.
This project aligns with Purdue Computes, a University-wide initiative to implement machine learning and artificial intelligence in the physical world.

Source: Xianfan Xu, xxu@purdue.edu
Writer: Jared Pike, jaredpike@purdue.edu, 765-496-0374
Bayesian optimization with Gaussian-process-based active machine learning for improvement of geometric accuracy in projection multi-photon 3D printing
Jason E. Johnson, Ishat Raihan Jamil, Liang Pan, Guang Lin & Xianfan Xu
https://doi.org/10.1038/s41377-024-01707-8
Multi-photon polymerization is a well-established, yet actively developing, additive manufacturing technique for 3D printing on the micro/nanoscale. Like all additive manufacturing techniques, determining the process parameters necessary to achieve dimensional accuracy for a structure 3D printed using this method is not always straightforward and can require time-consuming experimentation. In this work, an active machine learning based framework is presented for determining optimal process parameters for the recently developed, high-speed, layer-by-layer continuous projection 3D printing process. The proposed active learning framework uses Bayesian optimization to inform optimal experimentation in order to adaptively collect the most informative data for effective training of a Gaussian-process-regression-based machine learning model. This model then serves as a surrogate for the manufacturing process: predicting optimal process parameters for achieving a target geometry, e.g., the 2D geometry of each printed layer. Three representative 2D shapes at three different scales are used as test cases. In each case, the active learning framework improves the geometric accuracy, with drastic reductions of the errors to within the measurement accuracy in just four iterations of the Bayesian optimization using only a few hundred of total training data. The case studies indicate that the active learning framework developed in this work can be broadly applied to other additive manufacturing processes to increase accuracy with significantly reduced experimental data collection effort for optimization.