Self-optimizing AI technology increases the efficiency of 3D printing

A Washington State University study published in the journal Advanced Materials Technologies could enable more seamless use of 3D printing for complex designs in everything from artificial organs to flexible electronics and wearable biosensors.

In the study, researchers learned the algorithm to identify and print the best versions of kidney and prostate models. In the process, 60 increasingly improved versions of these organ models were created.

“You can optimize the results, saving time, cost and labor,” said Kaiyan Qiu, co-corresponding author on the paper and Berry Assistant Professor in the WSU School of Mechanical and Materials Engineering.

3D printing has gained momentum in recent years, enabling engineers to quickly transform customized designs into a variety of products, including wearable devices, batteries and aerospace components. However, they still face the challenge of finding the right settings for their printing projects. Factors such as material selection, printer configuration and the pressure of the nozzle have a significant impact on the final product.

“The sheer number of potential combinations is overwhelming, and each trial costs time and money,” said Jana Doppa, co-corresponding author and Huie-Rogers Endowed Chair Associate Professor of Computer Science at WSU.

Qiu has already spent several years researching the development of complex, lifelike 3D-printed models of human organs that can be used to train surgeons or evaluate implants, for example. The models must accurately replicate the mechanical and physical properties of the real organ, including veins, arteries and other fine structures.

To optimize their 3D printing process, Qiu, Doppa and their team used an AI technique called Bayesian optimization. This enabled them to optimize three different goals for their organ models: the geometric precision of the model, its weight or porosity and the printing time. Porosity is particularly important for surgical exercises, as it influences the mechanical properties of the model.

“It’s hard to balance all the objectives, but we were able to strike a favorable balance and achieve the best possible printing of a quality object, regardless of the printing type or material shape,” said co-first author Eric Chen, a WSU visiting student working in Qiu’s group in the School of Mechanical and Materials Engineering.

The algorithm was initially trained to print a model of a prostate for surgical specimens. Thanks to the general applicability of the algorithm, it could also be used to produce a kidney model with minor adjustments.

The interdisciplinary project, which brought together researchers from different fields, underlines the importance of AI in modern manufacturing and shows how machine learning can drive the development and production of biomedical devices.