Researchers have successfully applied machine learning to create nano-architected materials that possess the toughness of carbon steel while being as light as Styrofoam. This team has outlined the development of nanomaterials that boast a remarkable balance of high strength, lightweight properties, and the ability to be customized. These advancements could have significant implications for a variety of industries, including automotive and aerospace.
A team from the University of Toronto’s Faculty of Applied Science & Engineering has harnessed machine learning to craft nano-architected materials, achieving the durability of carbon steel with the weight of Styrofoam.
In a new study published in Advanced Materials, Professor Tobin Filleter leads a group that explains how they constructed nanomaterials that uniquely blend outstanding strength and lightness while allowing for customization. Their innovative methods could be advantageous across different sectors, from automotive to aerospace.
“Nano-architected materials utilize high-performance shapes—similar to using triangles to construct a bridge—at nanoscale sizes, capitalizing on the principle that ‘smaller is stronger.’ This leads to some of the best strength-to-weight and stiffness-to-weight ratios among materials,” says Peter Serles, the paper’s primary author.
“However, conventional lattice shapes and structures often feature sharp edges and corners, which can create points of high stress, resulting in premature local failures and breakage, thus limiting their overall effectiveness.
“Recognizing this issue made me see that machine learning could be the ideal solution.”
Nano-architected materials consist of tiny building blocks or repetitive units that are only a few hundred nanometers in size—over 100 of these lined up would equal the thickness of a human hair. In this case, the building blocks, made of carbon, are configured into intricate 3D structures known as nanolattices.
To enhance their materials, Serles and Filleter collaborated with Professor Seunghwa Ryu and PhD student Jinwook Yeo at the Korea Advanced Institute of Science & Technology (KAIST) in Daejeon, South Korea. This collaboration was established through the University of Toronto’s International Doctoral Clusters program, which promotes doctoral education through international research partnerships.
The KAIST team applied a multi-objective Bayesian optimization machine learning algorithm, which learned from simulated geometries to anticipate optimal designs that could better distribute stress and improve the strength-to-weight ratio of nano-architected structures.
Serles then utilized a two-photon polymerization 3D printer at the Centre for Research and Application in Fluidic Technologies (CRAFT) to produce prototypes for testing. This advanced additive manufacturing technique allows for 3D printing at both micro and nano scales to create enhanced carbon nanolattices.
The resulting optimized nanolattices demonstrated more than double the strength of previous designs, enduring a stress level of 2.03 megapascals for each cubic meter per kilogram of density—roughly five times stronger than titanium.
“This is the first instance of employing machine learning for optimizing nano-architected materials, and the enhancements we observed were astonishing,” states Serles. “The algorithm didn’t merely repeat successful designs from the training data; it learned from the modifications—what was effective and what wasn’t—allowing it to forecast entirely new lattice designs.
“Machine learning typically requires considerable data, but gathering high-quality data through finite element analysis is challenging. The multi-objective Bayesian optimization method only needed 400 data points, while other techniques could require 20,000 or more. This allowed us to work with a much smaller yet extremely high-quality dataset.”
“We anticipate that these innovative material designs could lead to ultra-lightweight components for aerospace applications, including airplanes, helicopters, and spacecraft, which could decrease fuel consumption during flight without compromising safety or performance,” notes Filleter. “This could ultimately help lower the environmental impact of aviation.”
“For instance, replacing titanium parts in an aircraft with this material could yield savings of 80 liters of fuel annually for every kilogram of material swapped,” Serles adds.
Other team members involved in the project include professors Yu Zou, Chandra Veer Singh, Jane Howe, and Charles Jia from the University of Toronto, along with international partners from the Karlsruhe Institute of Technology (KIT) in Germany, the Massachusetts Institute of Technology (MIT), and Rice University in the United States.
“This project encompassed various facets of material science, machine learning, chemistry, and mechanics, enhancing our comprehension of how to advance and deploy this technology,” says Serles, who now serves as a Schmidt Science Fellow at Caltech.
“Our future efforts will concentrate on scaling these material designs for cost-effective, large-scale components,” Filleter adds. “We will also investigate new designs aiming for even lower density materials that still exhibit substantial strength and stiffness.”
