Scientists from the Los Alamos National Laboratory have devised a way of using machine learning in order to discover new materials with specific properties using an informatics-based adaptive strategy in combination with experiments. The new approach will help scientists find new materials in a manner that is more cost-effective and less time consuming than current procedures.
"What we've done is show that, starting with a relatively small data set of well-controlled experiments, it is possible to iteratively guide subsequent experiments toward finding the material with the desired target," said Turab Lookman, a physicist and materials scientist at Los Alamos National Laboratory and senior author of the study.
"Finding new materials has traditionally been guided by intuition and trial and error," he continued. "But with increasing chemical complexity, the combination possibilities become too large for trial-and-error approaches to be practical."
Lookman and his team used machine learning to speed up this tedious process using a framework that guides experiments using uncertainty in order to hone in on a shape-memory alloy with low dissipation, a property essential for improving fatigue life in engineering applications.
"The goal is to cut in half the time and cost of bringing materials to market," Lookman said. "What we have demonstrated is a data-driven framework built on the foundations of machine learning and design that can lead to discovering new materials with targeted properties much faster than before."
The team's machine-learning algorithm is powered by Los Alamos National Laboratory's high-performance supercomputers, which allowed them to digitize the trial-and-error process.
Using the interplay of structural, chemical and microstructural degrees of freedom allowed the team to introduce a great degree of flexibility in comparison to standard procedures that create and screen databases using thousands of quantum mechanical calculations and do not incorporate uncertainties.
Although Lookman and his team focused on nickel titanium-based shape-memory alloys, their new approach can be used for any kind of material, from polymers and ceramics to nanomaterials, as well as target properties such as dielectric responses, piezoelectric coefficients and band gaps.
Machine learning is a great benefit to the process of material selection, especially when experiments or calculations reach a level of complexity that becomes very costly and time consuming.
The findings were published in the April 15 issue of Nature Communications.
