Mobile Navigation

IIOT Oil & Gas

View Comments

Materials properties discovery aided by machine learning

| By Chemical Engineering

A new materials-discovery platform that relies on machine-learning enables scientists and engineers to conduct large-scale searches and predict material properties from atomic structure data. Known as Xaedra, the artificial intelligence platform allows users to define desired properties and quickly identify materials that are likely to exhibit those properties.

“Xaedra allows us to cast a broader ‘net’ over the materials universe at the beginning,” says Charlie Baker, business development director at Lumiant Corp. (Kelowna, B.C.;, which recently launched Xaedra. It can help solve difficult materials challenges faster and cheaper than using density functional theory (DFT) modeling calculations, which attempt to solve quantum mechanical wavefunctions for many-atom systems.

In a key advance, Lumiant engineers have developed a way to present atomic crystal structure information about materials in a way that allows known, open-source neural-network algorithms to analyze them. In this way, each crystal structure is given an atomistic “fingerprint” that enables the machine-learning algorithm to be applied, explains Pawel Pisarski, lead developer of Xaedra.

Lumiant has loaded known material properties into its proprietary database, and these properties are used to train the machine-learning system to predict the properties of previously uncharacterized materials. It is done without DFT modeling, which can be time-consuming and expensive, the company says, although future enhancements of Xaedra may apply DFT calculations for further accuracy for some types of predictions.

One initial success of the Xaedra platform was its prediction that the composite material titanium silicide/titanium carbide could be synthesized by self-propagating high-temperature synthesis, a scenario that was not intuitively obvious to researchers. The novel ceramic material is now patented by Lumiant. Xaedra is being used on a host of other projects, including improving metal alloys by sorting and predicting thousands of alloy combinations that would be impractical to test in the laboratory, and in the field of spintronics for computer processors.