Using artificial intelligence to accelerate materials development
- PDF / 2,604,471 Bytes
- 10 Pages / 585 x 783 pts Page_size
- 10 Downloads / 233 Views
Using artificial intelligence to accelerate materials development By Philip Ball
S
cience is typically slow work. It can take years or even decades for exploratory work on, say, a new concept in materials to become a product ready for the market place. But advances in artificial intelligence (AI) have the potential to greatly accelerate that tortuous process. Computer algorithms are increasingly helping with the exploring and understanding, and the direction of experimentation, modeling, and simulation. They are working in parallel with human creativity and ingenuity to find and refine new materials for tomorrow’s technologies. Launching one effort to harness computational and datadriven resources, the Materials Genome Initiative, in 2011, US President Barack Obama laid out the objective. “The invention of silicon circuits and lithium-ion batteries made computers and iPods and iPads possible,” he said, “but it took years to get those technologies from the drawing board to the market place. We can do it faster.” Yet sometimes the choices available to materials scientists are enough to make you despair. Take so-called “high-entropy alloys,”1 which have high strength and are mixtures of five or more metallic elements; some may contain up to 20 elements. How can one ever hope to probe all the possible permutations and phases? Or take the exotic quantum mechanical properties discovered during the past decade or so in complex materials with compositions such as Ca10Cr7O28 and YbMgGaO4.2,3 How can we find out in any comprehensive, systematic way what other new and potentially useful behaviors might exist in combinations of elements that no one has thought to look at before? It is unfeasible to trawl blindly through all the options experimentally. Previously,
the options were narrowed down largely through intuition. But human intuition becomes severely tested by both the range and complexity of the possible choices. Yet computer algorithms can now develop a kind of intuition too through the same process that we tend to use: looking for patterns and regularities in what we already know. This is machine learning (ML), an aspect of AI that aims to digest and generalize existing knowledge to find new solutions to problems. It is used today in all manner of applications in which a large amount of data exceeds human capability to assimilate it all—from genomics and drug design to analysis of financial markets and the development of gameplaying algorithms. It seems increasingly likely that some of the outstanding challenges in materials design will be solved this way too. The potential effect of AI in materials science, however, extends well beyond the discovery of new substances and compositions. Far from being merely a tool for automated materials exploration, said Benji Maruyama of the Air Force Research Laboratory in Dayton, Ohio, AI might supply nothing less than a new way of doing science, helping to improve, streamline, and guide the process of acquiring new knowledge about the materials universe (see the Materials Rese
Data Loading...
