Neural nets could speed up development of new materials.

What’s new: A deep learning system from Sandia National Laboratories dramatically accelerated simulations that help scientists understand how changes to the design or fabrication of a material — say, the balance of metals in an alloy — change its properties.

How it works: The researchers trained an LSTM to predict how the properties of a material evolve during the process known as spinodal decomposition, in which a material separates into its constituents in the presence or absence of heat.

  • The authors trained their model using 5,000 simulations, each comprising 60 observations over time, of the microscopic structure of an alloy undergoing spinodal decomposition.
  • They simplified these observations from 262,144 to the 10 most important using principal component analysis.
  • Fed this simplified representation, the LSTM learned to predict how the material would change in subsequent time steps.

Results: In tests, the model simulated thermodynamic processes, such as the way a molten alloy congeals as it cools, more than 42,000 times faster than traditional simulations: 60 milliseconds versus 12 minutes. However, the increased speed came at a cost of slightly reduced accuracy, which fell by 5 percent compared to the traditional approach.

Behind the news: Machine learning has shown promise as a shortcut to a variety of scientific simulations.

  • Alphafold figures out 3D protein structures, a capability that could accelerate drug development.
  • DENSE has sped up physical simulations in fields including astronomy, climate science, and physics.

Why it matters: Faster simulations of materials can quicken the pace of discovery in areas as diverse as optics, aerospace, energy storage, and medicine. The Sandia team plans to use its model to explore ultrathin optical technologies for next-generation video monitors.

We’re thinking: From Gorilla Glass to graphene, advanced materials are transforming the world. Machine learning is poised to help such innovations reach the market faster than ever.


Subscribe to The Batch

Stay updated with weekly AI News and Insights delivered to your inbox