877-412-3651 
MIT researchers have created a machine-learning method for modeling chemically disordered metal alloys, with the goal of improving predictions of how materials behave before they are made and tested.
MIT said the method could help companies working in aerospace, energy, and computing, where new materials are often needed but testing them can add cost and time. Current simulation methods can struggle with the complex chemical arrangements found in many solid materials, especially metal alloys.
“The real challenge in our field is modelling these chemically disordered phases,” Freitas said. “Chemical disorder means there’s a huge variety of local chemical environments, which is hard for the machine-learning model to learn.”
Material properties depend heavily on how chemical elements are arranged inside a material. Two materials can contain the same elements but behave differently if their internal chemical arrangements differ.
The MIT team focused on improving the training datasets used by machine-learning models that simulate materials atom by atom. Existing methods often use brute-force computation to create training data, which MIT said can require more than 100,000 hours of computation for a single material.
Building on previous work from Freitas’ group, the researchers used information theory to generate datasets that capture a wider range of local chemical environments. The method swaps atoms in samples to reduce repeated examples and expose the model to arrangements it might otherwise miss.
“We kept optimizing the training set so it captured as many different local environments as possible,” Freitas said.
The researchers tested the method across chemically diverse metal alloys. Models trained on the new datasets predicted material properties more accurately than models trained through random sampling or another common sampling method. MIT said the team’s models were also more accurate than much larger models created by companies including Google and Microsoft.
The team tested predictions against experimental data for atomic ordering in alloys. Xiao led simulations showing that the models could predict phase diagrams that closely matched experimental data. Phase diagrams show which phases are stable across different temperatures and chemical compositions.
The researchers are now using the method to study how alloy composition affects mechanical properties and radiation tolerance. The research was supported by the U.S. Air Force Office of Scientific Research.
Article & Image Source: MIT
