In sectors such as aerospace, energy, and computing, companies are constantly seeking new materials to enhance performance. However, understanding how these materials behave in rockets or computer chips requires first manufacturing and testing them. This is because even the most powerful simulation techniques struggle to model the complex chemical arrangements in most solid materials, adding costs and time to material innovation.
A team of MIT researchers has now created a method to accurately model metal behavior, regardless of the complexity of their chemical arrangements. Central to this approach are machine-learning models that make material simulations faster and more accurate. The researchers improved these models by constructing training datasets that capture the diversity of atomic environments in chemically disordered materials.
Their findings, published in Science Advances, demonstrate that this approach can predict material properties for a diverse range of metal alloys under various conditions, and it can also be utilized to develop new materials, particularly in scenarios where experimentation is costly.
Modeling Metals
Material properties are largely determined by the internal arrangement of their chemical elements. Even if two materials have the same mix of elements, different arrangements can lead to significant differences in brittleness or deformability. Capturing this distinction requires simulating materials atom by atom, relying on models that describe atomic interactions.
Over the last two decades, machine learning has emerged as the most accurate method for constructing these models, but it struggles when chemical arrangements are disordered.
"The real challenge in our field is modeling these chemically disordered phases," Freitas states. The existing leading approach for generating training data often demands over 100,000 hours of computation, and does not transfer well when material composition changes.
Previously, Freitas' group developed a method to measure the chemical complexity of solid materials by analyzing the frequency and spacing of small atom groups. In this study, they leveraged that capability to build better training datasets using a mathematical approach known as information theory.
"We kept optimizing the training set to capture as many different local environments as possible," Freitas explains. By training on their datasets, the models predicted material properties more accurately than those trained using random sampling or other popular methods.
From Lab to Industry
The method works, in part, by identifying hidden patterns in the sample data, described in the paper as "subtle energetic biases toward certain local chemical configurations." These small energetic differences are crucial as they influence which phases form in an alloy, how those phases evolve with temperature and composition, and ultimately, the material's properties.
The researchers are now applying this technique to study how changes in an alloy's composition affect mechanical properties and radiation tolerance, aiming to design materials that maintain strength and damage tolerance in harsh environments.
The goal is to ensure these predictions are useful in the contexts where material decisions are made, as Freitas emphasizes. This research was supported by the U.S. Air Force Office of Scientific Research.
Blogger's Review: This research significantly enhances the accuracy of modeling metal alloy behavior through innovative machine learning techniques, breaking the limitations of traditional simulations. The future of material design will be more efficient and intelligent, propelling industry advancement.